Xtreme whatever the heck you want to call this

[alpha754293]

There really isn't just a section of the forum for generally xtreme stuff, so I ended up dumping it here.

System specs:
Intel Q9550 (2.83 GHz@3.2 GHz)
4x 2 GB of DDR2-800 RAM
Gigabyte EP45-UD3P (I think....)
WD 250 GB 7.2krpm SATA
WD 36 GB 10krpm SATA
2x Hitachi 2 GB 7.2krpm SATA
OCZ Vertex 3 240 GB SSD SATA
BFG GTX260OC (I think)
case, power supply, optical drive, blah blah blah...

Anyways, so this is just a video that I just created where I am trying to spin a car model around.

http://www.youtube.com/watch?v=rIlGZuyaPSw

It IS an actual production vehicle (or soon to be SOP'd). Please keep the car and the car company anonymous. But that is using semi-live (CGR) and partially live data (some of it will have the full geometric definition/specification, which others might still be CGRs).

The point of showing this is a) a very tiny snippet of what I do, and b) a kind of demonstration of "this is what engineers do" and perhaps, more importantly on this site, how computer hardware fairs when you literally, throw a car at it.

And I didn't even have the full vehicle completely loaded in yet. This is just MOST of it. (Even if I were to load more of it in, it would be really hard to observe the slowdowns because at that point, the CPU is desperately struggling to keep up.)

I couldn't get any sort of accurate part count because of the mix of live and semi-live models, but there's probably well over 10,000 as shown.

car_frt_left_3qtr2.jpg

[Panther_Seraphi]

Use MSI afterburners capture feature seems to have less load than fraps

[Kayin]

Did you download that car? (cue eye rolls.)

[alpha754293]

Use MSI afterburners capture feature seems to have less load than fraps

I don't know if it would matter that much...

The session crashed shortly after when I tried to load in more data (of the car).

[alpha754293]

Did you download that car? (cue eye rolls.)

In a manner of speaking...yes.

For this purpose/application, I can probably really use an Ivy Bridge OC'd to 5 GHz at least. (Probably 7 GHz would be preferable since it's only a single process application anyways.)

[Liam_G]

For this purpose/application, I can probably really use an Ivy Bridge OC'd to 5 GHz at least. (Probably 7 GHz would be preferable since it's only a single process application anyways.)

I have been designing a computer case for a while now in Solidworks and with some models I see similar and sometimes a lot worse slowdown when just panning and rotating around.

My System specs:

CPU: Q9550 @ 4.0ghz w/ 1.368v (471*8.5)
RAM: 4x2 Gigs Corsair 1066 @ 942mhz (471*2)
MOB: Asus P5Q Deluxe
HDD: 2 x Seagate 250gb RAID 0
GPU: 2 x Asus hd6850 Crossfire

My real issues are when i am trying to rebuild complex parts, one example is a hexagonal mesh for the case side panel vent, it can bring my computer to a halt for hours. In the end i either have to make do with textures or just suppress the detail after i have modeled it or convert to a parasolid. I find i have to compromise my model detail, some would call it "smarter" modelling but i feel with the power our CPU's have these days we shouldn't have to compromise any level of detail.

Which makes me wonder, just how much difference will, for instance an i7 3960x @ 5ghz make, considering the main problem comes from the fact that the programmers haven't figured out how to make the program fully multi threaded yet... Will Intel or AMD be able to produce a highly overclockable and highly threaded CPU any time soon, maybe a quad core Piledriver might be highly overclockable for single threaded apps but the IPC prob will still have it lower than i7 3770k in overall speed. I would kill for a CPU that could overclock a single core/2 threads to 7ghz under water cooling for single threaded programs while still having 8 or more threads for the stuff that does take advantage of it, wishful thinking for now though. In my dreams i think of having a chip that will do 10ghz single core 5ghz all cores... we will probably have working quantum computers before that happens though.

[alpha754293]

I have been designing a computer case for a while now in Solidworks and with some models I see similar and sometimes a lot worse slowdown when just panning and rotating around.

My System specs:

CPU: Q9550 @ 4.0ghz w/ 1.368v (471*8.5)
RAM: 4x2 Gigs Corsair 1066 @ 942mhz (471*2)
MOB: Asus P5Q Deluxe
HDD: 2 x Seagate 250gb RAID 0
GPU: 2 x Asus hd6850 Crossfire

My real issues are when i am trying to rebuild complex parts, one example is a hexagonal mesh for the case side panel vent, it can bring my computer to a halt for hours. In the end i either have to make do with textures or just suppress the detail after i have modeled it or convert to a parasolid. I find i have to compromise my model detail, some would call it "smarter" modelling but i feel with the power our CPU's have these days we shouldn't have to compromise any level of detail.

Which makes me wonder, just how much difference will, for instance an i7 3960x @ 5ghz make, considering the main problem comes from the fact that the programmers haven't figured out how to make the program fully multi threaded yet... Will Intel or AMD be able to produce a highly overclockable and highly threaded CPU any time soon, maybe a quad core Piledriver might be highly overclockable for single threaded apps but the IPC prob will still have it lower than i7 3770k in overall speed. I would kill for a CPU that could overclock a single core/2 threads to 7ghz under water cooling for single threaded programs while still having 8 or more threads for the stuff that does take advantage of it, wishful thinking for now though. In my dreams i think of having a chip that will do 10ghz single core 5ghz all cores... we will probably have working quantum computers before that happens though.

Well, I remember reading the dev blog for POV-ray quite some time ago (I think when 3.07 was still in beta) and they were saying that some of the intrinsic math functions are actually REALLY hard to parallelize.

Like if you have a + b = c, how do you parallelize that. (And now suppose that a, b, and c, are vectors or matrices...the process of dividing it up for the parallelization might be more expensive than doing it all on one core). So I dunno. I'm not a programmer, so I don't really know. You would think that with the prevalence of multi-core systems that they would have been working it for quite some time by now and that there WOULD be a CAD package available now that's fully multi-processor aware, but alas, that does not seem to be the case.

(And I refrained from calling it multi-threaded becuase they ARE multi-threaded, but just not multi-processor capable (or necessarily even aware of multiple processors)).

As for the hex mesh problem - I think that it really depends on how you've modelled it.

I remember raising this issue/question to one of my friends like...back in 2004-2007 timeframe and he was like "well...you would rarely load in a full vehicle anyways, cuz that's just dumb", and to his credit, he does have a point that even as a vehicle systems engineer/vehicle systems integrator, you would hardly ever load the whole thing in all at once. And it WOULD be nice/ideal (say for vehicle engineer) that if the powertrain got all their stuff sorted out first, then body, then chassis or something like that (not necessarily in that order, but it's a A -> B -> C dealio), but the reality is that instead of an 18 month long development timeframe/schedule, it'll be like 10 years. And I still think that you'd have just as many problems now throughout the process as you would doing it the way we do it now (which is everything happening and being all developed concurrently and in parallel). But where it would be handy if we could load the full vehicle in is when the different groups are all arguing and fighting for one reason or another (space and packaging is often a main issue. The chassis guys wants it one way, but then the powertrain guy has the exhaust running right through the rear cradle and subframe AND the driveshaft - blah blah blah). So being able to move stuff around dynamically, and then being able to rerun all of the simulations, engineering, and analysis would be ideal.

If you're doing it for visualization purposes, then you can probably keep it off for most of the development time. But if it's something that depends or is integral to it (like in the model that I've got right now, it's lost a lot of the links/references to nuts and bolts, which for my purposes of just looking at the data, isn't really critical/necessary, but for someone else who might be worried about clearance or access, then yea, they would be rightfully worried about it, and would need them all to be there and in place).

It also depends on how you've instanced them.

One thing that the CAD program that I'm using allows me to do is use CGR (I think it's short for computer graphics representations or something like that). Basically, it's like a parasolid for visualization purposes only, but if I have the source/underlying CAD for it, I can double click on it, and that's when it would actually open the whole thing up and would let me edit it, and stuff, and then when I'm done with it, it will updae the CGR model, so that I don't necessarily have to have the live model loaded in for me to do finish doing what I'm doing. So it also depends a little bit on the program that you use.

So I guess that it depends a little bit. But there may be sometimes like a smart way of modelling, and also a not-so-smart ways as well. For what I've done in the past, a lot of it was driven by how parametric and how flexible the model needed to be.

[alpha754293]

And there ARE performance differences between CAD packages too.

Like for the life of me, I've NEVER been able to figure out why UG NX needs to redraw the entire model whenever you switch from say...shaded to wireframe view. And it doesn't even use the GPU!!! It uses the CPU for the redraw! Whyyyyy????? That's like the dumbest thing ever.

And in contrast, CATIA doesn't do that. Or if it does, it does it with the GPU, NOT the CPU. (Switching between view modes is a visualization thing, not a modelling thing, but NX it's like it treats it like it's a modeling thing....and I've NEVER understood why it does that.)

[Liam_G]

Yeah, i know that it is hard to fully multi-thread the app because of not being able to split up certain functions, the a+b=c thing. What i am thinking they could do is allow you to set up many functions at the same time.... for instance, in solidworks, to model the mesh i spoke of i have to do an extruded cut to a solid, and in that sketch i have about 6000 hexagons derived from a linear pattern, that can only be calculated on one core...

but say the program was designed to allow you to set up 4, 8, 16 or even 32 extruded cuts/sketches for dual socket systems at the same time, none of them derived from each others dimensions so it didn't have to wait for the first one to complete, and only when you had set up all the sketches then you could allow the program to run it, effectively engaging as many available cores/threads as you have, 1 per sketch, so instead of the CPU having to figure out how to split into multiple threads we just set it up to run that way, effectively breaking the work down into smaller parts for the CPU to process... Possible? worth thinking about. I'm a bit of a solidworks noob so there may already be something like this.

Also, breaking the extruded cut into smaller portions and then running one by one doesn't work because of what you mentioned about NX, when i do one extruded cut it first rebuilds all previous cuts before doing the last one, thats why i thought setting up all sketches and then building them all at once would be faster, you only need to build it once.

I'm not so bothered about slow performance when you have a massive assembly loaded, in that case i am happy to turn off features and hide some detail to speed it up, but when it comes to modelling individual parts that is when i am looking for fast and fully detailed/accurate models.

[alpha754293]

Yeah, i know that it is hard to fully multi-thread the app because of not being able to split up certain functions, the a+b=c thing. What i am thinking they could do is allow you to set up many functions at the same time.... for instance, in solidworks, to model the mesh i spoke of i have to do an extruded cut to a solid, and in that sketch i have about 6000 hexagons derived from a linear pattern, that can only be calculated on one core...

but say the program was designed to allow you to set up 4, 8, 16 or even 32 extruded cuts/sketches for dual socket systems at the same time, none of them derived from each others dimensions so it didn't have to wait for the first one to complete, and only when you had set up all the sketches then you could allow the program to run it, effectively engaging as many available cores/threads as you have, 1 per sketch, so instead of the CPU having to figure out how to split into multiple threads we just set it up to run that way, effectively breaking the work down into smaller parts for the CPU to process... Possible? worth thinking about. I'm a bit of a solidworks noob so there may already be something like this.

Also, breaking the extruded cut into smaller portions and then running one by one doesn't work because of what you mentioned about NX, when i do one extruded cut it first rebuilds all previous cuts before doing the last one, thats why i thought setting up all sketches and then building them all at once would be faster, you only need to build it once.

I'm not so bothered about slow performance when you have a massive assembly loaded, in that case i am happy to turn off features and hide some detail to speed it up, but when it comes to modelling individual parts that is when i am looking for fast and fully detailed/accurate models.

I forget if my email's on my profile, but you're more than welcome to send me your model if you want me to take a look at it.

I think that the other part of the problem is that typically in CAD design, the result of one object or feature depends on the result of another. So, it's kind of like a Markov chain.

I'm not really sure if there's really a good way to parallelize 6000 hexagons only because to us, it's a array that the result can be computed in parallel, but the computer might not see it that way, so it might actually be doing it one hexagon at a time.

I'll probably have to take a look at your model to see if there's a way to come up with a faster representation method, but still maintain full accuracy.

At least in your case, you have a CHANCE of speeding things up. Unfortunately, in a good portion of automotive engineering, there isn't. Or it's minimal at best. And even with loading just mostly CGR (representations, which is even a step down from parasolids), I get what you see in the video --- very low framerates.

Granted, the video card is a bit older, and it's not a professional CAD card, but if there was a way for me to capture Windows Task Manager along side, you'd be able to see that it's mostly the CPU that's doing the work at that point, and no so much the GPU. (Or maybe that the CPU is taking the GPU overflow, I dunno...)

Suffice to say that regardless, it taxes the system quite a fair bit.

And given this forum, where you have people OCing a fair bit - I, for one, would LOVE to be able to see how their OC'd systems perform when given this type of a workload.

[alpha754293]

I think that the closest thing that I've come to doing that would be doing a trim by surface. So you basically create hexagonal prisms as just shell surfaces, and then at the very end, you're going to say "cut the solid with these surface". Then, I think that it might treat all of the cuts as like a single "stamping" operation. I don't know if that would work better for you or not.

Surfaces in the CAD program that I use may or may not be related or dependent on each other. You can set up it in either way. But the program itself is still only for a single processor, so it doesn't really matter all that much.

[alpha754293]

More pictures. Once again, do not mention the name of the car or the car company if you know it.

front.jpg

It took me a little while to actually come up with all of the settings for the vehicle occupant accomodation (VOA) workbench. I WAS going to try and position the manikin manually but that would probably have been quite a tedious task, so I ended up using that instead, since it's a lot quicker and easier. But because the full vehicle model is so slow and cumbersome to manipulate, I had to fire up a second session where I just have the seats, the steering, the pedals, and also the interior trim so that I can get a rough idea as to where some of those points in space are (like the ARP, AAP, and AHP). It did a reasonable job. The thing that was a little surprising to me was that in editing the steering wheel grip diameter, the VOA model didn't update itself to accommodate for that, so there actually IS a little bit of an interference between the hands and the steering wheel, but overall, it looks about the same. And he slouches. lol...which is probably NOT good for RSI.

Took a bunch of pictures of him looking around. And if you KNOW which car it is, and you google the pictures of the actual car, it's kinda cool to see the real thing vs. CAD. (And how similiar it is.) The thing that I had to actually check a few times was the glass that's by the side mirrors and also the B-structure underneath it.

I AM missing some of the CC and IP trim, but I suppose another way of looking at it is that you get to see some of the stuff that goes on underneath "the hood" per se, without necessarily having to rip and tear up your car apart. It's a neat little car. It's PACKED with stuff, like...somewhat common, but ridiculously so.

And now, it makes me want to go out and by one since I know stuff about it.

P.S. The engine's pretty neat too, but unfortunately, I won't be able to do much with the data (since they ARE JUST visual representations) so I can't extract any useful data out of it (to feed into my engine combustion CFD simulations).

And I thought/figured that maybe SOME of you guys might be interested since I'm guessing that the vast majority of you probably have never really seen an almost-entire vehicle CAD data package in detail, let alone having a sense or a feel for how demanding it really is on computer systems. And this is JUST the final results too! Now imagine if this was actual live data, and think of all of the engineering man-hours that went into every tiny aspect of it from concept to production. To me, the most amazing and telling sign has always been the underbodies and B-structures because every hole, every nook, every crevice has or serves a purpose or a function. All the analysis, the simulations, the crash analysis, the studies, the forming studies, the packaging studies where you're trying to fit everything into one neat little package that we call "a car" -- what can I say? I love my job!

right.jpg
left.jpg
right2.jpg
001.jpg
002.jpg
003.jpg
004.jpg
005.jpg
006.jpg

I did ended up having to fire up one of my older workstations in order to be able to load the vehicle representations in because my Q9550 just wasn't able to handle/take it. (Probably not enough RAM). Ironically, my older workstation, which is a dual AMD Opteron 254 (2.8 GHz, 1-core), has 16 GB of RAM, so it was much better table to take the load than a Q9550 with 8 GB of RAM.

Topped out at 10 GB used (about 5-ish GB left) and it STILL managed to generate a 21.7 GB swap file when I woke up this morning.

Again, this is JUST the visual representations of the vehicle, and it isn't even the actual live data. I'm trying to load the actual live body-in-white data now but between the front subframe, the front suspension, and the rear subframe and suspension, it's already eating 8 GB of RAM and created an 8 GB swap file. And even then, some of the data are just dumb solids.

*edit*
I WAS going to try and "pop" the hood, but it would have taken WAYYYY too much time to try and manipulate the model so that I would be able to get the hood into the open position (and I'm not sure if they're using gas shocks or a more traditional prop rod -- I would guess that it would be the latter given that's it's like their flagship vehicle). So instead, I just hid the hood. And then also took off the engine cover as well so you guys can take a little bit of a virtual peek.

[Liam_G]

Yeah, you do have a cool job!! That looks like a pretty insane assembly to be running on your hardware, especially the older AMD Rig. Whats your virtual memory size set at on your q9550 rig? Might be faster than your AMD rig if you have it set big enough on your SSD. I can't wait to upgrade to x79 and an i7 3930k, 32 or 64gb of RAM coming up!! And fast RAM too, like 1866mhz, twice my current speed and up to 8 times the amount, droool!!!! Not to mention a decent improvement in IPC with the 3930k over the Q9550, i'm thinking about 30-40% from looking at Superpi 1M scores, maybe more if I crank the 3930k to 5ghz.

Oh and no worries about my mesh model, it's all done and dusted now. I did do some research last night though to try and find a more efficient way to do it in Solidworks and came up empty. Could possibly mirror the pattern, seems to be slightly less load than one huge pattern. One thing i noticed with the massive linear pattern, the 6000 hexagons is only half the cut, it can handle that in about 10 minutes, it's when i do a second extruded cut with same amount of hexagons that my system grinds to a halt and generally crashes before it completes the feature, mainly due to my impatience and doing other tasks on the computer instead of just clicking "Exit Sketch" and leaving it till it is done. Solidworks is not good at handling large numbers of edges, and that is over 200000 edges just in one very simple part, so it's understandable that it is slowing my system down. Funnily enough, when i turn "edges" off in the viewport and just look at shaded surfaces it functions better in general, panning etc, just not rebuilding.

[@rne]

Ever tried quadro or FireGL? I am sure you did at work

[alpha754293]

Yeah, you do have a cool job!! That looks like a pretty insane assembly to be running on your hardware, especially the older AMD Rig. Whats your virtual memory size set at on your q9550 rig? Might be faster than your AMD rig if you have it set big enough on your SSD. I can't wait to upgrade to x79 and an i7 3930k, 32 or 64gb of RAM coming up!! And fast RAM too, like 1866mhz, twice my current speed and up to 8 times the amount, droool!!!! Not to mention a decent improvement in IPC with the 3930k over the Q9550, i'm thinking about 30-40% from looking at Superpi 1M scores, maybe more if I crank the 3930k to 5ghz.

Oh and no worries about my mesh model, it's all done and dusted now. I did do some research last night though to try and find a more efficient way to do it in Solidworks and came up empty. Could possibly mirror the pattern, seems to be slightly less load than one huge pattern. One thing i noticed with the massive linear pattern, the 6000 hexagons is only half the cut, it can handle that in about 10 minutes, it's when i do a second extruded cut with same amount of hexagons that my system grinds to a halt and generally crashes before it completes the feature, mainly due to my impatience and doing other tasks on the computer instead of just clicking "Exit Sketch" and leaving it till it is done. Solidworks is not good at handling large numbers of edges, and that is over 200000 edges just in one very simple part, so it's understandable that it is slowing my system down. Funnily enough, when i turn "edges" off in the viewport and just look at shaded surfaces it functions better in general, panning etc, just not rebuilding.

The Q9550 has the VM set at 16 GB. But I think that more importantly, is that the system a) is running Windows 7 Professional x64 vs. Windows XP Professional x64 (which that alone, consumes like 2 GB of RAM), and b) that my older AMD rig just has more RAM than my Q9550. (The EP45-UD3P can't take more than 8 or 16 GB I think. And the cost of adding more RAM to it now just doesn't makes sense. That plus I don't have any other system that would be able to take the DDR2-800 unbuffered, non ECC RAM.)

I actually just added an Intel 520 180 GB SSD and also an older Intel SATA 3 Gbps RAID HBA (SRC28X) to is since the board only has SATA 1.5 Gbps on it being an older board and all. And I've put the swap file on that, and because I was playing around with just loading the live data in, and I think I might have gotten like 1/5th of the body and chassis stuff (IF that) in before I started running out of RAM and racking up a 20 GB swap file; so I killed that, put the SSD in, and also turned on CGR/cache management so that now when I load the files in, it actually generates the visual representations of the files onto the SSD (no quota set), and will only load the complete live data when I need it to. (And once I'm done with that, I can switch back to visualization mode, it'll update the CGR and sorta unload it from RAM - sorta...to speak.)

But given how demanding a full vehicle is (or even a part of a vehicle), I'm now seriously contemplating on getting 128 GB of RAM on my next system instead of 64 GB, because I have a feeling that I would be able to eat that up as well. Every delicious byte.

Honestly, that's why I was starting to think (cuz I had another sorta crazy-ish benchmarking thread just not too long ago where I took a 3930K up to 4.5 GHz), but because this is only or mostly a single processor application anyways, that I might be able to get away with using the 3770K instead. It's cheaper, lower power consumption. I don't know what the verdict is on OCing that yet because I did read about how people were noticing higher OC temps, so...I dunno. Maybe that might be easier to push to 5 GHz since it's only quad-core rather than a hex-core. *shrug* who knows.

I dunno. The model that I've got in the vehicle is DEFINITELY more than 200,000 edges. LOL. I think that when I had the live data in (1/5th of the body or less), that alone had like some 10,000 curves (that might just be like bounding edges or curves needed to create the geometries, but all of the boundaries of all of the surface segments is highly likely to be substantially more. I forget exactly. Point being that there was a LOT more "geometric entities" than 200,000 edges. I think that it might have something to do with how Solidworks does the linear pattern. (i.e. does it do it cut by cut since the result of one is dependent on the other in terms of computer/Solidworks speak). You would think that if you were to clone the hexagon in the sketch however many times you need to would be faster than doing one hexagon cut and then cloning the cut, but I dunno. I've never really attempted to commit suicide by doing anything that complicated with it. (All the complicated stuff, I leave to the big boys/Solidworks bigger, older, brother/cousin.)

[alpha754293]

Ever tried quadro or FireGL? I am sure you did at work

I'm actually using a 3D Labs Wildcat Realizm. On my Socket 940 AMD rig it's a Realizm 200 (which still kicks butt). On my Socket F AMD rig, it's a Realizm 800.

At work it's a Quadro 4000 right now. And the specs and the benchmarks say that it should be faster/better, but when I'm actually doing stuff with it, I feel that the only thing that really bothers the Realizms is when you have full scene AA turned on. Otherwise, I've done engine designs (for engine/SI/combustion CFD on a 4.6 L V8) using just the Realizm 200 and it's perfectly fine.

I've tried the very OLD old old...FireGLs back in the day and they sucked. Haven't tried any of the new ones though.

[alpha754293]

Yeah, you do have a cool job!! That looks like a pretty insane assembly to be running on your hardware, especially the older AMD Rig. Whats your virtual memory size set at on your q9550 rig? Might be faster than your AMD rig if you have it set big enough on your SSD. I can't wait to upgrade to x79 and an i7 3930k, 32 or 64gb of RAM coming up!! And fast RAM too, like 1866mhz, twice my current speed and up to 8 times the amount, droool!!!! Not to mention a decent improvement in IPC with the 3930k over the Q9550, i'm thinking about 30-40% from looking at Superpi 1M scores, maybe more if I crank the 3930k to 5ghz.

Oh and no worries about my mesh model, it's all done and dusted now. I did do some research last night though to try and find a more efficient way to do it in Solidworks and came up empty. Could possibly mirror the pattern, seems to be slightly less load than one huge pattern. One thing i noticed with the massive linear pattern, the 6000 hexagons is only half the cut, it can handle that in about 10 minutes, it's when i do a second extruded cut with same amount of hexagons that my system grinds to a halt and generally crashes before it completes the feature, mainly due to my impatience and doing other tasks on the computer instead of just clicking "Exit Sketch" and leaving it till it is done. Solidworks is not good at handling large numbers of edges, and that is over 200000 edges just in one very simple part, so it's understandable that it is slowing my system down. Funnily enough, when i turn "edges" off in the viewport and just look at shaded surfaces it functions better in general, panning etc, just not rebuilding.

I switch to wireframe mode as well, since moving around with the full car (or most of the full car) is VERY VERY slow. Even with wireframe, although the wireframe is a bit more tolerable than the shaded mode.

You might be able to try playing around with your AA settings if you've got them on. You sacrifice a bit of image quality, but maybe that might help speed things up a little bit for you.

[alpha754293]

@Liam_G
Just did a quick check. I've got just under 12,000 lines just in the rear subframe alone.

And it's got over 302,000 edges as well (an edge being anything 1D entity (curved or otherwise) such as surface segment boundaries; also for the rear subframe alone.).

[Liam_G]

Yeah, we are talking about two totally different beasts here. The complexity of that car assembly massively eclipses my PC Case assembly, although that 200,000 edges was just one part, total of whole assembly i have no idea, significantly more lines/edges than that, prob more like 2,000,000, not really relevant though, We are obviously dealing with two very different loads on the system.

I did some messing around with the full assembly loaded and really, can only get a slow down when i am trying to edit a part within the assembly, causing it to rebuild the whole assembly. Or when I have edges/shaded on, or wireframe. Wireframe in solidworks is slower than shaded mode without edges. I think I am only using about 4-6gig of ram with full assembly loaded, so compared to your 20+gig stuff i'm dealing with a light load, doesn't stop me getting annoyed with the system sometimes, honestly if i was dealing with your car assembly on my system i just wouldn't do it, can't tolerate an unresponsive system for too long, and it seems as though you have access to better equipment at work...

Where were you playing with the 3930k? Is it yours or at work? If you have access to that why are you still messing with your q9550 or the AMD system? I think with x79 or z77 they are both a little lack-lustre in OCing, 4.5ghz for 3930k or 3770k unless you have nice custom watercooling, even then 5ghz on either CPU requires too much voltage for 24/7 longterm. Mind you, I will happily push volts though a 3930k to get 5ghz, i think some ppl are too cautious, but really if i'm ocing i'm prepared to lose a chip, so may as well go for the big 5g. I would love to achieve 5.5ghz on 3930k on one core for Solidworks, that would be nice And if Solidworks made better use of Cores/threads i would splurge on a dual socket system, but so far as i know you can't overclock the latest xeons hopefully there is some leeway there like with the i7 3820...

This thread has got me dreaming about system upgrade, hard to justify when i only need the power to satisfy a hobby though, maybe i should do some contract CAD work again :P

[alpha754293]

Yeah, we are talking about two totally different beasts here. The complexity of that car assembly massively eclipses my PC Case assembly, although that 200,000 edges was just one part, total of whole assembly i have no idea, significantly more lines/edges than that, prob more like 2,000,000, not really relevant though, We are obviously dealing with two very different loads on the system.

I did some messing around with the full assembly loaded and really, can only get a slow down when i am trying to edit a part within the assembly, causing it to rebuild the whole assembly. Or when I have edges/shaded on, or wireframe. Wireframe in solidworks is slower than shaded mode without edges. I think I am only using about 4-6gig of ram with full assembly loaded, so compared to your 20+gig stuff i'm dealing with a light load, doesn't stop me getting annoyed with the system sometimes, honestly if i was dealing with your car assembly on my system i just wouldn't do it, can't tolerate an unresponsive system for too long, and it seems as though you have access to better equipment at work...

Where were you playing with the 3930k? Is it yours or at work? If you have access to that why are you still messing with your q9550 or the AMD system? I think with x79 or z77 they are both a little lack-lustre in OCing, 4.5ghz for 3930k or 3770k unless you have nice custom watercooling, even then 5ghz on either CPU requires too much voltage for 24/7 longterm. Mind you, I will happily push volts though a 3930k to get 5ghz, i think some ppl are too cautious, but really if i'm ocing i'm prepared to lose a chip, so may as well go for the big 5g. I would love to achieve 5.5ghz on 3930k on one core for Solidworks, that would be nice And if Solidworks made better use of Cores/threads i would splurge on a dual socket system, but so far as i know you can't overclock the latest xeons hopefully there is some leeway there like with the i7 3820...

This thread has got me dreaming about system upgrade, hard to justify when i only need the power to satisfy a hobby though, maybe i should do some contract CAD work again :P

Well...I don't know if the number of edges (regardless of whether it's a computer case or a car) makes that much of a difference. It might, maybe, depending on the complexity of the edge. But it's still an edge that the graphics card has to draw. And for me, my wireframe mode is slightly/marginally faster (it's sooo hard to tell when you're already working with such a large/complex model - since I probably can't really objectively tell the difference between 0.1 fps vs. 0.2 fps).

I was going to take a look at your model so that maybe we might be able to come up with a more efficient/faster method of modelling it. *shrug* it was a thought.

That 302k edges was JUST for that subassembly. I haven't actually done a total count on the whole vehicle (and not sure that I can because I don't think that I can since I don't have the full vehicle as live models).

The visual representation ONLY, so far takes about 13 GB of RAM. Although, 5 that might be because I loaded the entire rear subframe assembly in full, so it might only really be between 8-10 GB. (It's hard to tell, but it jumped from like 2.8 GB when I loaded the rear subframe initially in visualization mode only first, and then when I switched it to design mode (which loads the full model in, not just the visual representation), it jumped to like 7.8 GB.

I know that if I only load four of the assemblies in (front suspension and subframe, rear suspension and subframe, front bumper, rear bumper) as live data - I'd be already pushing like 10-12 GB of RAM used. So, they're quite "heavy" models. Which also means that I KNOW that I won't have enough RAM (I don't have any system that has enough RAM) to be able to load all of the live data that I have.

Uhh....when you're doing vehicle engineering, you learn to be patient.

Some of the simulations can take days depending on what it is; so...it is what it is. And because you're always constrained by budget and hardware and stuff - mehhh...we can ALWAYS make the problem harder.

It was a loaner system. I got the data after the fact.

I think that EVGA is coming out with the LGA2011 successor to the SR-2 (dual LGA1366). And I THINK that people were telling me that you can OC that, but don't quote me on it.

I would also think that if you were say to limit your 3930K or 3770K to just using two cores in the BIOS, that you should be able to push the chip even higher/harder. I dunno. But considering how the whole TurboBoost thing works, I would think that it should work along the same theory right? *shrug* no idea.

A lot of the stuff that I do at some point either becomes RAM limiting or CPU limiting. Rarely do I hit both at the same time, with equal "gusto".

If it makes you feel any better, I think that I'm likely going to be dropping close to $17k in hardware (probably over the course of a year) since I also do simulations as a "hobby" and in order for me to be able to do said simulations, I need to do the geometry prep myself as well (which is going to be another $2k-3k on top of that).

The combusion CFD that I was doing on just a measly 63k elements took 12 hours. The original single-cylinder model that I was working with I think had like somewhere between 400k-700k elements. So, yea, that's expected to run about a week, straight. And that's JUST the combustion aspect of it. Add the piston motion on top of that and it'll likely push a week to a week-and-a-half.

Now copy that 8 times, throw them all outta sync (i.e. in phase based the firing order), and then add the intake and exhaust manifolds on top of that...and I'm probably looking at like a 2-month straight run.

Fun stuff.

(And I've already tested that ONE 8-core CPU can cut the time down to 8 hours from the original 12 hours, so it's quite possible that TWO 8-core CPUs would be able to drop that down to between 4-6 hours.) Such an expensive hobby....LOL...

[Liam_G]

Just curious, why are you running simulations of a combustion engine for a hobby, why not build the engines? Especially when it will set you back 20k to do it.....

I only run simulations to get an idea of what will happen before i validate that testing in real life. I will eventually justify spending money for my solidworks hobby because in the end i need to upgrade my system, play games and i do build the stuff i design....

So are you building engines? :P because that would be cool, i used to draw plans for 2 stroke motorcycle engines as a kid and always wanted to build an engine from scratch, however small and simple it was, i even designed a model in Autodesk inventor when i first learnt CAD in highschool

[alpha754293]

Just curious, why are you running simulations of a combustion engine for a hobby, why not build the engines? Especially when it will set you back 20k to do it.....

I only run simulations to get an idea of what will happen before i validate that testing in real life. I will eventually justify spending money for my solidworks hobby because in the end i need to upgrade my system, play games and i do build the stuff i design....

So are you building engines? :P because that would be cool, i used to draw plans for 2 stroke motorcycle engines as a kid and always wanted to build an engine from scratch, however small and simple it was, i even designed a model in Autodesk inventor when i first learnt CAD in highschool

Suppose that it's $20k per build, if I screw up the build, then it's $20k a piece.

Whereas with the CFD simulations, I can get it like 80% of the way there, that way I might only have to build it once or twice at the most.

That and also I haven't found a place where I would be able to do the machining properly (no access to 5+ axis CNC). And I haven't found a place where I would be able to get the heads and block cast at a decent price either. (Foundry work is dirty and for a one-off, expensive).

I think that the so-called "old school" garage engine builders worked really well from carbureted probably up through port fuel injected engines.

But if you're dealing with homogenous charge gasoline compression ignition (HCCI-gasoline) or controlled auto-ignition (CAI) or high pressure piezo injectors (I think that the injection pressure is something like 2 MPa); what you're really dealing with is a bomb. A $20k bomb.

And without a LOT of VERY expensive equipment to study the fuel spray and atomization process, or being able to measure the combustion temperatures/flame front/emissions testing (and subsequently cleaning) equipment - CFD and computers and electricity cost is cheap in comparison.

That and I'm pretty sure that my neighbors wouldn't like it if I'm running the engine lab at like 2 AM whereas I can run the computer 24/7 and if it gets too loud, I can always stick it in the basement with a dehumidifier.

The CFD won't tell me EVERYTHING. But it can probably get me 80% of the way there at best.

That and also, if I want to get into engine development, without CFD nowadays, it's nearly impossible.

You know that CFD has its merits when even NASCAR uses it (to study aerodynamics) (of the entire racing field).

And like it is with crash simulations - the models and simulations have gotten so good that some of the OEMs pretty much do ALL of the development work with FEA, and then when it's finally ready to go, they may only run the actual crash test (which costs anywhere betweem $250k-500k each) ONCE and if they pass - that's it. It's signed off and good to go.

CAN you build a car without CAE? Yes. But can you build a car that gets 54.5 mpg combined without CAE? I would be hard pressed to find that. (Or if you did, it might only seat 1 instead of say...4 or 5).

Another example is the heavy truck industry. I forget how many tens of millions of gallons of fuel they burn every year transporting stuff, but it's up there. And with the cost of fuel (energy) always going up, there are trucks that are out on the road that are like 30 years old. It's so old that there are NO CAD models for them, and definitely NO FE models. (We do some of it here at work).

And now, they're all up in CFD's grill and FEA because they need to milk every ounce of fuel economy they can get out of it.

The other thing that I've noticed -- there are A LOT of people who say they know engines. They can recite specs, and part numbers/engine codes, etc. And probably ALL of them can spin wrenches better than I can; no doubt or question about that. But if I were to ask them what's their optimium bfsc zone, they probably have absolutely NO idea what I'm talking about. Or some might. But I suspect that it might only be a very tiny handful.

Just imagine what it would be like if you were to show up to a Concourse or a street/drag race where your engine has variable compression ratio (because you have an offset crank) or you have infinitely variable valve timing because you're using emag valves, in addition to HCCI with GDI and forced induction where the leading edges of the intake blades are modelled after the GE NX90.

Either the person you're talking is VERY lost (and looks it) or they'll think that it's the coolest thing ever.

And the fact that you have two 3.5" thick binders where you know EVERYTHING there is to know about your engine.

Or another question that I would LOVE to ask someone who says "they know engines" is "did you run a FFT to get your PSD plot for your dynamic NVH characteristics?"

[Liam_G]

So your a theory man, thats cool. I'm a bit of both, I like to get my hands dirty, I learn better by doing something, that's just the way my brain works (PS, I know nothing about car's, they don't interest me, motorbikes and motorbike engines is the extent of my interest in that area, so I understand most of car engines but really only where they are similar to bike engines, which in principle is a decent amount.) But at the same time I love the accuracy and flexibility of designing in CAD, much faster and easier to make amendments to design and you have the added bonus of running simulations. I like to know how things work, every time I see some piece of tech or machinery I want to know how it works, and exactly how it is made, once I have unraveled the mystery I move on.

But I only really enjoy CAD when I know it is going to allow me to build something, maybe you should separate your hobby and work a bit, keep it simple at home, build a steam engine, restore an old car or bike, draw up an existing engine in CAD and then rebuild it, maybe run some simulations to see how to port it and gain some power. It's hard to tell what is home or work with you. The only person who really cares how much you spend on this stuff is you so if that's what you really love to do then go for it!

[tcclaviger]

You sir sound like a perfect candidate for a 3930k, 64GB of DDR 2133, and "By Core" OCing so that you can push say all 6 cores to like 4.8 (5.0 is not at all guaranteed stable on all 6), and then use per core to push it 5-5.1 when using only 2 cores. For thermal reasons and CPU safety it makes far more sense than shooting for 5.0x6core, especially when the app is primarily single threaded as you stated earlier. You may not even need to push it so high, the 3930K at 4.5 is dramatically stronger than the 9550 at 3.2.

Also, for this kind of work load I would stay far away from AMD.

Also, I am not sure you really want to put that swap file on the SSD, sounds like it will be doing an insane amount of operations on it, and at 20gigs, I would expect it drastically effect the lifespan of it. That is why I suggest the massive 64gb ram for a 2011 platform.

As an alternative, can you possibly use a different application that doesn't lean so heavily on the CPU for rendering and pushes more of it to the GPU(s)?

Perhaps something like this:

[alpha754293]

So your a theory man, thats cool. I'm a bit of both, I like to get my hands dirty, I learn better by doing something, that's just the way my brain works (PS, I know nothing about car's, they don't interest me, motorbikes and motorbike engines is the extent of my interest in that area, so I understand most of car engines but really only where they are similar to bike engines, which in principle is a decent amount.) But at the same time I love the accuracy and flexibility of designing in CAD, much faster and easier to make amendments to design and you have the added bonus of running simulations. I like to know how things work, every time I see some piece of tech or machinery I want to know how it works, and exactly how it is made, once I have unraveled the mystery I move on.

But I only really enjoy CAD when I know it is going to allow me to build something, maybe you should separate your hobby and work a bit, keep it simple at home, build a steam engine, restore an old car or bike, draw up an existing engine in CAD and then rebuild it, maybe run some simulations to see how to port it and gain some power. It's hard to tell what is home or work with you. The only person who really cares how much you spend on this stuff is you so if that's what you really love to do then go for it!

I'm a theory man because tooling is expensive. And screwing up is even more expensive.

If I had infinite money, then yea...I'd probably do a lot more of it physically.

I think that there are some things that you can learn more/better by doing simulations than in real life (because sometimes trying to get the data from physical experiments is REALLY REALLY hard) - like the flame front temperature during the power stroke.

I started doing the engine CFD because I think that I was in my junior year of college when I finally thought that I had enough information to be able to design my own engine. Clearly, I don't. (Thermo is a good approximation, but when you're solving for the chemical kinetics (in order to determine the amount of energy, and thus heat, is released due to combustion)) yea...I'm learning wayyyy more about engines than I ever thought I would be. And the strange thing is that I'm not really THAT much of an engines guy either. But it was an application of the technology and I wanted to learn the technology more than anything.

I'm a "car" guy. To the fullest extent/degree. So that ranges from having to draw up the bond graph for the state-space equation system for the entire vehicle, starting from the tire contact patch leading back up to the rest of the vehicle, to doing the suspension kinematics, and simulating the vehicle dynamics, to vehicle packaging and systems integration, to CAD design, to engines, and torque transfer devices (clutches, torque converters, that sorta thing), to transmissions, to drivelines, drivetrain/powertrain, to differentials, to subframe, body and chassis design/engineering.

I've done interiors (that was my co-op job). And as you can tell from the pictures earlier, I've done it with the manikin for human/machine interface. I think that really the ONLY thing that I haven't done is anything electrical/electronic (I'm not a ECE/CS). But if I had any clue about that, I probably would do that too.

And instead of looking at all of those components separately; I look at them together AND separately. Vehicle engineering is a HIGHLY dynamic problem. You one thing somewhere and it's going to affect something else, somewhere else. (Which means that you have to hit the 'recalculate/recompute' button). And that leads all into multi-disciplinary optimizations (MDO).

If you love what you do, the fine line that separates work and play no longer exists. Instead of waking up thinking "aww crap I gotta go to work"; you wake up thinking "I'm gonna go play."

I guess that in the capex list of equipment that I haven't been able to get access too (in addition to the 5+ axis CNC and stuff), the other things that I haven't been able to get a hold of either is an x-ray and also a 3D scanner.

If I COULD get access to a x-ray, I would be doing 2 mm cuts through a cylinder head so that I would be able to reverse engineer it, and then start doing the flow analysis on it.

And my friends and I have talked about doing everything from expandable foam to reverse casting and I've even thought about using jello to extract the port geometry. Reality is - having source CAD is best. (Of course). Follow by sequence of X-ray images. Otherwise, it's a destructive test.

There are people that can do that - build a steam engine, restore old cars/bikes, etc... but the problem that I've discovered is that you ask them a rather simple, basic, question about something that they should or OUGHT to know based on what they've done "hands-on" and they often times, can't tell you. Like I said, half the time, they don't even know what you're talking about. And this is just very simple, basic concepts. Like if someone lowers their car; a very simple, basic question would be what's their new roll center and instant center. A LOT of people who do that (either by themselves, or with friends or pay someone else to do for them) have absolutely NO clue/no idea what a roll center is or what's an instant center.

So, CAN people do that stuff? Yes. Does that mean that they actually understand it? Well...not really. (Because the next step from that is "how does the changes in the instant and roll centers affect your vehicle dynamics?"). And well, if you have no concept of what those things are, you won't be able to answer the follow-up question.

(And it matters because it affects you as the driver.)

I actually took my old '98 Grand Am once autocrossing. A lot of other people showed up in a lot more powerful cars and they were knocking over the little pylons like there was no tomorrow. But beacuse I knew the limits of my vehicle (lack of) performance, at one point I thought that I hit a cone and after I was done, I was talking to a few guys that was standing trackside and they were like "no, you were like an-inch-and-a-half away from it, but no..." and then I checked the times - no deductions.

Or another time where it was raining, and where I live, one of the main streets has a tight left-hander. And I entered the turn a little too quickly, which put the car in a bit of a drift, but because next to me was a mom and 2 kids in a minivan and the curb, I had to drift IN LANE so as to avoid hitting either. I did it. It was an interesting feeling having to drift IN LANE (which actually requires a fair bit of control) - but again, because I understood my car and vehicle dynamics, I was able to do that.

And I've taken my car right to the edge of the tire performance limits/envelop a few times where I can feel it bounce in and out of lateral slip. And this year at the Detroit Auto Show, Ford had one of those driving simulators and I think it was a Fiesta; and the official time was like 54 seconds, and I managed to come in at 56. (Funnier thing was that there were a bunch of like upper teens that were all talking a big game about what they would do and how they would drive it, etc. etc. etc. and when they actually did it, they were coming in at like 1:05, 1:04, and 1:10 or something like that. I jump in. Didn't really say much. (They were complaining about how the car sucks and it's got like no power and no brakes, etc. etc. etc.) and then I jump in, and the first thing I do is get a feel for the acceleration and braking and manage to pull in a 56 on the first try. So...it helps. (I've had something similar happen to me where some of the guys took me to a go kart track that I've never been before and I think that I was like a second behind the lap record or something; and the guys were like "Have you been here before?" and I'm like "no." and they were quite surprised for a first timer. (And I was like doing some stupid on purpose cuz it's supposed to be fun as well; although if I weren't messing around so much, I probably could have done even better.) They were like..."haha...we should put you in the formula car and see what happens." and I'm like..."no no no no no...." fun times.