More CPU for more GPU - Benchmark scores
YESTERDAY we went through the motions to see what happens to the 3Dmark scores when you pair up a simple dual core CPU with one, two or four GPUs - yes the CPU was Intel and graphics was Nvidia: all this on Asus mobo and card platforms.
Now, I replaced the dual core entry with an identically clocked quad-core one, the QX9650, having double the cache as well, while keeping everything else identical. How were the graphics, CPU and overall scores impacted? Was the PhysX card still of any use in that second CPU benchmark?
This time, I limited the benchmark scope somewhat, minus the extreme Vantage round - 9800GX2 got more of a showing here, as it was far more stable even in quad-GPU setups, and it kinda made more sense to combine quad-core CPU with quad GPU, it matches doesn't it?
After the usual round of hang-ups and freezes - yes Striker II Extreme has its fair share of these despite our update to the freshest BIOS 0703, and some of our readers seem to have the same experiences - here are the results:
And, here's how the PhysX stands against doubled raw CPU power:
As you can see, the CPU still has an impact on the scores - keep in mind there was NO CPU overclocking here, for all practical purposes: 3.2 GHz is pretty standard stuff for any quad-core Yorkfield Penryn.
Take a look at the CPU 2 test in 3Dmark Vantage: it seems that quad core 3.2 GHz is a good match for the Ageia PhysX, speed wise. But when you put them together, it's a record result, nearly 30 interactions per second. In both dual and quad-core CPU setups, Ageia PhysX brings along extra speed equal to additional two 3.2 GHz cores.
For a comparison, in 3Dmark06, there was very little jump when moving from two to four GPUs (exception is the HD resolution), but more benefit when going from dual-core to quad-core CPUs.
Here's the single 9800GX2 on dual core in WUXGA
Then, dual 9800GX2 in Quad SLI on that same dual core, again WUXGA
Finally, dual 9800GX2 Quad SLI, Quad core CPU, WUXGA:
Funny how rounded some of the aggregate scores are?
Overall, within the very limited 3Dmark scope, yet counting the real app experience, my point is: if going for dual or quad GPU and such gadgetry, I'd say it does make sense to match it with a full quad core CPU. After all, both Intel and AMD quaddies are now in the $300 class for a pretty well performing - and in the Intel case, beautifully overclockable - products.
At least you know that you'll be able to run pretty much any application well, whatever you throw at your PC. As for PhysX, well maybe it was about time to move it to GPU or CPU or - both. In the meantime, we'll share with you more on the PCmark Vantage comparison, real soon now... ยต
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
This is the most pointless set of articles ever, use 3dmark which separates GPU and CPU tests then combines them for a final score and then we are suppose to draw some conclusion from it? Daft I call it (except for the part where asus gives you a free system for this joke).
If you want to see the effects of number of cores vs number of GPU's on games you have to test with games that use the GPU and CPU massively at the same time, not some benchmark that runs separate tests for each function, duh.

So going by 3DMark06's 3Dmark score you only get a 15.65 percent increase when going from a single 9800GX2 with dual core CPU to a dual 9800GX2 Quad SLI, Quad core CPU. That is pretty lame for the extra cost and power used. I would of expected at least a 35% increase.
What junk is this?
3DMark has been dissed for years as a junk benchmark.
Even by The Inq themselves, and now it proves something?

This is too much, even for The Inq.....
Get Mike back in here, he would never have approved such rubbish.
Wow, Pay more than computer costs & run worse. Amazing, Must be transplantee.

Maybe problem is fundumentally CONTROLLERS just arn't going to allow that much higher of scores. Maybe its limit HDD cannt perform above. Yet something tells me any serious Gamer should consider Another Product entirely for Money. What if 2,000 stream processors also continue slide back. When Vantage first appeared, all sorts of ideas on why & what great numbers would come to, Came to NOT Much, is fact. So BE Dissappointed for price.

Arrrggggg, Me Baby Whites Seem Sickly.Kriel artists abound in ye waters.
Drashek
How about doing the same test with AMD branded stuff with crossfire. See if the scores fluxuate with the different CPU's then.

Intel is not the only player, and yes, I am an AMD fanboy.
What the results seem to indicate is that a one to one matchup between CPUs and GPUs gives better performance, which seems intuitively correct. Which tells me that a chip integrating a CPU, GPU, and possibly a physics engine, or a GPGPU, should give fairly linear scaling as you double or quadruple the number of these integrated chips.

So the emphasis should be on the CPU/GPU integration. It will be interesting to see what AMD develops under the Fusion project.

But the key factor is still going to be the ability to structure the application to take advantage of the hardware architecture. Hopefully, more will be done to automate this process.
With all due respect Nova, I think your reasoning might be flawed.
I think that solely on the basis that you draw your conclusions from the synthetic scores the calculations of which were almost comlpetely arbitrarily decided by Futuremark.

As far as I know, the benchmark could automatically assign a bonus for more cores just by counting them, and not by actually measuring the increase in performance.

I think using a couple of different benchmarks would provide results that are a bit more "scientifically plausible" ...