CPU vs GPU, the final chapter
EARLIER THIS WEEK we saw that, even within a single benchmark vendor, Futuremark, the two subsequent test generations, 3Dmark 06 and 3Dmark Vantage, scale quite differently depending on the CPU and GPU configuration changes.
What about a generic PC performance benchmark that takes into account all of the PC componentry - how would it scale then? Will the - more Intel preferred - test mix change the CPU vs GPU influence dramatically?
To stay within the family, I ran the new PCmark Vantage, 64-bit version of course, on this Vista 64 configuration. 4 GB of OCZ heat-piped RAM made sure that memory swapping doesn't become the spoiling factor.
I limited the run to five particular configurations, combining dual core or quad core 3.2GHz CPU with single or dual (two or four GPUs total) GeForce 9800GX2 cards.
As you can see, the differences are there - but this time, it is the CPU change that leads the way to higher scores. The GPU improvements, however, aren't totally unnoticeable - if you want to call some obvious score reductions that way. Namely, in both cases, when you add the second graphics card, the memory score would go down, more so in the dual-core case.
In summary, the previous "chef's recipe" stays: go ahead with a balanced combo. Good dual-core CPU with a single GPU, midrange quad-core with dual GPU, and top quad-core or more with either top-end dual GPU (read: dual GTX280) or quad-GPU setup. ยต
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
The numbers there make it look like there is little point in an extra 2 cores and a second GX2, the gains are pitiful in fact. Quad still sucks for gaming, ah well.
Could you guys post a conclusion article somewhere where you dumb it down so laypeople like me can understand it?
And there was much rejoicing.
What are results with a 4/2 core CPU running without a Nvidia graphics card and with the best intel graphics chipset?
Game1/Game tests: why does 2core1gx score higher than 4core1gx? (I think all cores run @3.2G)
Futuremark has been dissed as junk benchmarks years ago.
All rubbish what they produce.
And it has been proven many times, so sad The Inq uses them again.
You are putting your reputation in the shredder writing such crap.
Well actually, it's the benchmark that is pitiful.
I plunked down for a quad-core when I found out that Supreme Commander on my dual-core, single-GPU system with 2GBS of DDR2 RAM was gasping for air.
So now I have a quad-core Q6600, a single 8800GTS and 4GB of DDR2 RAM, and Supreme Commander runs a lot better !
Oh, and I run XP SP2.

Flame on.
The scores demonstrate that the test code is poorly threaded and/or that windows vista has a poor task scheduler. Poor task schedulers have always been a feature of Microsoft OS's, so no amount of great threading code in the test can overcome the fact that the OS is useless at this. The more cores the worse it gets and the return on investment drops.

You will need a good Unix platform like Solaris to really get the advantage of SMP configurations. Linux is better than windows at this, but its by no means the best either, and being better than windows in this arena is not exactly a big bragging right.

Thats why some of the scores defy common sense.