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: Ali Chen who wrote (227269)	3/2/2007 7:06:16 PM
From: pgerassi	Respond to of 275872

Dear Ali: Purchases for productivity workstations are driven by performance of SINGLE APPLICATIONS, not multiple copies of the same application. Flat out lie! Even single applications like a CAD/CAM package use multiple threads running many different programs simultaneously. I have built such applications. Evidently you have not! I even shown you a single instance of an Oracle RDBMS which had 6 threads open running 6 different programs all cooperating to allow users to work on just one data set. That doesn't even include any client code or OS threads to help the subtasks along. Its a fantasy to think that even a single application runs just one thread without any help. There are processes that do such mundane things like scan network traffic looking for its IP address, regulating virtual memory, read and writing to disk, and looking at KB and mouse traffic. But no, Ali thinks it just runs one thread doing one task all of the time. And even if you were right, 100 low end SC C2D's (you did say that workstations do only a single thread at a time) have less profit than four top end QC CPUs. Lets see, 4 times $3K is $12K revenue (DC parts go for $1500) and over $11.8K in profit. 100 SC Core 2 Solos are about $100 each for $10K revenue and $7K profits. Opteron QCs win given your assumptions being true. But rarely does a corporation hand out 100 workstations. Perhaps a handful while the rest are just good enough PCs or previous hand me down PCs. You don't need a workstation to do office type work. A $300 or $400 Celeron or Sempron is good enough with chips that fetch $40-60. That is the bulk of the desktop market. It is different from boldly running multiple applications of the same kind who do not interact other than through cache thrashing and memory bandwidth contention. Not if most power users run multiple things at the same time which cause cache thrashing and memory bandwidth contention. That is typical for server code and power user usage. The working sets don't fit into the caches which is why memory latency is such a key to real world performance. Intel's band-aids of bigger caches, more aggressive prefetching try to mitigate this as the real world saps the theoretical performance. If it was so atypical that cache thrashing and memory BW contention happen, then Intel could get away with a tiny L1 and no L2 cache at all. The last time they tried that, they got their head handed to them. Heck servers used to thrash main memory and swap a lot. Many businesses were sold faster servers that ran so good because they had enough memory to keep from needing to swap. Many times these businesses were surprised that their slow server doubled in speed when a simple thing like third party memory was added. And if adding a few GB of memory stops swapping, then adding a little cache won't help much. The working sets are far larger than the cache. In such cases only lower memory latency has a good effect on performance with memory BW being a close second (effectively lowers a portion of latency). Most agree that in those cases, multisocket Opteron has a big advantage there. So you lose either way. Your assumptions are WRONG or your case shows Opterons get more money and profits. Pete