Why does phenom perform better than an intel quad at high resolution in Lost Planet?

[Eson]

So BenchZowner, you've come here all drenched in gasoline to piss in AMD territory. I guess you asked to be flamed.

First of all, very impressive confidence you have!
Althought you knew you werent welcome here you still thought Jakko was trying to impress you, as if you thought people were looking up to you.
While Jakko nicely told you that he didnt want an other Intel vs. AMD flame wars, you imedietly start to insult him and flame AMD. Very good start for a nice topic.

You state several times that you dont care yet still you are trying to convince us that we are wrong and you are right.
However thats where you lost me. You claim that it doesnt matter since the game is GPU limited but its wrong to say AMD is just as good as Intel.
As if the 'GPU limited' argument is only valid when AMD got the upper hand and thus making the 'win' nullified. While for Intel it counts as a 'win'.

All your posts are 98% off topic because this topic was about WHY PHENOM WAS IN A LEAD AT SOME POINT IN THIS BENCHMARK!
Not whatever the benchmark is GPU limited or not, the Phenom is still in the lead at some point in the bench and so why is that?

With the results of your actions meny people propably have come to the conclusion that you are simply a less intelligent troll, hence why they call you stupid.
Althought some of your post where not insulting or flaming and showed understanding of a nice discussion. You still seem to have a long way to go.

[boogle]

You were obviously talking about Phenom vs Core 2 Quad.
You were talking about it in a thread that some people believe that Phenom runs faster in real-life conditions a specific game.
And you say "Since the Phenom is faster in this game, what makes you think that it may not be faster in another game?"

He didn't say that. He said that in any given competition, even if one contender seemingly wins all the time, there will still be occasions where the other contender will triumph given the right circumstances.

He didn't say one was better than the other, he didn't even try to predict how one contender will do. He simply attempted to answer the question posed in this thread: Why is this contender beating the other contender in this paticular area? The answer being, of course, that given the right circumstances any contender will win.

When coming into a discussion, please try to remain neutral and leave your own personal world view at the door. Otherwise it'll twist what you read and then make people think you're stupid because you seemingly can't read.

[savantu]

...

So here the native quad advantage shows up -- from single thread to 4 threads, the scaling factor of the phenom scales with respect to the single threaded performance better.

Jack

Scaling doesn't matter.

This is a must read :

http://www.realworldtech.com/forums/...79593&roomid=2
http://www.realworldtech.com/forums/...79593&roomid=2
http://www.realworldtech.com/forums/...79593&roomid=2

and the masterpiece :
http://www.realworldtech.com/forums/...79593&roomid=2

Don't know if you've came across this thread ; but it's worth reading .

[systemviper]

Hi, i want ONE amd system and i am thinking of a Foxcon motherboard,
any help which is he best Foxcon MB and which is hte best chip.

regards
steve

[Jakko]

Scaling doesn't matter.

This is a must read :

Scaling does matter I think.
A quote from the article you posted:

It's not that Intel "scales badly", it's that Intel does
so much better on single-core benchmarks thanks to
the bigger usable cache.

What this person forgets is that it doesn't really matter why something scales or doesn't scale.
Yes intel uses a cache intensive system that improves single core performance dramaticly, and doesn't do that much for adding more cores, but really what does this have to do with anything?

The final result is still good or bad core scaling.

Next time I hear somebody talk about "bad scaling", when
it's just that the performance is higher to start with,
I'll find a chainsaw and do unspeakable acts to the
idiot. Scaling is meaningless.

Next time I hear somebody talk about AMD's poor single core performance, when it's just that the scaling is better in the end, I'll find a chainsaw and do unspeakable acts to the idiot. Single core performance is meaningless

Sounds ridiculous eh?
And it is.

[JumpingJack]

Scaling doesn't matter.

This is a must read :

http://www.realworldtech.com/forums/...79593&roomid=2
http://www.realworldtech.com/forums/...79593&roomid=2
http://www.realworldtech.com/forums/...79593&roomid=2

and the masterpiece :
http://www.realworldtech.com/forums/...79593&roomid=2

Don't know if you've came across this thread ; but it's worth reading .

Scaling is relative in the context that I state. The Phenom scales over the threads better, this is what the data says... no arguing that.

However, if single threaded performance has a 20% lead, and the Phenom only scales 5% better utilizing 4 ... the overall absolute performance is still overwhelmingly lopsided.

The only thing the scaling for each shows is that the native approach yields better scaling... for absolute performance, it is not near enough to catch up.

Whether it is important or not -- it is only important to me as a learning experience to help understand how CPUs work, that is the short of it. Both AMD and Intel are bound by more constraints than 'let's make the faster CPU'... the factors going into design decisions include -- time to market, die size, costs, yield, competitive forces, etc. etc. While AMD is techologically more 'elegant' (I will not argue against that), it is Intel that has the technologically important characteristic -- good raw absolute performance, quad core 1 to 1.5 years early, etc. etc. and the effects are (unfortunately) showing up in AMD's bottom line.

Considering this is an AMD section though, this is not about -- Intel trouncing AMD ... not at all, the Phenom is a fine CPU, good performance, and AMD has priced it appropriately such that it is a good purchase for anyone who want to purchase AMD for the sake of purchasing AMD -- or for what ever reasons -- I would not criticize someone for choosing/taking this processor for example.

Jack

[Jakko]

Oh and thanks Jack for wanting to do that test in the future.
I do have a 790X board with a brisbane 5000+ BE, and similar ram to what you used, but sadly I only have a 3850 at the moment, probably not enough to make the snow level gpu limited in the way the gpu was the limiting factor in your benchmarks.

What I find strangest is that

1. at very low resolutions (completely cpu limited) intel wins
2. at a middle ground (high resolution, standard details, gpu limited) amd wins
3. at the very highest details and resolutions, it's a tie.

I could understand how, when the gpu has to work very hard(#3), bandwidth requirements between the gpu and cpu are lower than a scenario in which the gpu can pump out much more FPS at a fairly high resolution, due to limited detail level.(#2)

But it stops making sense when intel wins at those low resolutions. Isn't that the situation in which the bandwidth/architecture advantage seen at #2 should show as well?

[JumpingJack]

Oh and thanks Jack for wanting to do that test in the future.
I do have a 790X board with a brisbane 5000+ BE, and similar ram to what you used, but sadly I only have a 3850 at the moment, probably not enough to make the snow level gpu limited in the way the gpu was the limiting factor in your benchmarks.

What I find strangest is that

1. at very low resolutions (completely cpu limited) intel wins
2. at a middle ground (high resolution, standard details, gpu limited) amd wins
3. at the very highest details and resolutions, it's a tie.

I could understand how, when the gpu has to work very hard(#3), bandwidth requirements between the gpu and cpu are lower than a scenario in which the gpu can pump out much more FPS at a fairly high resolution, due to limited detail level.(#2)

But it stops making sense when intel wins at those low resolutions. Isn't that the situation in which the bandwidth/architecture advantage seen at #2 should show as well?

Well, it is hard to draw generalized conclusions ... this is one game, with two different scripted scenes ... one thing to take away from this data set is that a gaming experience is really dictated by a complex set of factors (I am diverging here from a CPU/architectural discussion true).

In any given benchmark, at any given set of settings the GPU may be important, the CPU may be important ... overall, it is the GPU that determines the experience for the most part -- choosing the CPU should be done carefully though, either CPU in today's gaming environment will handle any game (only possible exception is Crysis). It is hard to declare any CPU a 'winner' with such a very narrow subset of conditions/scenes/rendering conditions.... the best way (impractical way) would be to determine through the entire story line of a game, how often either CPU limits the gaming out put to below the playable FPS.... this is simply not possible, i.e. you can never guard against the 'worst case condition' as none of the review sites nor any of the built in benchmarks can possibly account for worst case in actual game play.

Back on topic -- I think you are right, ultimately the BW argument will ultimately be the appropraite explanation, be it at the HT (analogous FSB) point, or a the PCIe point of connection.... if we were to generalize a statement, it would fall out very similar to what has been said time and again -- CPU limited work loads, Intel has a stronger core -- throughput limited workloads AMD's interconnect shows it's advantage.

jack

[rozzyroz]

Scaling doesn't matter.

This is a must read :

http://www.realworldtech.com/forums/...79593&roomid=2
http://www.realworldtech.com/forums/...79593&roomid=2
http://www.realworldtech.com/forums/...79593&roomid=2

and the masterpiece :
http://www.realworldtech.com/forums/...79593&roomid=2

Don't know if you've came across this thread ; but it's worth reading .

I dont own an amd system atm, but i always come to visit this forum to see how things are going. the posts you linked to are very enlightening to me (I respect LT's views), but looking at the setup, I don't think its a scaling advantage. with the test being completly GPU dependent, I would think that the performance advantage has to be attributed to the motherboard/ chipset in some way. it could be the amd system getting a boost somehow, or the intel system being hindered in some way.

perhaps running the intel system on a different chipset to see if there is a similar falloff at higher levels, and run the amd system on a different chipset to see if the performance curve is the same.

[mAJORD]

Actually, yes... I just picked up a i9300 myself, so getting someone to run benches to check reproducibility across different configs would be interesting to.

What MB, memory, OS?

EDIT: What I am finding (initial info) is that the 6 Meg (1/2 the cache) at the same clock has a bigger impact than what I would have thought.

Mem: Ballistix 1066 @ tbd timings, maybe do some runs on both systems @ diff timigns

Mobo: Intel: GB EP35-DS3L , AMD: GB 790FX

Vid: 3850 512mb

OS: Vista, edit SP 1 + no additional updates, should be better
I'm installing clean OS on both systems, exactly the same. Defragging between prog installs.

[boogle]

Scaling doesn't matter.

This is a must read :

http://www.realworldtech.com/forums/...79593&roomid=2
http://www.realworldtech.com/forums/...79593&roomid=2
http://www.realworldtech.com/forums/...79593&roomid=2

and the masterpiece :
http://www.realworldtech.com/forums/...79593&roomid=2

Don't know if you've came across this thread ; but it's worth reading .

I don't agree with his conclusions whatsoever, he's clearly talking in a server context (and even on a desktop I still don't agree). He's saying if you add more raw power to a system, it shouldn't scale linearly. That's not very cost-effective is it?

Additionally, he's saying that the CPUs should be matched to the infrastructure. By that logic, we should be back with 386s, since there's no chance a HD will be able to saturate a faster CPU. Even a 386 might be too fast.

I know he's getting at Intel being faster in single-threaded apps is a good thing, and that scaling beyond that, it's still faster than AMD. That's fine, and most definitely true. But trying to pretend that scaling isn't important with multi-CPU server/super-computer situations is a joke. Incidentally in servers Phenom beats Intel handedly, seems that scaling suddenly matters.

[Devil's Prophet]

I don't agree with his conclusions whatsoever, he's clearly talking in a server context (and even on a desktop I still don't agree). He's saying if you add more raw power to a system, it shouldn't scale linearly. That's not very cost-effective is it?

More power as in more cores? More parallel power in situations that need serial power? Or what are you saying that isn't right?

Additionally, he's saying that the CPUs should be matched to the infrastructure. By that logic, we should be back with 386s, since there's no chance a HD will be able to saturate a faster CPU. Even a 386 might be too fast.

That's why they invented RAM and even RAMdisks. Now solidstate is also getting closer.

I know he's getting at Intel being faster in single-threaded apps is a good thing, and that scaling beyond that, it's still faster than AMD. That's fine, and most definitely true. But trying to pretend that scaling isn't important with multi-CPU server/super-computer situations is a joke. Incidentally in servers Phenom beats Intel handedly, seems that scaling suddenly matters.

Doesn't he mean the absolute numbers are more important than the scaling numbers on their own? Just like you say in the above of this quotation.

Anyway, are you sure you're not pulling his words out of context in a charade of sarcasm? Because in my experience that happens A LOT in here.

[boogle]

More power as in more cores? More parallel power in situations that need serial power? Or what are you saying that isn't right?

Doesn't he mean the absolute numbers are more important than the scaling numbers on their own? Just like you say in the above of this quotation.

Anyway, are you sure you're not pulling his words out of context in a charade of sarcasm? Because in my experience that happens A LOT in here.

From my understanding he's saying that you should have loads of performance with whatever the software was initially designed for, and the more nodes you add the less of an improvement you should get (diminishing returns).

Being a software developer he should know rule #1 with server apps: The software should scale with an infinite number of CPUs.

I agree with his comments about absolute power, especially with Intel favouring single-threaded apps - this is a good plan from Intel. However, when he starts talking about scaling shouldn't scale evenly, he's lost me. Completely. If you don't scale evenly, there's a bottleneck somewhere and that bottleneck needs to be resolved. Its NOT a good thing!

[Oldguy932]

Not sure if it applies to this thread, but I'm going to be getting the parts in on wednesday to build a Q9450/X48 system to compare it and the 9850 at crossfire and crossfire alone.

Intel system is...
Q9450 cooled by TRUE with Scythe Slipstream 1900rpm 110cfm
Asus Rampage Formula
Kingston 2x1gb DDR2-1066 5-5-5-15 1.8v
2x Sapphire 3870's in crossfire
Seagate Barracuda 250gb
Windows Vista Ultimate x86

AMD system is...
9850 BE (my week 811) cooled by TRUE with same fan
Asus M3A32-MVP Deluxe Wifi, bios 1101
Kingston 2x1gb DDR2-1066 5-5-5-15 1.8v
2x Sapphire 3870's in crossfire
Western Digital 250gb with perpendicular recording
Windows Vista Ultimate x86

Only differences will be the case, hd, and cd rom other than the necessary cpu/mobo in order to minimize variables. I'll just do tests at 2.66 and 3ghz because my board doesn't let me take the 9850 higher than 3ghz stable even though it does it at 1.325v.

[gosh]

AMD performs better on high res
Intel Q9450 vs Phenom 9850

http://www.overclockersclub.com/revi...l_q9450/14.htm
http://www.overclockersclub.com/revi...l_q9450/13.htm
http://www.overclockersclub.com/revi...el_q9450/8.htm

AMD scales better using more cores
Processor performance list

[roofsniper]

as far as games go cpus really don't have that great of an effect on fps. the difference might be by just a few frames. mostly the difference is something that the computer can't tell you. its something you will have to see for yourself. now you can get an intel cpu clocked 4ghz and get a phenom clocked at 3ghz. the intel will have a higher fps but does it mean its better? if you actually look at the game instead of the fps you will notice that the amd system runs a lot smoother even tho the fps is lower. benchmarks and results can't be the only proof of why something is better. in this case the benchmark favors one but you can look at it and clearly tell the other is better.

[gallag]

now you can get an intel cpu clocked 4ghz and get a phenom clocked at 3ghz. the intel will have a higher fps but does it mean its better? if you actually look at the game instead of the fps you will notice that the amd system runs a lot smoother even tho the fps is lower.

Is fps not what makes a game smooth? I can see the argument in saying windows is smoother etc but in a game does fps not define smoothness?

[Devil's Prophet]

Is fps not what makes a game smooth? I can see the argument in saying windows is smoother etc but in a game does fps not define smoothness?

Remember microstuttering with SLI/Crossfire setups? Maybe this is something equivalent.

[BenchZowner]

Holy thread-digging batman

As for the "lower FPS but smoother"... please, don't kill me ( that's definitely funnier than this phrase ).

When you have:

Lower minimum FPS
Lower average FPS
Lower maximum FPS

And probably bigger drops from average FPS to minimum FPS...how can you feel that the game runs smoother than the other setup with higher min FPS, avg FPS and max FPS and less fluctuation ?

Stop beating on a dead horse

and... like we said... in GPU Limited scenarios ( most of the games [ if not all ] when run under "normal gaming" settings ) the processor plays little to none ( none most of the time ) [ if you're running anything better than a A64 3500+ / P4 3.4GHz ] role.

[gosh]

Is fps not what makes a game smooth? I can see the argument in saying windows is smoother etc but in a game does fps not define smoothness?

How smooth the game is depends on how evenly frames are displayed. If you have 100 FPS but 90 of those is displayed on the first half of the second, 10 the second half. You have another computer that has 30 FPS but those are evenly displayed. Then the 30 FPS computer will be smoother than the 100 FPS computer.

Intel has one schizophrenic behavior. It is very fast when data is in the cache, it has high frequencies. But when it needs to communicate with other hardware it isn’t as fast, the latency goes up. On AMD the processor isn’t as fast but the processor is built to scale. Cache, hypertransport, IMC etc. Areas that will enable AMD to scale evenly.
You could say that AMD has worked on their weak parts. AMD don’t have any bottlenecks. Intel has worked on their strong parts but the bottlenecks still exists.

[gallag]

How smooth the game is depends on how evenly frames are displayed. If you have 100 FPS but 90 of those is displayed on the first half of the second, 10 the second half. You have another computer that has 30 FPS but those are evenly displayed. Then the 30 FPS computer will be smoother than the 100 FPS computer.

Intel has one schizophrenic behavior. It is very fast when data is in the cache, it has high frequencies. But when it needs to communicate with other hardware it isn’t as fast, the latency goes up. On AMD the processor isn’t as fast but the processor is built to scale. Cache, hypertransport, IMC etc. Areas that will enable AMD to scale evenly.
You could say that AMD has worked on their weak parts. AMD don’t have any bottlenecks. Intel has worked on their strong parts but the bottlenecks still exists.

I agree, If a game was running at 90fps then 10fps it would be very bad, But Intel cpus very ofter have a higher min fps so how could it seem less smooth?

[gosh]

I agree, If a game was running at 90fps then 10fps it would be very bad, But Intel cpus very ofter have a higher min fps so how could it seem less smooth?

I don’t think that you could see this effect on games that isn’t very new (are using threads and is heavy on the graphics). There might be situations on C2Q and if the game has more than one thread. If one thread is moved to another core (on another C2D, C2Q = 2 x C2D) and cache is invalidated then performance will go down because data used would need to be fetched from memory again. If the other thread is sending and retrieving data also then this will be slower, Intel is using the FSB for memory and I/O, more traffic and there is greaer risk for conflicts.

There could be situations where you have other programs running on the computer, these programs doesn’t use that much power (then are mostly resting) but sometimes they might and that will hurt Intel more.

[Tony]

I talk to a few tier 1 reviewers who do a ton of testing, one reported right after Phenom was launched games average frame rates were slower but the games ran smoother, now I don't take that much notice while testing so i just ask more reviewers to see what they feel....and they all say the same.

Now I thinks it not only CPU, I think its total platform and AMD have been working really hard on this for the past 18months. 4XXX GPU's have suddenly come good, ATI's driver dept look to be getting a move on to take the cards to new levels. I actually feel the driver writers were being held back by the 3XXX series cards and are now starting to get it together as they have more to work with.

So, AMD are just building a platform that plays games well, they dropped the need to compete in super pi etc and just concentrate where they think end users feel the system should perform.
I think it will get even better with the new GX chipset, all we need is someone to test it for us vs the older FX chipsets.

[roofsniper]

Is fps not what makes a game smooth? I can see the argument in saying windows is smoother etc but in a game does fps not define smoothness?

there are many things besides fps that make a game smooth. everyone here is familiar with tearing right? where you can even have TOO many fps and the game will not be smooth. just like gosh said its more because of the features amd has. no they may not have higher clocks and be running slower but they get it coming out at a steady pace and make it smooth. think of it as a fast reckless driver compared to someone just driving along. reckless driver will most likely be braking and accelerating over and over and changing lanes while the normal driver will be just going along. the jump from high fps to lower fps will make the games not as smooth. even if its a jump from 50-70 where fps is still high. so even tho intel pumps out more frames amd has them coming out at the same rate the entire time which is all a cpu should be doing anyway

[soundood]

I think there are Valid points on both sides,

one sais single otimized scaling, one sais 4 core scaling beyond your dreams,

let me ask you this, and i beg you to answer honestly Intel urinal stalker/AMD trouser sniffer,





how are they with teh pron?