realworldtech article on k10

[dmens]

http://realworldtech.com/page.cfm?ArticleID=RWT051607033728

 

[myocardia]

Very interesting. Now they just need get them into retailers' hands, so we can see how they perform in real life situations, with real software.

 

[pm]

It's a very good article. As always, David makes very insightful observations. Thanks for posting it, dmens.

 

[coolpurplefan]

Uh, did they just say what I think they said in that article? It seems they're saying single core won't be much faster. Sheesh, that kind of makes me think of the new video cards. Better performance in DX10 may be their only hope. Now it looks like they'll only dominate with Barcelona if software is multi-threaded.

 

[Griswold]

Originally posted by: coolpurplefan
Uh, did they just say what I think they said in that article? It seems they're saying single core won't be much faster. Sheesh, that kind of makes me think of the new video cards. Better performance in DX10 may be their only hope. Now it looks like they'll only dominate with Barcelona if software is multi-threaded.

Without reading it yet, I'm absolutely sure that this is the synopsis - why else would dmens link to a K10 article if it doesnt somehow relativize K10? We'll see if its true.

 

[pm]

Well, since I too am an Intel employee, I would like to point out that the reason that I thought that it was a good article has nothing to do with the performance section at all but because the author definitely did his homework - among other things, the author estimates the number of FO4 delays in the pipeline. It's well written and it has very insightful technical details about the lower-level operation.

I honestly thought that it was just a very interesting read.

I'm not sure where exactly you guys are getting the performance information from - but to me the performance section was the weakest section of the article... he seemed to be making vague guesses. The thing that makes it a good article is that it explains the new features in pretty good technical detail. If you want performance numbers, wait and see how the benchmark numbers look when systems are reviewed... but if you want to understand what was changed... then go read David Kanter's article.

 

[dmens]

I linked to it because the article has actual useful information and is a very worthwhile read, unlike faud's obviously fradulent sandra pics, or kyle's parroting of AMD marketing soundbites.

 

[Regs]

I only read 2-3 pages and I'm almost drowned with many good points. Like how AMD's prime target seems to be the server market. It confuses me because AMD just strike a deal with Dell which is a vendor that reaches millions of consumers. Now only if AMD could capitalize on something targeted for Dell consumers it would make a lot of revenue for them.

In any SWOT that should be their #1 opportunity.

Though the paragraph that stiked me more interested in out of all of them was the one about branch prediction.

According to our own measurements for several PC games, between 16-50% of all branch mispredicts were indirect (29% on average). The real value of indirect branch misprediction is for many of the newer scripting or high level languages, such as Ruby, Perl or Python, which use interpreters. Other common indirect branch common culprits include virtual functions (used in C++) and calls to function pointe For the same set of games, we measured that between 0.5-5% (1.5% on average) of all stack references resulted in overflow, but overflow may be more prevalent in server workloads

Improved branch prediction alone could give consumers outside the server market a little more incentive to upgrade from the K8.

 

[dmens]

well, now im really curious about k10 performance on single threaded workloads. the article statement regarding three-wide being a performance non-issue is confusing. that change on conroe resulted in significant gain.

 

[Special K]

I found the article very informative. After taking some of my advanced classes, all of those block diagrams now make perfect sense to me:thumbsup: