Analysis on IT trends and competitive strategies, with emphasis on micro processors, computer systems and networks. Based on latest news, backed up with real data, this site intends to provide a true and realtime picture of the fast changing IT landscape. This journal strives to be accurate on facts and sharp on criticisms. You may email your opinion to sharikou@yahoo.com or post comments here, be cool and intelligent.
Freelance journalist on IT matters. Some of my writings have been published on online IT journals. Any original content on this journal is Copyrighted, but it's free for non-commercial use. Any Trademarks used on this site belong to their respective owners. Some of the pictures are links. If there is any issue with the content of this site, please email sharikou@yahoo.com .
I just added the motherboard information for AnandTech's test of Opteron against Woodcrest. Anand used the MSI K8N Master2-FAR for the Opteron.
posted by Sharikou, Ph. D. @ 6/21/2006 12:57:00 PM
10 Comments:
Maybe this has been stated in the other thread, but it is worth reiterating.
I don't know if this particular motherboard is similar to past MSI motherboards. In the past, MSI would connect only one of the two processors to RAM and rely on HT for the other processor to access the RAM.
This design choice by MSI dramatically reduces the RAM bandwidth available to the OS. So bandwidth sensitive apps can suffer.
It is something worth checking for just to be clear what hardware is really being tested.
Whenever a single memory bus design Opteron motherboard is selected for benchmarking vs. Intel, it is usually a sign that the review is fraudulent.
Most dual physical processor Opteron motherboards have two separate memory busses, one per processor, for a true NUMA architecture and full performance, at least when running Windows Server 2003 R2, Windows XP x64, or an appropriate NUMA Linux with true NUMA kernel.
Let me emphasize that if you do not use an appropriate NUMA compatible OS, you are not properly testing an Opteron system.
As most dual physical processor Opteron motherboards implement two independent memory busses, this is what should be tested for any sort of comparison.
I could not find anything on MSI's site that indicated what sort of memory bus configuration was used by the AnandTech review motherboard.
If indeed Anand picked a single memory bus design for the tested Opteron motherboard, then he is guilty of fraud.
I hope the facts come forward soon.
The MSI K8N Master2-FAR does not employ NUMA, CPU0 gets 6 Banks of RAM and CPU1 has to go through CPU0 to get RAM. Maximum memory bandwidth is 6.4GB/s but with the latency of ccHT for 1 bank of RAM, that motherboard is the worst board to use for testing. Either Anand is a moron and didn't know what board to use, or he purposely chose a crappy board.
I checked out the manual for the MSI K8N Master2-FAR and according to MSI, this is to be a Workstation board, not server. Last time Anand did a review of this, they said it was a "pretty cheap" board that should give them "decent performance" to test against and noted that it has single memory slot; yet this time they're using it as a high performance, server board? Give me a break....
I'm not sure why AnandTech would use the MSI board. But after putting a few Opteron worstations together, I would way the MSI board is the bottom of the barrel when it comes to Opteron mother boards. I woulds have suggested the Tyan Tiger K8WE (S2877). This is the best workstation class motherboard IMHO. If you want server class they could have selected any of the Tyan Thunder motherboards.
I'm sure the facts will come out when real intel parts are available to the public. Even it the Intel parts show some advantage, I think will will stick with AMD because I hate to qualify things all over again. It will be a while before I switch back to Intel.
I agree, an person that would spend the money on 2 server CPUs and then only use a motherboard with ONE memory bus for 2 CPUs is an imbecile. So why even bother with a comparison??? Wouldn't an imbecile like that just go with the cheapest setup? Because they definitely DO NOT care about getting the best performance...
So who is guilty? Linux or the Opteron system? We had to test with Solaris on the Opteron to be sure. However, the Serverworks chipset of our MSI 1U server was not supported by x86 Solaris. So we went back to our homebuilt server, based on the MSI K8N Master2-FAR.
That motherboard was used only for testing MySQL + Solaris + Opteron performance, after the linux setup showed degradtion. MySQL shows a 20% performance improvement. MSI K2-102A2M 1U server was used for the rest of the review
Peter posts a somewhat confusing reply to my comment on his blog. But seems to agree with Anandtech that MYSQL on linux has a performance bug.
http://www.mysqlperformanceblog.com/2006/06/08/intel-woodcrest-vs-amd-opteron-for-mysql/
But Anandtech uses benchmarks brewed inhouse, and has not posted the optimizations it used for tuning Gentoo performance.
I dont even know why they are calling it a independent review. Jonan from Anandtech posts reply for a few comments on the article, but skips queries on why they dont post optimizations used with MySQL and Gentoo
It seems the more I learn about Anand's so-called "reviews", the more I see Anand stacks the deck against AMD.
For example, no major vendor uses Gentoo Linux for NUMA performance. Gentoo is a developer's OS, not an enterprise server OS.
Instead of Gentoo, Anand should have used an appropriate *enterprise server* OS from Redhat or SUSE.
Furthermore, Anand's logic of throwing in results from running Solaris x86 on an old one memory bus Opteron board is severely flawed. If Anand had any integrity, he would have asked Sun for a review system with the appropriate Solaris x86 build. The current builds have many optimizations that probably don't get used properly on Anand's homebrew system.
All through Anand's reviews, there is simply no scientific method being followed. There is nothing except for Anand's half-baked hardware setups, wrong choice of OS, idiotic assumptions and endless pro-Intel conjecture.
In short, Anand should retire and join Tom Pabst in the Intel shill Hall of Fame.
This is just another proof point of why the Woodcrest announcement will be a resounding "thud". Opteron will pulverise on 4P, win handily on 2P and will be competitive on 1P. Wherever DirectConnect5 shows up, AMD wins.
As I remeber, amd issued a fair duel benchmark call to intel. Why, now if woodcrest is that good, intel doesn't accept the chalange??
Wodcrest is no more prototype, it is comercial product and i think that such duel would clarify some aspects. But intel doesn't even trow that to amd. They are just plain silent, and they leave it to anand and the rest to do their dirty work.
I used to be an avid Anandtech reader on their site/forums. Over the years (since the mid 90's), I've seen the evolution of their site.
I was there for the whole Staples ordeal (if you know about this, you're old skool like me)...
I banned Tom's HW for info gathering (although amusing at times), and Anand has joined my list of "NOT CREDIBLE" review sites anymore... no longer a visitor.
I'm in my 30's now, and learned thru the years that all these supposedly "review" sites are an extension of the Marketing departments of a lot of companies. Intel in particular...
In lieu of all that, as one poster commented, there are a lot of readers that just read that and take it as gospel without thinking. Heck, that's what Marketing is... to sway your decision regardless of the technique used. These review sites, are just that, marketing... :)
I'm eager to see non-futureware as well, and want to see what Conroes on the market will do. Bring the competition ON!!
I'll pick a new system come X-mas, and may the best company win. ;-)
Post a Comment
<< Home