Linux, Open-Source Affected In AMD Cutbacks?

[leeenux]

BD's about equal to X6 in many loads, yet more expensive and using more power under load.

That would be the FUD we're talking about, because what you're saying just isn't true.

Look here:

http://www.phoronix.com/scan.php?pag...ulldozer&num=6

There's no X6 in the benchmark, but the 2 x Opteron machine could be considered an excellent point of reference. 8 very similar cores, reasonably similar clocks to the X6. In a well threaded benchmark, it's actually about 20% beastlier than an X6. Notice that FX CPU is significantly faster.

Now look here:

http://www.phoronix.com/scan.php?pag...ulldozer&num=7

Bottom graph. The aging Core i7 920 soundly defeated it's newer cousin, the Core i5 2500K, despite a 700mhz clock deficit. Yes folks, that would be a regression, the kind that AMD has with Bulldozer, but supposedly Intel never has.

I trust you'll do the right thing and go trolling over the entire internet about what a failure Sandy Bridge is.

[deanjo]

Frankly I agree with deanjo. He's being honest, I couldn't see any FUD there. BD's about equal to X6 in many loads, yet more expensive and using more power under load. How does that spell "good"?

I actually postponed my purchases a couple of months to see how BD ended up. Guess what? Now typing this on an X6.

Ya I'll be hanging on to the 1090T it looks like until I can see what the Vishera respin will bring. Hopefully they will have an updated chipset by then as well.

[deanjo]

That would be the FUD we're talking about, because what you're saying just isn't true.

Look here:

http://www.phoronix.com/scan.php?pag...ulldozer&num=6

There's no X6 in the benchmark, but the 2 x Opteron machine could be considered an excellent point of reference. 8 very similar cores, reasonably similar clocks to the X6. In a well threaded benchmark, it's actually about 20% beastlier than an X6. Notice that FX CPU is significantly faster.

Actually that would be where you are wrong. Even the 1090T beats the 2 x Opteron quite handily in most, including the heavily threaded tests such as x264. NUMA has some hard penalties.

http://openbenchmarking.org/result/1...AR-X2641090T17

And that is with a 600 Mhz deficit and stuck with DDR2-800. Pushed up to BD's 3.8 GHz the 1090T starts pulling away.

[leeenux]

Actually that would be where you are wrong. Even the 1090T beats the 2 x Opteron quite handily in most, including the heavily threaded tests such as x264.

References or it didn't happen.

Please let me know what you think about the 2nd part of my post that you didn't quote, and how it relates to your anti-Bulldozer crusade.

[deanjo]

References or it didn't happen.

Please let me know what you think about the 2nd part of my post that you didn't quote, and how it relates to your anti-Bulldozer crusade.

Look at the linked results.

[deanjo]

http://www.phoronix.com/scan.php?pag...ulldozer&num=7

Bottom graph. The aging Core i7 920 soundly defeated it's newer cousin, the Core i5 2500K, despite a 700mhz clock deficit. Yes folks, that would be a regression, the kind that AMD has with Bulldozer, but supposedly Intel never has.

I trust you'll do the right thing and go trolling over the entire internet about what a failure Sandy Bridge is.

Lol, compare apples to apples. The i7 920 is capable of 8 threads. The i5 2500k has no Hyperthreading. Throw a 2600K in there and watch it thump the i7 920.

[leeenux]

Lol, compare apples to apples. The i7 920 is capable of 8 threads. The i5 2500k has no Hyperthreading. Throw a 2600K in there and watch it thump the i7 920.

So now hyperthreading is analogous to 8 real cores, and not just a gimmick that delivers -20% to +20% performance depending on the benchmark?

The 920 was 25% faster than the 2500k. Normalize the clocks, and it's 50% faster than the 2500k, or even more depending on turbo. So, in order to not just tie, but "thump" the 920, hyperthreading would need to offer twice the performance per core of a non-HT core, which has no chance in hades of happening. Nice try, though.

[Jf4frWil]

I will miss their slides...

[ooHe0Eejem1]

The way I understand it, Bulldozer is taking a beating not because it's that bad, but because Vista7 isn't really optimized for it and the Tom's Hardware like sites are giving it a beating because Windows is crap.

The Linux kernel already has patches that improve the performance quite a lot and they will be part of the kernel within the next three months.

Maybe AMD should quit recommending Vista7 since it seems it's not treating their new architecture that well.