Brilliant videos Particle!
I really like your case and how you manage cables!
It looks so familiar to my own computer! Only difference is that I have some HDDs on the floor because they can't fit into my case (TT Soprano) :p:!
Keep up good work!
Printable View
Brilliant videos Particle!
I really like your case and how you manage cables!
It looks so familiar to my own computer! Only difference is that I have some HDDs on the floor because they can't fit into my case (TT Soprano) :p:!
Keep up good work!
24464 was score with 12 cpus so that's like each cpu is getting 2038.666. on avg anyways. what is the speed up now that u have 1 ram in each cpu atleast?
Scaling is a bit off ,around 8.6x iirc from his previous post.So the one core score should be ~2800.Quote:
Originally Posted by i found nemo
"it takes a special kind of person to be excited by this"Quote:
Originally Posted by Particle
lol I know what you mean! Thanks of sharing you experience. :up:
This is true. Still, the scaling seems very low to me right now. I guess Particle has some tweaking to do with the memory or somethings. I would expect ~10x scaling at least. 28% is a lot of performance to be losing don't you think? Istanbuls are supposed to be very efficient in multi-CPU setups.Quote:
Originally Posted by informal
I think it was since he was running the board with one dedicated memory (for CPU0 only).Quote:
Originally Posted by Mechromancer
not in the vid. tha's AFTER he switched rams. but the post he had in here earlier was when rams were on cpu0. i'm just wondering how much, if any cpu scaling went up after giving each their own mem. ( even if single channel ) i get 2200 pts with single core @ 2.64 ghz ram @ 264 3-4-3-10
the poor scaling is the software not the hardware in this case. if the scene was larger and more complex you would see a much larger increase in performance
AMd talks a lot about their overall platform, it would be awesome if we saw overclockable dual-socket mobos supporting Istanbul.Quote:
Originally Posted by Tom128
Actually, it would be awesome just to see desktop Istanbul :D
watching the video you can quickly see how badly C10 is at trying to balance core loads. if they didnt try to render a whole row at once, it would probably scale much better.
Yes, like many people (including myself..) have said earlier in this thread, CB10's scaling doesn't work properly for more then 8 threads. The engine is showing its age, a way better test is, for example, Frybench
It scales with ram speed as well, which means my Dual Gainestown totally owns the Dual Harpertown here, even tho in CB10 the Harpertown wins. Try it particle, under 4 minutes is good :D
Guess that's not something my Opties are good at. Render time was 5m26s at 12 x 2508. If it's memory sensitive though, it might be suffering from my single channel DDR2-800 @ 780 with vanilla timings.
hi particle,
i havent posted in this thread yet, but i have been following it, i just wanted to say that has been very interesting to see this evolve into 12 core goodness, and too see the results,
i was also very surprised to hear that you sound exactly like you post, if that makes sense?
keep the results coming, nice to see a thread in the AMD section that has some meat in it.
Just to put your time into perspective, on my C2D T5500 (1.66GHz) laptop in 32bit Vista I've got 45min. 42sec. :eek:Quote:
Originally Posted by Particle
Very nice bench I must admit!
Thanks jcool for linky :up:
Your CB10 score is fine:
http://www.techreport.com/r.x/optero.../cinebench.gif
I'd get at least two ram sticks for each socket stat, Particle, otherwise you were better off with Shanghai :stick:
Perkam
Just to make sure everyone knows: To all of you who have been supportive and extending compliments--I really do appreciate it. Thanks a bunch folks. You guys are what makes this project fun to do in the first place. :up:
As for RAM, that has been and still is on my list of to-dos, but I'm going to wait until I get paid this Friday. :) I've spent a boatload of money already, and I'm below my normal reserve line due to some problems encountered during this build. The RAM I use is frequently out of stock anyway, and Newegg doesn't predict they'll have more until Friday so that times up nicely. Most ECC/Reg is 667 or slower. I've got the fancy shmancy 800 stuff! (Woo, fancy $35 RAM ftw) hehe I only bought two sticks initially so I could run in a minimal state if my mountains of regular RAM didn't work (it didn't).
vCore
I do still need some help on this one. I'd like to get it up to 1.30V and try my luck overclocking a little higher. I still haven't found a way around the (designed) behavior of the system to where if a VID above MaxVID (1.225V) is set, the system ignores it and re-writes 1.225V to the CurrentVID register. I need a way to overwrite the MaxVID register or something, whever it is. I emailed one of those Crystal CPU warriors from the TLB adventure, but he either hasn't noticed my PM or isn't interested. A hardware mod isn't out of the question, either, provided anyone knows how to do it. We need to get this truck rolling at 2.8 or 3.0 GHz I think. :)
Yep! Those are the CB10 scores I'm comparing his thread to. Now that 2435 is running 100mhz faster, but I noticed that CB10 is sensitive to memory speeds. When I went from 4GB@1066 to 8GB@800 there was a noticeable decrease in rendering time. At 800mhz I was able to tighten up my timings quite a bit and that helps too.Quote:
Originally Posted by perkam
Particle is a couple of good DDR2 ECC 800mhz kits away from extracting full power from those Six-Cores.
Particle, I actually want to buy two of the 4GB Kingston ECC 800mhz kits for my PII-940! ECC@800mhz > regular@1066 Iv'e found. How tight can you get the timings on that 800mhz ECC?Quote:
Originally Posted by Particle
I haven't played with the timings yet at all. I could see ECC memory in theory granting more overclocking headroom since it can correct single bit errors if they don't happen too frequently, I suppose. Not sure I'd want to be that far out on the edge though! :D You could probably use the ECC memory just fine if your motherboard exposes the options, just not ECC + registered. I'm not sure what the registered buffer does to overclocking stability.
Very interesting thread!:up:
It actually made me take a look at pricing on the Shanghai and Istanbul Opterons, and while two hexacores seem a little too rich for me price wise, I actually found two Shanghai 2376s to have very good pricing. I think I'll hold off upgrading for a few months to see if the usual suspects (Supermicro?) will release dual socket boards based on the new AMD chipsets, and how the Lynnfield release goes.
I'd love to see some Arma II benchmarks Particle, my 3,8 GHz E7200 suffers badly when there is a lot of AI, I need more cores I think:(:D I'm actually CPU limited in gaming at 1920x1200 with a 1 GB 4870 and a 3,8 GHz Core 2 Duo, who would have thought?
Thanks, Frodin. I too am hoping for some 800-series chipset boards this fall. As for Arma II, I have a bit of bad news there. It appears to be bugged out--it won't run with 12 cores in a system. It'll run with 6 cores/1 CPU just fine and a guy on a skulltrail system can run it (2 quads), but with 12 cores it just dies when the menu should be loading up.
PARTICLE! I have news from the PhenomMsrTweaker thread!
Give that a try and let us know if the service allows you to change core VID :D. Also, post in that thread so kink knows just who the heck you are. I seriously don't know how he missed your Istanbul thread.:rofl:Quote:
Originally Posted by kink
Yes, it is very memory intensive. My Gainestown at 3Ghz with Tri Channel DDR3-1066 C7 takes around 3:30-3:40 to complete.Quote:
Originally Posted by Particle
And here is the Harper on its 24/7 setting, which you managed to beat even though a pair of 4Ghz Harpers would be faster in most other apps. Even though they got QuadChannel FB they are somewhat limited by the FSB interface, even at FSB450.
Now I see why you don't like speculation, you have so much factual info :stick: :p:Quote:
Originally Posted by jcool
There is definitely some potential there in these Istanbuls. Cost/Core doesn't become competitive with Intel, though, until the Octa and Deca-core processors come out next year.
Perkam
do you think you have enough ram to run games?Quote:
Originally Posted by Particle