It's up.
We only had one board and one set of processors from AMD for this article, so of course they first went up to Jason in Canada for the server tests. He had them for about a week then sent them down to me for the desktop tests. I struggled with some board issues and eventually got the hardware down to Derek for his workstation look. Unfortunately he had some issues getting his Intel workstation platform up and running so the workstation performance numbers will have to wait for another article.
At the end of the day I was very impressed with the performance of AMD's Athlon 64 X2 as well as the Opteron dual core solutions, my main issue is price. Intel is really playing the role of driving the market here, since developers won't multithread their applications unless a large installed user base of multi-core/multiprocessor systems exist and Intel's pricing is the only thing that will drive that sort of mass adoption of dual core. So while we always criticize Intel and haven't been happy with their architecture in years, the industry needs Intel to make sure that by the time '06 and '07 roll around - we'll be playing multithreaded games in a more multi-core friendly OS with tons of multithreaded applications running in the background.
I do sincerely hope that the X2 4200+ and 4400+ will be available in reasonable quantities by the end of the year because they are the only hope for getting any sort of somewhat reasonably priced dual core CPUs out of AMD.
We only had one board and one set of processors from AMD for this article, so of course they first went up to Jason in Canada for the server tests. He had them for about a week then sent them down to me for the desktop tests. I struggled with some board issues and eventually got the hardware down to Derek for his workstation look. Unfortunately he had some issues getting his Intel workstation platform up and running so the workstation performance numbers will have to wait for another article.
At the end of the day I was very impressed with the performance of AMD's Athlon 64 X2 as well as the Opteron dual core solutions, my main issue is price. Intel is really playing the role of driving the market here, since developers won't multithread their applications unless a large installed user base of multi-core/multiprocessor systems exist and Intel's pricing is the only thing that will drive that sort of mass adoption of dual core. So while we always criticize Intel and haven't been happy with their architecture in years, the industry needs Intel to make sure that by the time '06 and '07 roll around - we'll be playing multithreaded games in a more multi-core friendly OS with tons of multithreaded applications running in the background.
I do sincerely hope that the X2 4200+ and 4400+ will be available in reasonable quantities by the end of the year because they are the only hope for getting any sort of somewhat reasonably priced dual core CPUs out of AMD.
Facebook
Twitter
RSS
16 Comments
View All Comments
Mojo_HK - Monday, May 9, 2005 - link
Your comments on driving volume w/ price to stimulate the development eco-system is interesting. Noting the cheapest Intel dual core is ~25O $, it would be interesting to see a performance comparision b/w AMD's equivalent priced single core product as well as Intel's product from it's P4P-HT line up. Point being - end user's try and buy the best value in their budget. What options exist at the 250$ price point - is dual core really the best bang for buck?Mojo
Howard - Monday, April 25, 2005 - link
#13,#14 never mind. There's no measurable difference using -j2 on WinXP, dunno what I was looking at before.Howard Chu - Monday, April 25, 2005 - link
#13, likely not, good point. There's so much overhead in running Cygwin already. MinGW/Msys would have better throughput. I use -j2 when building OpenLDAP on WinXP on my Centrino laptop, using Msys. It takes several times longer to do the corresponding build with Cygwin, with or without -j2.Red and black - Monday, April 25, 2005 - link
#12, note that they're using cygwin gmake on MS Windows. Is the 2xCPU setting of -j still a good idea given the horrible behavior of the MS Windows scheduler?Howard Chu - Monday, April 25, 2005 - link
Generally you can get away with "-j" set to twice the number of CPUs. Compiling is a fairly I/O intensive task - reading source files, writing object files, temp files, etc.; while one job is in I/O wait the processor can compile something else that's already been read in.By the way, I designed the current "make -j" feature in GNU make, and wrote the first implementation of it, so I'm pretty familiar with its advantages and limitations, in case you're wondering.
Matt Smith - Sunday, April 24, 2005 - link
#10SSE3 might have made a small difference, but the main reason encoding performance was way up was the dual cores. Intel's HT has helped them a lot in this area, and now that both have 2 cores things are pretty even.
Red and black - Friday, April 22, 2005 - link
Excellent article. I look forward to the mentioned look at linux multitasking. Thanks for including a compilation benchmark! Also, it would be neat to see how the compile performance changes with the "-j" parameter to "make" -- I think it should be set to no less than the number of cores.Was the better AMD media encoding performance because of new SSE3 stuff?
Matt Smith - Thursday, April 21, 2005 - link
#7, The entire chip is throttled down at the same time. :(Blake - Thursday, April 21, 2005 - link
Anand, what bios were you running on the Tyan K8WE?I've been looking into that board and it's really nice but I agree with you; the bios releases up until now have been horrible.
Tyan is supposedly releasing a new version of the board without the 6 pin power connector and maybe some other changes too.
Have you heard anything about that?
Dan - Thursday, April 21, 2005 - link
How does Cool & Quiet work on a dual core athlon64? Do both cores throttle down at the same time or can you have one core running at full speed and the other at half speed?10x