Latency Settings and their Impact on Memory Performance

[Trellabor]

Corsair just published a detailed Application Note on memory performance.

"Memory modules are currently available which support a wide variety of different speeds and latency settings. Speed is easy to understand - in general, faster is better. But what do the latency settings mean? And, what impact do they have on memory performance? This paper will provide a brief background on latency settings and what they mean. Then, we will move to the lab, where we will run a suite of benchmarks over a wide variety of latency settings, and we will measure the impact of these settings on benchmark scores. "



The pdf is posted on Corsair's Tech Notes page and can be seen here;


Corsair Memory

[Mindwarp]

Wow, that really explains alot. Great find!

Now we can tweak with a purpose!

[WiCKeD]

Cool! I like the global comparison. Good work from Corsair.

Interesting though, because I remember cuddles thread on XS showing the latencies independently were having no impact on Doom3. Maybe each one individually was not enough to alter the results? Take a look: http://www.xtremesystems.org/forums/...ad.php?t=48634

This is an excellent chart, although in my own tests on an XP system, tRCD did not have 1/4 of that impact - about the same as tRP... at least on 3DMark01. Their overall 3DMark01 score looks reasonable. Someone double-check these results.

[Dil-tech]

They're playing at higher resolutions Wicked. Those benches were done at 640x480, where it's ALL cpu limitation. At 1280x1024, or 1600x1200(and usually with AA) it won't be noticed due to the fact that you're coasting on gpu power and not cpu.

[WiCKeD]

Originally Posted by Dil-tech

They're playing at higher resolutions Wicked. Those benches were done at 640x480, where it's ALL cpu limitation. At 1280x1024, or 1600x1200(and usually with AA) it won't be noticed due to the fact that you're coasting on gpu power and not cpu.

1024*768 would probably have used less GPU power on a four year old benchmark than 640*480 on DOOM3 on any card... Not to mention the fact that 99% of people don't play in 640*480, so it would not have been a realistic test of how memory affects gaming. You know a test using AA might have showed an even greater impact, because the graphics card probably relies more on system memory, when it runs out of texture space on the card. Still that is a good point, because I didn't think about that when I did the test.

Anyway, none of that explains the difference in tRCD.

V Thanks, but I just pulled it from the PDF in the first post It is very handy, though.

[astrocrep]

Wicked that table is a pretty nice table...

Good find!

Rich