Generic CPU Benchmark

[Particle]

Hey guys. I wrote a CPU benchmark that tests a battery of common math ops on a large dataset. It should be a good mix of both CPU and RAM stress. While this didn't start as a serious project, I figured some of you might like a new bench to play with. It does actually measure performance in a realistic manner despite the humor in its interface, so it isn't a bad benchmark to try out.

The throughput is measured in KCUs per second. Higher numbers are better. This benchmark is multithreaded, and it will probably make your machine sluggish while it is running. It shouldn't take too long to complete--15 seconds is pretty common on modern hardware.

Build 105: Locked to x86 since it runs faster on AMD64 systems. Also reports individual thread finish stats.

[speedfreak86]

Vista64:

[Titan7171]

[roofsniper]

phenom 9600
corsair dominator 1066 @ 5-5-5-24-30
m3a32-mvp
nothing is overclocked

[Daveburt714]

Well.... I just had to be the 1st one to break 31,000 KCU's

CLICKABLE:

[Particle]

Good work, dave. I think that's the record so far.

[dengyong]

Cpu @ 4.3ghz

[SoulsCollective]

31.1k
Test seems to love low latency - about 3k boost going from CAS-7 to CAS-6 at same speed.

[speedfreak86]

31.1k
Test seems to love low latency - about 3k boost going from CAS-7 to CAS-6 at same speed.

I tried a lot of things just now in windows upping and lowering speed of the processor made less than 1k of difference. I agree with the above that memory latencies will make a big difference though.

[merkk1]

Phenom 9850BE @ 3.107 ghz

[dengyong]

Throughput score should be measured from a single core if it's only multiplied by the number of cores.

[m411b]

This is my bench, its on a C2D lappy - T7600@ stock 2.33Ghz - 10,829 KCU's. LOL


Something I noticed is that the test doesnt seem to be touching the CPU. It didnt make it jump 1% even. I think it is more memory oriented.

[Daveburt714]

31.1k
Test seems to love low latency - about 3k boost going from CAS-7 to CAS-6 at same speed.

It does seem to be pretty memory sensitive, but thats not a bad thing...
In fact, it probably makes it a better test for real world performance.

I started by tweaking timings first which showed slight improvements, but to actually break the 31k barrier I had to up the Mem Frequency... (I seriously doubt it was 24/7 stable, but it was stable enough... )

Geez SoulsCollective, I was hoping to hold the record a little longer than 1/2 an hour....

[dengyong]

Something I noticed is that the test doesnt seem to be touching the CPU. It didnt make it jump 1% even. I think it is more memory oriented.

Cpu is at 100% for the time it runs. Each block=5 seconds.

[roofsniper]

i usually use my logitech g15 to check my cpu load and for every benchmark i have ever ran it works and shows the load. for this one it pauses and w/e load it was on it stays at that. like i have ran the test with it saying 0% load just because this program eats up so much bandwidth.

[sirheck]

[m411b]

Cpu is at 100% for the time it runs. Each block=5 seconds.

i usually use my logitech g15 to check my cpu load and for every benchmark i have ever ran it works and shows the load. for this one it pauses and w/e load it was on it stays at that. like i have ran the test with it saying 0% load just because this program eats up so much bandwidth.

Yeah I see that now, didnt think of opening task manager.
Was using Everest OSD..

[SoulsCollective]

Nice result. Mind posting up your system settings? Sig is a little short on detail.

[sirheck]

Nice result. Mind posting up your system settings? Sig is a little short on detail.

I dont really understand the benchmark.
But here are my settings from CPU-Z and Realtemp.

[dengyong]

If you hold the Alt key when you hit PrtScn (print screen) you'll only capture the active window and won't have to crop your image.

[sirheck]

If you hold the Alt key when you hit PrtScn (print screen) you'll only capture the active window and won't have to crop your image.

Yeah im lucky just to be able to post a pic
I always use (snag-it) for this type of shistuff.

[SoulsCollective]

I dont really understand the benchmark.
But here are my settings from CPU-Z and Realtemp.

*pic snip*

What PL setting and strap?

[sirheck]

What PL setting and strap?

IIRC everything in my rampage bios is stock.
The volts for the cpu are up a little.

But things like trans booster,gtl ref and all that are (auto)

I had to manually set certain things on my old maximus SE
to keep the Q66 folding stable at 3.6 but almost everything
in BIOS on this rampage is stock/auto.

Sorry im old and its been a while since i set it all.

Could post a pic of BIOS if you like.
Probably better to PM so not to jack this thread to much.

[dengyong]

Yeah im lucky just to be able to post a pic
I always use (snag-it) for this type of shistuff.

I use paint with the shortcut on quick launch. Saving as jpeg keeps the file size small enough to not need an image host.

[SoulsCollective]

IIRC everything in my rampage bios is stock.
The volts for the cpu are up a little.

But things like trans booster,gtl ref and all that are (auto)

I had to manually set certain things on my old maximus SE
to keep the Q66 folding stable at 3.6 but almost everything
in BIOS on this rampage is stock/auto.

Sorry im old and its been a while since i set it all.

Could post a pic of BIOS if you like.
Probably better to PM so not to jack this thread to much.

Hmm, ok. It's not thread-jacking, we're trying to work out what makes this bench tick, and thus why your score is better than anything else posted so far.