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: graphicsguru who wrote (238330)	8/7/2007 11:16:18 PM
From: Sarmad Y. Hermiz	Read Replies (1) \| Respond to of 275872

you're trying to reason with a foul-mouthed shill.

To: graphicsguru who wrote (238330)	8/7/2007 11:48:48 PM
From: pgerassi	Respond to of 275872

Graphicsguru: At how many FBDIMMs per channel? Most commercial applications get the most performance at max memory which is 2 FBDIMMs per channel. Thats 30W additional power usage per Clovertown over RDDR2 for Opteron. You are also assuming continuous usage of the server. At a lot of idle, the 30W FBDIMM disadvantage smokes any Clovertown power reduction. And most commercial servers have 33-67% idle per week. Third which Opteron stepping are you comparing to G0 stepping of Clovertown? Early Toledo based ones or the later Windsor based ones? If you have to compare 4 socket servers, you must use the P4 based Xeon MPs versus Windsor based Opterons as there is no C2D based Xeon MPs. There the P4 Xeon MPs use more energy in both idle and under load plus are lower in performance to boot. If you compare using Penryn based Xeon MPs, then you must be against Barcelona based Opterons which use less power and perform better. You would still need to use the 2 or even 4 FBDIMM per channel power disadvantage for Xeon MPs. !6 RDDR2 DIMMs would allow a lot more memory than 4 FBDIMMs. Even 8 FBDIMMs would have a large performance penalty for commercial server tasks. This 16 FBDIMMs force a 25W per quad core Penryn disadvantage at load. That will push the overall power consumption of 4P Penryn QCs servers above 4P Barcelona QC servers at same clock. The performance will likely be in Barcelona's favor as well. 8 channels of RDDR2-800MHz will be more than 4 channels of FB-DDR2-667MHz in both lower latency and higher bandwidth. Lastly, Sun's 4600 shows that some of your file serving towers are also Opteron powered. So are many file serving boxes from other vendors. They have more DRAM performance and I/O performance per socket. Come Q4, this advantage will skyrocket in K10's favor since they will also have far greater I/O performance. Pete

To: graphicsguru who wrote (238330)	8/8/2007 4:54:47 AM
From: pirasa2	Respond to of 275872

2006 x86 server sales was about 7.5 million units. That makes, at roughly 2 CPUs per box, 15 million CPUs. One should add to that drop-in CPU upgrades/replacements, which should bring the number up by 1-2 million. I am inclined to believe that the overall effect of Celerons being counted as server CPUs in the final count that you alluded to must be relatively insignificant. In Q4 2007, the server market is expected to grow healtily yoy. It is a seasonally good quarter anyway. So a 4.5-5 million in CPU units is a conservative estimate for Q4 2007 total market size. If Barc captures just 8% of that, thanks basically to pent up demand and drop-in upgradibility, that makes 360K-400K going into servers. Add to that the number of units going into HPC machines. And finally, add the Phenom sales.