official 3D MARK 11 Score Thread

[Qb2k5]

I just find it interesting in order to reach the 10.5K+ Performance mark you need to be running atleast two gpus.

[Nunz]

Good, means it's actually a benchmark and not just a quick ugly stress test. It looks good and pushes my GPUs, that's all I care about.

[moshpit]

Originally Posted by Nunz

Good, means it's actually a benchmark and not just a quick ugly stress test. It looks good and pushes my GPUs, that's all I care about.

+1

While I only have a single GPU, and will only run singles from now on, I do agree on the wanting a test that looks good and pushes my GPUs nicely.

[demo]

P16802

http://3dmark.com/3dm11/1192579

[shadow001]

Originally Posted by demowhc

P16802

http://3dmark.com/3dm11/1192579

Impressive score indeed, but i'll be coming for you soon enough as i'll be installing the water blocks on the GTX 580's this monday.....


If you're wondering, both 6 core Xeons are merrily running at 4.3 Ghz with just 1.37 volts and even under full load for all 12 cores for hours in linpack, never exceed 65*C and the fans are only operating at 60%, so the system is pretty quiet overall...

[Barnzz]

P6473 - 1 x 6970 - screenshot only as futuremark servers were down at the time

11345 - 2 x 6970's - http://3dmark.com/3dm11/1193720
14831 - 3 x 6970's - http://3dmark.com/3dm11/1190982

[demo]

Cheers buddy

I was wondering how your build was going, cant wait to see some scores and more pics

[shadow001]

Originally Posted by demowhc

Cheers buddy

I was wondering how your build was going, cant wait to see some scores and more pics

http://i765.photobucket.com/albums/x...s/DSCN0049.jpg


How it is currently, but there's still some wire cleanup~management left, but the radiators at the front for the video cards are permantly installed, and i'm just wating for these to arrive :





If you're wondering, i'm already doing close to P16000 without overclocking the video cards.....Imagine when they will be

[demo]

Nice blocks, I was considering the EK blocks cause they are cheaper but I see you spare no expense.. you do it right

Look foward to some scores, Ill blow my shiit up trying to beat you

[shadow001]

Originally Posted by demowhc

Nice blocks, I was considering the EK blocks cause they are cheaper but I see you spare no expense.. you do it right

Look foward to some scores, Ill blow my shiit up trying to beat you

Well, the blocks are installed and ran a couple of sessions to see where the limit is in terms of clocks for the cards, and so far i got up to 925Mhz/2200 Mhz on all 3 cards at 1.138 Mv and running the heaven demo in a repeating loop, and the fun part is that with the radiator fans and water pump set at maximum speed(it does become noisy like this), load temperatures for all 3 cards were 60*C even...


I did see a little bit of artifacting in some areas every now and then, so the cards might be giving me a message that despite the temperatures not being a problem, they can't take much more or i need to find a means to increase the Voltage to the GPU core, as the cooling can handle it....We'll see.


I'm already happy with a 150Mhz GPU overclock on the cards at such low temperatures anyhow...

[shadow001]

Still in descovery of what this setup can do though, but not bad for a first run..


http://3dmark.com/3dm11/1220414;jses...pLAFMCbCpbmNqn


P16194

[Roadhog]

Originally Posted by shadow001

Still in descovery of what this setup can do though, but not bad for a first run..


http://3dmark.com/3dm11/1220414;jses...pLAFMCbCpbmNqn


P16194

Your score is extremely low for 3, 580's.

[moshpit]

Originally Posted by Roadhog_

Your score is extremely low for 3, 580's.

Now don't you two start that up over here too! I've seen the Nvnews thread with you two going round and round about Furmark testing. To be honest though, I do agree with the need for using it as a stability stress test similar to Prime95 or OCCT when overclocking.

And besides, not like I can really bitch at anybody for going round and round with shadow001, I do so regularly

[shadow001]

Originally Posted by moshpit

Now don't you two start that up over here too! I've seen the Nvnews thread with you two going round and round about Furmark testing. To be honest though, I do agree with the need for using it as a stability stress test similar to Prime95 or OCCT when overclocking.

And besides, not like I can really bitch at anybody for going round and round with shadow001, I do so regularly

It's your favorite hobby, don't deny it...


Anyhow, like i said in the other thread, i'm still very much in discovery mode just how fast the setup can be and that was just the first round...There's more to come.

[moshpit]

Originally Posted by shadow001

It's your favorite hobby, don't deny it...

Arguing with anybody is my favorite hobby, you're just my newest project

[Roadhog]

Originally Posted by moshpit

Now don't you two start that up over here too! I've seen the Nvnews thread with you two going round and round about Furmark testing. To be honest though, I do agree with the need for using it as a stability stress test similar to Prime95 or OCCT when overclocking.

And besides, not like I can really bitch at anybody for going round and round with shadow001, I do so regularly

No, but look at demowhc's score.

http://3dmark.com/3dm11/1192579

What I was trying to say is that shadows score should be about the same as that, which its not.

[moshpit]

Originally Posted by Roadhog_

No, but look at demowhc's score.

http://3dmark.com/3dm11/1192579

What I was trying to say is that shadows score should be about the same as that, which its not.

To be honest, it looks right to me. Shadows CPUs are reading as a flat 4Ghz, while demowhc is running 4.62Ghz, significant CPU speed difference adding up to a minor 600 point difference. The core count on shadows machine is meaningless in this comparison since he's WAY beyond what the benchmark can use. Another example of why dual socket wasn't the best choice for this rig unless he plans on some serious Db mining.

I can promise he still doesn't get full use even in the most intense encoding jobs, one whole processor socket worth of cores sits idle without batch processing of many transcoding jobs simultaneously. I don't think he's got that much encoding work to do. One would need a full time job of transcoding media for that many cores to be even REMOTELY useful to encode/decode work.

The GPU clocks are close enough in the comparison of both rigs that this can PURELY be chalked up to having severe SMP overhead issues with that many cores (or possibly is just constrained how far he can overclock his CPU). THIS is what I warned him about. BUT, seeing as 600mhz of CPU speed MAY bring him even with the 600 points he's behind, I don't see a real problem with his score.

I've said it before and will say it MANY times again (mainly because it's fun jacking with shadow ), this build was WAY off target for the stated point of it lasting a long time. It won't last any longer then a single socket system with just one of those CPUs before it feels like a dog.

[Roadhog]

Originally Posted by moshpit

To be honest, it looks right to me. Shadows CPUs are reading as a flat 4Ghz, while demowhc is running 4.62Ghz, significant CPU speed difference. The core count on shadows machine is meaningless in this comparison since he's WAY beyond what the benchmark can use. Another example of why dual socket wasn't the best choice for this rig unless he plans on some serious DB mining.

I can promise he still doesn't get full use even in the most intense encoding jobs, one whole processor socket worth of cores sits idle without batch processing of many transcoding jobs simultaneously. I don't think he's got that much encoding work to do.

The GPU clocks are close enough in the comparison of both rigs that this can PURELY be chalked up to having severe SMP overhead issues with that many cores. THIS is what I warned him about.

Maybe his GPU memory is overclocked too far?

[moshpit]

A little factoid that TICKLES me about shadow's build. I built dual and quad socket Xeon machines FOR the express purpose of transcoding live video streams and storing them. I'm not remotely fooled into believing he built this rig with encoding jobs in mind. Fact is, he's WAY overdone it on processor core count for even a 32 camera, H.264 surveillance setup. I built many many many of those, and can even claim credit for having designed Salient Systems RM1000R3 systems when I worked there, not to mention directly influenced the design for the RM1500R3.

I'm happily retired now

[moshpit]

Originally Posted by Roadhog_

Maybe his GPU memory is overclocked too far?

I doubt it, but could indeed be a factor. I honestly think the 600mhz difference looks more like a 600 point difference to me, but memory throttling could certainly play a possible secondary, or even the main role. Hard to say.

[demo]

Actually Roadhog is right. Shadows score is low in the GPU department for the clocks, only hitting 17k while I and others are in the 19-20k range at similar clocks. The overall score isnt bad though and 3Dmark11 likes CPU and RAM... I recently got a 500 point boost just from RAM timings.


Your CPU score is very good Shadow even without knowing the RAM timings, perhaps the GPU oc actually reduced score to due error checking, or the dual nf200 chips are rearing their ugly heads..

[shadow001]

First of all, and this is something everyone is ignoring even though i mentioned it already this is just the first run i made and less than a day after i installed the water blocks for the video cards, and yes there's plenty of tweaking room left as the setup is completely stable at these clocks over several hour periods running anything....


Let's break this down to actual figures shall we?...Bolded for who's higher.


Graphics score....17924 for me and 19044 for demowhc...just a little over 1000 points extra for him.
Physics score....14550 for me and 13047 for demowhc.
Combined score...10422 for me and 11581 for demowhc.


As for the actual figures,there's only a 3 to 6 Fps difference in graphics tests 1/2 and 4, and only in test 3 that gap grows to a more significant 10 Fps, but on the other hand, my GPU's are 30Mhz slower, and when i do that(and i will), the gaps will become even smaller regardless of who wins or looses overall.


As for the physics test, i have a 6 Fps lead there, wich proves the point that mmore than 12 threads are used, otherwise i wouldn't be able to beat an i 970 CPU clocked 600 Mhz faster, now would i?


In the combined test, he leads by 5 Fps, but again to be apples to apples comparison, i have to match the GPU clocks to his run, and win or lose the gap will become smaller still....It's basically win or loss margins that will probably end up extremely narrow...


My longer term goal is to hit 1 Ghz on the GPU's since the water cooling can handle though, but a bios flash for the video cards may be needed so i'll have to reaserch that further....Like i said at the top of the page, this is only the first run of many to come over the next weeks and months...

[shadow001]

Better now?...And guess what, the clocks, or Bios settings or windows settings didn't change at all....Only a driver change to the latest beta:


http://3dmark.com/3dm11/1223730;jses...Muq z0U70DeTU


16800


What does that say about your hardware conclusions in those last few posts of yours Moshpit...

[Roadhog]

Originally Posted by shadow001

Better now?...And guess what, the clocks, or Bios settings or windows settings didn't change at all....Only a driver change to the latest beta:


http://3dmark.com/3dm11/1223730;jses...Muq z0U70DeTU


16800


What does that say about your hardware conclusions in those last few posts of yours Moshpit...

Bro, you lowered your memory clock speed too.

[shadow001]

Originally Posted by Roadhog_

Bro, you lowered your memory clock speed too.

I'll do another run in a bit then....Just for apples to apples comparison for just the driver update.

[Roadhog]

Originally Posted by shadow001

I'll do another run in a bit then....Just for apples to apples comparison for just the driver update.

50mhz is a huge change for memory.

[shadow001]

Originally Posted by Roadhog_

50mhz is a huge change for memory.

Well,once i do another run with the memory clocked back to 2150 like the first run and the overall score doesn't drop back to 16.2k like it did, but stays the same at 16.8k or even goes higher, we'll know it isn't a hardware issue at all....

[Roadhog]

well whats the holdup slacker.

[shadow001]

Originally Posted by Roadhog_

well whats the holdup slacker.

Ran the benchmark 3 passes back to back to make sure i get the same result, and there's no change between running the memory at 2100Mhz or 2150Mhz at the performance setting:


http://3dmark.com/3dm11/1224172;jses...8mW FYb6r64wM


16793....Dropped 7 points from the previous run.

[Roadhog]

Originally Posted by shadow001

Ran the benchmark 3 passes back to back to make sure i get the same result, and there's no change between running the memory at 2100Mhz or 2150Mhz at the performance setting:


http://3dmark.com/3dm11/1224172;jses...8mW FYb6r64wM


16793....Dropped 7 points from the previous run.

Yeah, thats the only way you can test the stability of DDR5 is to overclock and then see if performance drops. Though the gt580 has ddr5 rated at 5000mhz, who knows if its supplied enough voltage to run at that speed.