Windows 8: Does AMD's Bulldozer Architecture Benefit?

[Crashman]

[citation][nom]dozerman[/nom]techreport.com/review/21865/a-quick-look-at-bulldozer-thread-scheduling Something isn't right here. Techreport clearly shows a good single-threaded bump in performance in software with less than five threads when affinity is manually scheduled.[/citation]WHEN AFFINITY IS MANUALLY SCHEDULED? WHO DOES THAT???

The point of the article was that AMD said Win8 would fix automatic thread scheduling.

 

[dozerman]

[citation][nom]Crashman[/nom]WHEN AFFINITY IS MANUALLY SCHEDULED? WHO DOES THAT??? The point of the article was that AMD said Win8 would fix automatic thread scheduling.[/citation]

The point is that if you can see thirty percent performance increase by manually scheduling threads, it should be a piece of cake to identify the situations where this is optimal and do the scheduling automatically. Why isn't this the case?

 

[Crashman]

[citation][nom]dozerman[/nom]The point is that if you can see thirty percent performance increase by manually scheduling threads, it should be a piece of cake to identify the situations where this is optimal and do the scheduling automatically. Why isn't this the case?[/citation]AMD told us to expect that. The article investigated whether the reality of Win8 matched expectations set by AMD, and it didn't.

 

[denbot]

The test is flawed, running BF3 with background processes will take total ram to and over 9gb.

In this test bed the ram and GPU would hold the game back not making full use of the CPU, as seen in the DIRT test results on the MED quality settings there's a gain of 10FPS, if that was a shooter 10 fps would be a nice difference an unplayable game a playable game.

This is likely to be down to the RAM/GPU not been fully used on Med..

I think this needs a retest with at least 12gb or RAM and a higher spec GPU or multiple 6950's.

Hope to see the results

 

[dozerman]

[citation][nom]denbot[/nom]The test is flawed, running BF3 with background processes will take total ram to and over 9gb.In this test bed the ram and GPU would hold the game back not making full use of the CPU, as seen in the DIRT test results on the MED quality settings there's a gain of 10FPS, if that was a shooter 10 fps would be a nice difference an unplayable game a playable game.This is likely to be down to the RAM/GPU not been fully used on Med..I think this needs a retest with at least 12gb or RAM and a higher spec GPU or multiple 6950's.Hope to see the results[/citation]

I have personally seen this take place with my system at home. This whole test feels rigged.

 

[Crashman]

[citation][nom]dozerman[/nom]I have personally seen this take place with my system at home. This whole test feels rigged.[/citation]If by rigged, you mean "configured to represent real-world high-end builds", then yes, it is rigged to represent reality.

 

[ronch79]

Regarding the last line of the article, yes, we all know AMD *is* aware of the shortcomings of its architecture and is working to improve upon it. Bulldozer is not as fast as SB/IB in single threaded apps, but it does have eight cores. I believe AMD's Mike Butler (BD Chief Architect) did admit during the BD press event that BD's most important contribution to x86 computing is that it allows scaling with core counts, not per-core IPC. The BD engineers had different goals in mind. AMD thought multi-core was going to be the way of the future and will keep scaling but they made the wrong bet. If every kind of software was easy to parallelize then BD would be in a much stronger position today than it is. As it is, it's a bit unfair to keep harping about BD's relatively weak single-thread performance given what the design is really all about.

 

[Crashman]

[citation][nom]ronch79[/nom]Regarding the last line of the article, yes, we all know AMD *is* aware of the shortcomings of its architecture and is working to improve upon it. Bulldozer is not as fast as SB/IB in single threaded apps, but it does have eight cores. I believe AMD's Mike Butler (BD Chief Architect) did admit during the BD press event that BD's most important contribution to x86 computing is that it allows scaling with core counts, not per-core IPC. The BD engineers had different goals in mind. AMD thought multi-core was going to be the way of the future and will keep scaling but they made the wrong bet. If every kind of software was easy to parallelize then BD would be in a much stronger position today than it is. As it is, it's a bit unfair to keep harping about BD's relatively weak single-thread performance given what the design is really all about.[/citation]The article concerns the validation of AMD's promises more than it is about AMD's technology.

 

[psiboy]

Thomas I'm with Silverblue. I wouldn't mind seeing if, with Windows 8 and the 8350, hardware can take full advantage of the software instead of the other way around. I was a little dubious about blaming Microsoft in the first place - this isn't quite the Vista scheduler and Phenom all over again.

I'd love to see all this data and comparisons done with the fx-8350 as well and rolled into the graphs for the fx-8150!

 

[littleleo]

It would be nice to see this updated with the new AMD models after 8.1 is released.