Silicon Investor (SI) -- The First Internet Community

STOCKTALK

We've detected that you're using an ad content blocking browser plug-in or feature. Ads provide a critical source of revenue to the continued operation of Silicon Investor. We ask that you disable ad blocking while on Silicon Investor in the best interests of our community. If you are not using an ad blocker but are still receiving this message, make sure your browser's tracking protection is set to the 'standard' level.

Technology Stocks : Advanced Micro Devices - Moderated (AMD) -- Ignore unavailable to you. Want to Upgrade?

To: fastpathguru who wrote (255211)	8/5/2008 2:10:50 AM
From: graphicsguru	Read Replies (1) \| Respond to of 275872

with the same core clock speed but 2.5x (800 vs. 320) cores . . . . it achieves 1.5-1.8x performance. The difficulties of scaling from 320 to 800 cores are vastly different from the difficulties of scaling from 8 to 32. Eventually, you start to run out of things to do in parallel, and load balancing gets tricky. But if you can still get a decent improvement going from 320 to 800, that suggests that near linear speed-up from 8 to 32 should be a piece of cake.

To: fastpathguru who wrote (255211)	8/5/2008 2:26:41 AM
From: wbmw	Read Replies (1) \| Respond to of 275872

Re: There are only a few possible explanations for the graph: A) The chart is not representing anything meaningful performance-wise, B) The chart is bogus, C) The non-core portions of the chip are over-engineered by an order of magnitude, D) Intel has come up with magic crystals that allow them to conquer Amdahls Law. E) the chart is measuring rasterization performance - one of the most embarassingly parallel workloads out there. Amdahl's Law does apply to the serial portion of the graphics engine, but not to the raw processing of millions of independent triangles into pixels.

To: fastpathguru who wrote (255211)	8/5/2008 7:00:11 AM
From: Dan3	Respond to of 275872

Re: that slide If the 4870 scores 107 on the benchmark that Larrabee scores 5.5 on, that scaling could be just what's expected... That's about the overall performance ratio we usually see with Intel graphics compared to NVidia or AMD high end, so it seems to be a reasonable assumption. Perhaps we'll hear some alternate speculation regarding the actual absolute performance of larrabee relative to its real world competiton.... :-) ATI and NVidia have some low core count graphics parts, how is their scaling from 8 to 32 cores when memory bus, etc are under almost no load? The flat scaling may be an indication of very low absolute performance. Larrabee's competitor may be Geode, rather than 4870. Think it's up to the challenge? :-)