Advanced Micro Devices - Moderated (AMD) Message Board

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)

AMD 238.58

-6.8%

11:13 AM EST

Public Reply Prvt Reply Mark as Last Read File

Previous 10 Next 10 Previous Next

To: graphicsguru who wrote (243221)	11/2/2007 8:45:58 PM
From: chipguy	Read Replies (1) of 275872

The issue for selling into HPC is pretty simple: FLOPs/$. IPF chips have excellent FP performance but they are pretty pricy compared to the x86 server chips that sell into HPC. IPF chips command a price premium because of their RAS and scalability features intended for mission critical systems. That means that IPF chips can't compete against x86 chips in raw FLOPs/$. For some specific classes of applications IPF systems today beat x86 in delivered application perf/$. That is why IPF has taken a relatively stable ~15% share of the HPC market over the past two years even while the onslaught of cheap x86 clusters has gutted RISC's share. The performance boost going from Madison to Montecito was biggest by far for important commercial workloads. This is hardly surprising given Madison's weaknesses vs its primary competitors and IPF sales revenue breakdown by workload, i.e. who is paying the bills. The move from Montecito/Montvale to Tukwila will likely give the biggest performance boost to HPC workloads from the significantly improved FLOP/bandwidth balance. IPF will probably fall further behind x86 in raw FLOPs/$ with Tukwila because it integrates a lot of high value mission critical system level functionality on die that's not needed in HPC but I think the subset of HPC application where its perf/$ is superior to x86 will be much more extensive than that of Montecito and might drive IPF to a 20%+ HPC market share. Where it gets really interesting is around 2009 or 2010 with Poulson when IPF will finally catch up with Intel's x86 chips in process technology (32 nm) and could be ahead of competing x86 and RISC offerings. It will probably still be more expensive than Intel's x86 chips in raw FLOPs/$ but it could easily be better in delivered application perf/$ over a huge portion of the HPC spectrum.

Report TOU Violation

Share This Post

Public Reply Prvt Reply Mark as Last Read File

Previous 10 Next 10 Previous Next