I'm sure Johan's review will confirm what Cray and others have learned already that the Opteron 6200 and 4200 series are far better CPUs than most people expected, and that's a good thing.
Yeah, I think the only reason AMD launched bulldozer for was for holiday sales. Traditionally, AMD has been designing high end CPUs for the server market first, desktop second. For good reason, as this approach has gained them traction in a more lucrative market. These CPUs will likely shine in this regard.
And that's where the big money is. One company can buy more CPUs in one order than consumers buy as a whole in months. While everyone talks about the high-end consumer CPUs, they represent a fairly small volume. The big demand is in budget CPUs and server CPUs.
well, if servers are where the big money is, AMD had better hope that Bulldozer is much more competitive than it was on the desktop and helps them gain some market share. What I dont understand is how it can be such a great server chip when the power consumption is so high. Guess the review will clarify the situation. Will be anxious to see it.
The last figures I saw if I recall correctly showed AMD with I believe only about 5% of the server market.
You can't understand it because it's probably not true. Sure there are cases where AMD's design is better such as huge clusters like Cray, but for the majority of 2p and 4p servers, AMD is behind in performance and efficiency.
I think the problem is that they don't do very well at high frequencies. These server chips have a more "sweetspot" frequency where they are much more power efficient. And here they can trade frequency for more cores without losing performance, because servers can use them. Only problem is they are expensive to make, but that's AMD's loss. The chips themselves are just fine.
Interesting. I was about to say I want that CPU in a silent(-ish) 1U chassis like the Asus RS100 but with hotswap drives. Alternatively I'd be down with the MicroServer, too.
Hell of a CPU for a 2S system if you ask me. 1S would be fine for me, too.
The 4200 is a dual memory CH system while the 6200 is 4ch memory. So the 6200 could stand out a lot more, performance wise, on some server task. I think that could be a hugh "main difference between 4200 and 6200".
Also if the 4ch memory system is working well could AMD put that in the desktop system for Trinty to help the onboard video.
... is not in this first article. As always, it took some BIOS updating, some troubles because we updated the hypervisors and other software... I'll spare you, but Murphy's law hits us harder than ever. So we had to play somewhat safe, this review will focus on virtualization, rendering and some misc. benchmarks.
We do have a FPGA benchmark that we fully understand now, and more HPC benchmarking is under review. Expect that later, but still this year. :-)
I would like to see FOSS server benchmarks. Say Apache -> PHP/Perl/Java -> MemcacheD -> MySQL/PostgreSQL on Centos 6 using default distribution packages. Then virtualization under KVM. My own experience is much of the Intel advantage disappears when running FOSS but I only have my own machines to test against. There are many smaller companies that use FOSS aggressively because spending $100K on annual server licenses means the owner's salary is reduced by $100K.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
22 Comments
Back to Article
Beenthere - Monday, November 14, 2011 - link
I'm sure Johan's review will confirm what Cray and others have learned already that the Opteron 6200 and 4200 series are far better CPUs than most people expected, and that's a good thing.shiznit - Monday, November 14, 2011 - link
Source?Kristian Vättö - Monday, November 14, 2011 - link
http://blogs.amd.com/work/2011/11/13/with-the-laun...Should be related, at least.
Beenthere - Monday, November 14, 2011 - link
Over double the prior server performance in the same socket...http://www.crn.com/news/data-center/231902878/late...
MonkeyPaw - Monday, November 14, 2011 - link
Yeah, I think the only reason AMD launched bulldozer for was for holiday sales. Traditionally, AMD has been designing high end CPUs for the server market first, desktop second. For good reason, as this approach has gained them traction in a more lucrative market. These CPUs will likely shine in this regard.Kristian Vättö - Monday, November 14, 2011 - link
And that's where the big money is. One company can buy more CPUs in one order than consumers buy as a whole in months. While everyone talks about the high-end consumer CPUs, they represent a fairly small volume. The big demand is in budget CPUs and server CPUs.frozentundra123456 - Monday, November 14, 2011 - link
well, if servers are where the big money is, AMD had better hope that Bulldozer is much more competitive than it was on the desktop and helps them gain some market share. What I dont understand is how it can be such a great server chip when the power consumption is so high. Guess the review will clarify the situation. Will be anxious to see it.The last figures I saw if I recall correctly showed AMD with I believe only about 5% of the server market.
shiznit - Monday, November 14, 2011 - link
You can't understand it because it's probably not true. Sure there are cases where AMD's design is better such as huge clusters like Cray, but for the majority of 2p and 4p servers, AMD is behind in performance and efficiency.Beenthere - Tuesday, November 15, 2011 - link
You continue to talk foolishness.Over double the system performance from the same socket means lower TCO in addition to the substantial lower power consumption from 35w-115w.
http://www.crn.com/news/data-center/231902878/late...
http://www.dailytech.com/article.aspx?newsid=23267
http://www.theregister.co.uk/2011/11/14/amd_optero...
HATING does not change reality !
ajp_anton - Monday, November 14, 2011 - link
I think the problem is that they don't do very well at high frequencies. These server chips have a more "sweetspot" frequency where they are much more power efficient.And here they can trade frequency for more cores without losing performance, because servers can use them. Only problem is they are expensive to make, but that's AMD's loss. The chips themselves are just fine.
saneblane - Monday, November 14, 2011 - link
You do the same thing on every forum, talk whole day without proof. Where is the proof, or link or anything. Either put up or shut up.bigboxes - Monday, November 14, 2011 - link
He's too busy file sharing to care. ;)zdw - Monday, November 14, 2011 - link
I'd love to see a box like the HP MicroServer with a 4256 EE in it.Daniel Egger - Tuesday, November 15, 2011 - link
Interesting. I was about to say I want that CPU in a silent(-ish) 1U chassis like the Asus RS100 but with hotswap drives. Alternatively I'd be down with the MicroServer, too.Hell of a CPU for a 2S system if you ask me. 1S would be fine for me, too.
erson - Monday, November 14, 2011 - link
Kristian, perhaps you could add those two columns for a better overview.Kristian Vättö - Monday, November 14, 2011 - link
Added! AMD's CPU database didn't have this info, hence I originally left it out.gevorg - Monday, November 14, 2011 - link
The 16-core Opteron for half a grand is quite amazing.Marlin1975 - Monday, November 14, 2011 - link
The 4200 is a dual memory CH system while the 6200 is 4ch memory. So the 6200 could stand out a lot more, performance wise, on some server task. I think that could be a hugh "main difference between 4200 and 6200".Also if the 4ch memory system is working well could AMD put that in the desktop system for Trinty to help the onboard video.
Kristian Vättö - Monday, November 14, 2011 - link
It's pointed out in the article I linked ;-) I added it anyway.alpha754293 - Tuesday, November 15, 2011 - link
Johan,Please run more HPC benchmarks.
Thanks.
JohanAnandtech - Tuesday, November 15, 2011 - link
... is not in this first article. As always, it took some BIOS updating, some troubles because we updated the hypervisors and other software... I'll spare you, but Murphy's law hits us harder than ever. So we had to play somewhat safe, this review will focus on virtualization, rendering and some misc. benchmarks.We do have a FPGA benchmark that we fully understand now, and more HPC benchmarking is under review. Expect that later, but still this year. :-)
MossySF - Tuesday, November 15, 2011 - link
I would like to see FOSS server benchmarks. Say Apache -> PHP/Perl/Java -> MemcacheD -> MySQL/PostgreSQL on Centos 6 using default distribution packages. Then virtualization under KVM. My own experience is much of the Intel advantage disappears when running FOSS but I only have my own machines to test against. There are many smaller companies that use FOSS aggressively because spending $100K on annual server licenses means the owner's salary is reduced by $100K.