• entries
    422
  • comments
    1540
  • views
    488987

Benchmarks, like statistics, can be (ab)used to pr

Sign in to follow this  

147 views

You'll have seen in my previous journal entry that my new computer glows blue. To quote ApochPiQ:
Quote:
Everyone knows that more blue lights == more fastar.


Out of curiosity I ran some benchmarks on my two machines to see which came out ahead. I was assuming that a "3800+" would have more grunt than a "3.06ghz" chip (even though the AMD runs at a lower clockspeed). I was also assuming the AMD chip would run circles around my Intel one when it came to memory bandwidth - if only for the fact one runs PC3200 and the other runs PC2700.

For a Dhrystone benchmark the AMD jumps ahead by 20.32% which is consistent with a 23% difference between 3066 and 3800 ratings.

However, for a floating point Whetstone benchmark the AMD chip dropped nearly 3% - other SSE2 based FP benchmarks show between that and 30% drops for the AMD. Based on my bencmarks it seems the AMD Athlon sucks at floating point...

RAM Bandwidth tests were slightly more favourable towards the AMD. The PC3200 in my new machine pushes over 4.5gb/sec transfer rates - an impressive jump of 92% over my older machine [oh]


Ultimately these results probably mean nothing. My understanding is that SSE/SSE2 has always been Intel oriented, so maybe these tests were skewed in Intel's favour to start with. I've read about AMD having a superior memory interface on their chips (no idea if that includes mine) but equally I've no idea if I've benchmarked the RAM, the CPU or both...




In other news, I'm working on some code that'll convert a D3DX9 mesh to a D3DX10 one. Seems that the D3DX10 library is still no where near as complete as the more mature D3DX9 library, so if I can hook the two together then that makes things easier [grin]
Sign in to follow this  


6 Comments


Recommended Comments

“Bad SSE performance” is a typical problem with AMD chips. The reason for this is that many programs don’t use any SSE on AMD CPUs at all. This happens because they use a SSE detection function from Intel that first checked if the CPU manufacture is Intel and then looks for the feature bits.

The advanced of the memory interface comes from the integration in the chip instead of the Northbridge. This reduces the latency in random access situations. Additional it improves the overall memory bandwidth in multi CPU systems.

Share this comment


Link to comment
AMD's SSE is faster than Pentium 4 SSE, but not fast enough to compensate for the clock difference, unlike other parts of the AMD chip. Maybe K8L will fix this.

Share this comment


Link to comment
Interesting replies - thanks [smile]

Quote:
The reason for this is that many programs don’t use any SSE on AMD CPUs at all.
I did notice that the D3D10 RefRast only mentions its using 3DNow! extensions - my P4 machine mentions D3DX is using SSE2 extensions.

Cheers,
Jack

Share this comment


Link to comment
Quote:
Original post by jollyjeffers
I did notice that the D3D10 RefRast only mentions its using 3DNow! extensions - my P4 machine mentions D3DX is using SSE2 extensions.

It's a good question whether this 3DNow! is optimal these days. It made sense in the past to use 3DNow! when it's supported, but this may no longer be true, and may be a waste of code.

Share this comment


Link to comment
Actual FPU AMD owns Intel's FPU, but with SSE instructions AMD falls a bit behind. The benchmarks today usually test SSE/SSE2/SSE3. If you could run tests without them you would see AMD is quite a bit faster.

Share this comment


Link to comment

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now