Following on from my first two blogs, there is one more question I am often asked by gamers, LAN house/cyber cafe owners and just friends and family…. Do I just need a good GPU and CPU? Does the rest of the PC hardware make a difference or is it just wasted money?

As I discussed in my first blog, in my opinion, balance between CPU and GPU is almost always important. For non 3D applications, the CPU is often more important than GPU, for 3D gaming, GPU is most important, but a better CPU helps out in real games, not so much in synthetic benchmarks. What other hardware components actually make a difference?

So, I assembled a good rig, using “good” components as a starting point. Without counting case, monitor, OS or peripherals, the total cost was about $900 without tax. Here is the specification and costs:

I then benchmarked this rig with three tests – results below.

Next, I substituted cheaper, less capable components one at a time. So, for example, I tested with just with cheaper GPU. Then I put the better GPU back in and tested with just the cheaper CPU; I continued this until I had tested all the weaker components one by one. Now we can see what effect trading any single component for a weaker component makes to the overall good, balanced PC.

One further note: I did not change the motherboard in these tests, but started with a good one; fully capable of supporting all the performance components. I believe that if we had chosen a cheaper, lower speed and quality motherboard, it too would have had an impact on the tests.

The cheaper components were:

• GPU: ATI Radeon™ HD 3850 – $100
• CPU: AMD Athlon™ 64 X2 5200+ – $65
• Memory: 2 GB Same PC8500, (1066Mhz) – $50
• Memory: 2 GB PC8000, (800MHz) - $40
• HD: SATA 320 Gig, 7200 (8mb buffer) – $70

OK, so we then performed 3 tests on each hardware configuration:

• Test 1: 3DMark Vantage Performance Test (probably the most accepted synthetic test suite; created by a consortium of OEM’s, Microsoft and many hardware companies such as Intel, AMD and Nvidia) – overall score
• Test 2: Crysis, fps SAME DEMO LOOP, (I took the minimum fps to see where the system was “waiting”).
• Test 3: Transcoding a 4 minute, MPEG2 DVD quality format video at 780 X 480 resolution to MP4 @ 480 X 208 final resolution (minutes – remember smaller is better)!

Results (click on the graphs to enlarge the image):

3DMark Vantage – 3D synthetic benchmark:

Crysis – 3D game:

Transcoding test:

What do we see in the three tests?

1. Starting with the 3DMark Vantage test, we see quite simply that nothing except the 3D card makes any significant difference in this test. Simple and clear. I honestly believe that the 1% or 2% ups and downs to the scores are just noise.
2. In the Crysis test, an intensive 3D game, we again see that the 3D card is by far the most important determinant of minimum fps during the time trial. However, here we see some other interesting points:
   • Both quantity of system RAM and speed of system RAM are both important. Faster RAM helps and more RAM helps too.
   • CPU is also important – though not the biggest issue. A mid priced quad core CPU helps over a similar speed dual core CPU – I believe that Crysis must be compiled to use more than 2 cores in a CPU. (In my opinion, as we move towards 2009, we will see many more games utilizing all four cores in quad core CPU’s)
   • In this game, as in 3DMark Vantage, we see that having a bigger buffer hard drive seems to make little or no difference at all.
3. Finally, the Video Transcode test. This test is a non 3D test. In this case transcoding was done using XviD4PSP software, version 5.0.34. Here we looked at the time to transcode a video – so, smaller is better. Wow! Everything counts here! Starting with best case set up it takes 10 mins to do Transcode the 4 min video. Swap ANY hardware out and you move to 15 mins or more!
   • Let’s start with the CPU, here is the most important piece of hardware; changing for a quad core AMD CPU to a dual core slows the test down by close to 150%!
   • Memory, again more and faster memory make significant differences to the speed of transcoding
   • Now we get to things that surprised me: A bigger buffer on the hard drive also makes a significant difference
   • Finally, we see that there is some GPU support, when we moved from an ATI Radeon HD 4800 series to an ATI Radeon HD 3800 series, we see that it takes almost 40% longer to complete the task.

Overall conclusions:

So, let’s keep this simple:

• For some synthetic benchmarks such as 3DMark Vantage, graphics tests and overall scores, only the GPU really counts. But who cares? 3D game players play games, not benchmarks.
• Crysis, one of the most intensive 3D workload games we have: Here, again, we see that the 3D card is the most important part of the mix; but that the CPU counts significantly, followed almost equally by quantity and speed of system RAM
• For a non 3D application that is compute intensive, we see that all the key components have a significant effect on the overall scores.

So, where does this leave us?

In my opinion, if you want a truly great 3D game machine only, spend your dollars on GPU first, CPU second, memory third, (don’t forget a good motherboard) and just buy as much hard drive space as you need.

If you want a truly great general purpose machine, that you do NOT intend to play 3D games on, then spend your money in this order: CPU first, memory second, GPU third (even surprised me how much of a differences this make), better or faster hard drive forth. (Again, don’t forget the good quality motherboard)

Finally, If you are like me, and you run 3D games and video transcoding and everything else, seems that you need to balance the PC purchase overall, put any low end part in the overall platform and you will see a big difference. Again, balance is king!

Notes:

• All prices were taken from Newegg the week of 1^st September. I took cheapest price where there were similar products. I did NOT factor in any mail-in rebates or other promotions.
• All tests were performed over the last month. All tests were performed twice and an average score was taken. My accomplice and I are not professional testers; we hope that we made no mistakes!

Paul Ayscough is Director of Advanced Marketing at AMD and an evangelist for graphics and gaming.  His postings are his own opinions and may not represent AMD’s positions, strategies or opinions. Links to third party sites are provided for convenience and unless explicitly stated, AMD is not responsible for the contents of such linked sites and no endorsement is implied.