AMD Zambezi news, info, fans !

[BeepBeep2]

I hope some one test Fx vs I7 on Linux as well , really want to understand what is going wrong with Fx , is it equally bad under Linux as well.

Phoronix noted a trinity 2.5 Ghz sample doing "very well" against a 2.5 deneb based laptop...

[Mechromancer]

Does Windows need a patch to make the AMD processors work correctly AGAIN?

[Nintendork]

If at least the power consumption is real, then we have that this particular FX doesn't face any stress in some benchs. Some pages behind there was a post mentioning how FX power consumption was extremely low despite running CB11.5.

Since this violates the NDA then the "reviewer" should have no problem measuring how many watts this FX eats in all benchs vs the SB chip.

Doesn't make sense.

Also, why include the useless superpi? Doesn't even worth any decent reviewer's time.

[BeepBeep2]

If at least the power consumption is real, then we have that this particular FX doesn't face any stress in some benchs. Some pages behind there was a post mentioning how FX power consumption was extremely low despite running CB11.5.

Since this violates the NDA then the "reviewer" should have no problem measuring how many watts this FX eats in all benchs vs the SB chip.

He accidentally swapped the graph.
That's 130w for FX, ~100w usage for SB.

30% more power usage, failing MT perf and really failing ST perf. Thuban does better on a 45nm process...though consumes a small bit more.

[xVeinx]

Phoronix noted a trinity 2.5 Ghz sample doing "very well" against a 2.5 deneb based laptop...

That doesn't mean much. Phoronix isn't a completely bogus site by any means, but they aren't exactly "scientific" in their benchmarking either...

[mav2000]

Well if this is true then I am sticking with my 1090t. And thats that. This has really broken my heart.

[PatRaceTin]

some more leak to come

[Baam]

Well if this is true then I am sticking with my 1090t. And thats that. This has really broken my heart.

Me too. Might grab a 960T to play with till a revision or Pile-driver comes.

[spursindonesia]

I'm with you Dimitriman .. I can't understand the logic they used to get where they are now. They knew ,back in 2008,with what they are dealing with from the other side. They could even make projections with what will intel come up come 2010/2011. With all this foreknowledge they produce chip that on paper has 33% more cores,but since each core is weaker,especially in SIMD workloads (which are crucial aspect of modern day CPU performance),it can't outperform the old design by more than 20%. Looking at highly MT scores,Thuban is just behind it in handbrake(video transcode) and cinebench(3d rendering)-and these 2 are supposed to be strongholds of Bulldozer design. As for ST performance,they talked about super prefetch and other things that supposed to offset the deeper pipeline and they talked about whole flexfp executing 256bit for single thread code etc. None of these things help Bulldozer outperform a 2007 core(K10) that got a very mild facelift in 2008/09 (in Deneb iteration). It's pretty disappointing.

Now there is this clock speed headroom which is a definite plus of the design,but users will run in the wall around 4.8Ghz on air. This is 17% faster than average Thuban OC so in best case you are just matching Thuban's "pure speed"(aka IPC) in apps like games (which won't support 8 threads anytime soon so core count advantage won't kick in) and poorly threaded workloads which are just the majority of desktop application types. Where you can actually use 8 cores/threads ,then this 4.8Ghz OC vs Thuban's 6C running at 4-4.1Ghz will probably be around 33% faster,but this is just in a few applications .

Zambezi core in FX4100 has exactly 25% lower SIMD/fp performance than Deneb core... Fits kinda well with that 80% claim from the slides. But AMD said they expanded the unit and some other improvements so that SIMD workloads at least won't be slower than family 10h. We see that they missed the mark by 25% since it takes Phenom @ 3.3-3.4Ghz to match FX4100 @ 4.2GHz. Pretty disappointing IMO. Even in single thread department Zambezi's "mighty 256bit" FPU fails to beat Deneb core from 2008. Wtf is going on AMD?

Again I must wonder how in the world is Interlagos going to outperform MC @ 2.5Ghz in legacy (non recompiled) HPC workloads by more than 35% ?? According to all the leaks it should barely tie it or with help of Turbo beat it by 10-15% tops (if that much). Pretty puzzling. Not to mention AMD state that peak Flops are the same for legacy and AVX (recompiled) binaries,so no magic there either.

YUP, one word fits this launch, "DISAPPOINTING".

[Leeghoofd]

Also, why include the useless superpi? Doesn't even worth any decent reviewer's time.

Still an excellent test for ram stability, gain with extra ram speed, tightening subtimings... Eg : maxxmem is bugged, AIDA/Everest soft can output impressive nrs, yet they don't reflect in any other benchmark... Always nice to know how it performs under SuperPi

[radier]

http://lab501.ro/procesoare-chipsetu...ldozer-preview

[Apokalipse]

Why do people still believe those results are accurate for retail chips?

[Hondacity]

Why do people still believe those results are accurate for retail chips?

why would amd hold back their performance numbers? its a few days before the launch.

[xdan]

Still many unbelievers...
The results are true, Mosntru and Matose i think that they are the most reliable sources.
BD is what it is, a FAIL.
FX4170 will also fail to beat X4 955.

[Apokalipse]

Still many unbelievers...
The results are true, Mosntru and Matose i think that they are the most reliable sources.

the results are probably true for their particular configuration, which is to say there's something missing from retail configuration.

AMD wants to keep Intel in the dark, and have shown they're capable of keeping a secret (Eyefinity, RV770 shader count)

[Smartidiot89]

Why do people still believe those results are accurate for retail chips?

That is a retail sample...

[Apokalipse]

That is a retail sample...

And there's more than just the silicon that affects performance.

[Smartidiot89]

And there's more than just the silicon that affects performance.

Okay, I will put this in your language then... This is a retail configuration.

[xdan]

And there's more than just the silicon that affects performance.

Yes the arhitecture which was wrong from the begining.
What if the penalty from sharing resources between cores is biger than AMD was saying.
Finally three more days to the complete truth.

[Apokalipse]

Okay, I will put this in your language then... This is a retail configuration.

Which is mostly irrelevant for the CPU performance.
There's much more to it that what's listed there.

[Smartidiot89]

Which is mostly irrelevant for the CPU performance.
There's much more to it that what's listed there.

Ok wise guy, you can choose not to believe me then

[TESKATLIPOKA]

BeepBeep2

Phoronix noted a trinity 2.5 Ghz sample doing "very well" against a 2.5 deneb based laptop...

from the site

This Trinity APU is quad-core and running at 2.50GHz. The current quad-core Llano offerings are clocked at 2.6GHz (A6-3650) and 2.9GHz (A8-3850), while this Trinity part is clocked slower, it's numbers are nice compared to my A8-3850 Linux system.

http://www.phoronix.com/scan.php?pag...ty_early&num=1
Lower clock(16% difference) and module penalty should mean not just +10% but a rather big IPC increase.

xVeinx

That doesn't mean much. Phoronix isn't a completely bogus site by any means, but they aren't exactly "scientific" in their benchmarking either...

I don't know how good is their benchmark testing but Llano and SB2100 are comparable and the same situation is in Windows, so I am pretty optimistic but I hope It won't end like current BD launch.

[informal]

I think this is it,performance wise. There is no magic dust. Something is wrong with this thing since this cannot be the same design that was presented 1 year ago at HC22.That design had at least comparable performance versus the old core. This one is not only slower in ST,but sky high Turbo doesn't help it to beat the old core. Then on top of that comes poorer MT scaling and this further more lowers its performance in well multithreaded applications. And then comes the very poor SIMD performance,think Bobcat level stuff. I doubt we will hear that anything is wrong with Bulldozer and that this is the thing they were describing all along...

The funniest thing is soon AMD will release a 3Ghz QC Llano desktop chip that will maybe even have a slight Turbo for cpu cores. Imagine this 32nm 3Ghz K10 keeping up with the 4Ghz FX4110 stock vs stock in many benchmarks,it's going to be very funny to see the marketing spin on that.

[Apokalipse]

Ok wise guy, you can choose not to believe me then

Things that can affect CPU performance include:
* BIOS settings/features
* microcode
* CPU thread scheduler (OS), or various patches for the OS
* program compiler, and what instruction sets are used
* CPU revision/stepping

Some things are known, but some things aren't.

[nex_73]

Ok wise guy, you can choose not to believe me then

I think Apokalipse means more like:
Fine we know what stuff (hardware) that is used, but what about bios/uefi (agesa microcode version) & settings?
Also we know OS, but what driverversions (mostly chipset & video) are used and settings?

Did you se the PM from me requesting the "performance per watt" comparison in your upcoming review?
Would be fine if you have time to squeeze in a Win 8 dev preview test in it to

Edit: he was faster