AMD Zambezi news, info, fans !

[informal]

There is another sisoft entry for "non-existent" Opteron ES 16C @ 2.6Ghz. This time results are even better than the last time when the model was listed as running at 2334Mhz. Now it is listed at 2.6Ghz and scores are almost proportionally higher.
Details for Device 2x AMD Opteron 6282 SE
Processor Arithmetic Benchmark 2x AMD Opteron 6282 SE @ 2.6Ghz 243.275GOPS (previous score @ 2.34Ghz was 231.9Gops )
Processor Multi-Media Benchmark 2x AMD Opteron 6282 SE @2.6Ghz 628.234Mpix/s (previous score @ 2.34Ghz was 585.9Mpix/s)


It appears that Turbo is not functional and newest integer throughput results for 4x Opteron 6272 @ 2.1Ghz confirm this ( it scores 377.7 GOPs with 2x more cores => 377.7/2=188.8Gops for 32 cores @2.1Ghz clock? => 188.8 x 2.6/2.1Ghz= 233.8Gops or virtually the same as "non-existent" 2x Opteron 6282SE @ 2.6Ghz ).

For comparison with MC 2P system @ 2.5Ghz :
Processor Arithmetic Benchmark 2x AMD Opteron 6180 SE @2.5Ghz 202.098GOPS
Processor Multi-Media Benchmark 2x AMD Opteron 6180 SE @2.5Ghz 331.207Mpix/s

Note that Turbo appears to be not working for integer throughput test in the case of 6282SE @ 2.6Ghz. Still it is faster than 2.5Ghz MC system by a good 243/202=1.2 or 20% in Processor arithmetic test and 628/332=1.89 or 90% faster in Multimedia test. Multimedia test uses AVX in case of new Opteron which gives around 11% better results on Bulldozer versus SSE test. So 1.89/1.11=1.7 or 70% faster than MC in legacy SIMD (SSE) throughput.
Looking at these latest numbers for 6282SE (which is supposed not to exist according to AMD ), FX 8150 @default with Turbo should be in the ballpark of : ~86-90GOPs and 205-210Mpix/s(184-189Mpix/s if you don't use AVX but SSE), in the arithmetic and MM tests, respectively. For comparison 1100T @ def. gets ~66.6 GOPs and 110-115 Mpix/s.

edit: note that these are still "ES" results and not from a retail/final platform

[Olivon]

You will find the original source by yourselves ...

[Andi64]

Shouldn't an hypothetical Phenom II X8 @ 4.8Ghz score between 10 a 10.5 points in CB11.5?

[ReKcOlNu]

If they're not going to bundle it with any CPU's, why even bother?

If it is sold seperatly then the warranty/RMA can be limited,... If the cooler leaks they'll replace it but any damage caused by the leak will not be covered.

[informal]

You will find the original source by yourselves ...

Yes I'm aware of those. I am speaking about Sisoft sandra opteron and Zambezi results,not C11.5.

[Manicdan]

Shouldn't an hypothetical Phenom II X8 @ 4.8Ghz score between 10 a 10.5 points in CB11.5?

the old architecture gets about .3 pts per ghz per core, round down a little for inefficiencies

so up to 8.4pts at 3.6ghz for an octo PII
11.3 pts for 4.8ghz

keep in mind that in the CB11.5 thread weve seen a half point spread even at the same ghz. but we also see people getting mid 7pts in mid 4ghz. so these results look bad

[Manicdan]

double post

[Olivon]

Yes I'm aware of those. I am speaking about Sisoft sandra opteron and Zambezi results,not C11.5.

Why you feel concerned ?

I just post informations that I just find at the moment, continue speaking about Sisoft Sandra if you want dude ...

[informal]

No offense but that information you've "just" found is old information (and you know it). We have known this already for about a week or so. It was posted in this very topic. What I posted in my post was latest sisoft entry for Interlagos.

[Olivon]

You're right informal, but that's the first time I see the visual about the thing and I share as I always did.

[Macadamia]

You will find the original source by yourselves ...

That slide deck looks like the biggest pile of malarkey I've ever seen.
Logos that don't align and have double strokes, vector arrows that blur out while text is still sharp, ridiculously huge text on slides with very little explanation.. missing footnotes, no benchmarks, and AMD hasn't used Cinebench screenshots for benchmarking to date (it's always a table + specs on the same slide)


Takes a Czech to create such dumbassery- and to the nationalities of whoever believes that... hah.

[Canis-X]

First time that I've see it as well.

[r.p]

haha the cinebench slide is really funny:

cine_fun.jpg

[informal]

haha the cinebench slide is really funny:

cine_fun.jpg

Now that you mention it , it's really odd that AMD will use intel hardware running OS X and put this into their own official slide deck. One would think they would use opteron or phenom or athlon for that screen capture of c11.5. Or even more logical is that they would use actual Zambezi with the actual score of 5.95 showing up in the screenshot. Strange stuff. This of course doesn't mean that the slide deck is "fake" or not real,just that it is odd.

[Oese]

fake 100%

[undone]

Not really fake, but means nothing.
I have a stronger feeling that naming scheme would be changed when Zambezi is released, just like obsolete 8130p, so we might needn't bother how 'FX-8150' perform.

[rog]

fake 100%

No it's not. AMD used the screen from here: http://www.maxon.net/uploads/pics/ci..._screen_16.jpg

IPC is lowered with Bulldozer.

[flyck]

Now that you mention it , it's really odd that AMD will use intel hardware running OS X and put this into their own official slide deck. One would think they would use opteron or phenom or athlon for that screen capture of c11.5. Or even more logical is that they would use actual Zambezi with the actual score of 5.95 showing up in the screenshot. Strange stuff. This of course doesn't mean that the slide deck is "fake" or not real,just that it is odd.

slides are from oo..., the one who shall not be named. So we can just quit the descussion about those posts. Whether it is true or not, he isn't worth the effort.

There is another sisoft entry for "non-existent" Opteron ES 16C @ 2.6Ghz. This time results are even better than the last time when the model was listed as running at 2334Mhz. Now it is listed at 2.6Ghz and scores are almost proportionally higher.
.
.
.
Note that Turbo appears to be not working for integer throughput test in the case of 6282SE @ 2.6Ghz. Still it is faster than 2.5Ghz MC system by a good 243/202=1.2 or 20% in Processor arithmetic test.

Thats brings integer alot lower per clock than phenom.
http://www.sisoftware.eu/rank2011d/s...284f7cafa&l=en

Same core count, lower frequency and higher performance. So like you mentionned, BD seems to have alot of problems integer wise, fpu wise they are pretty good.

[Oese]

No it's not. AMD used the screen from here: http://www.maxon.net/uploads/pics/ci..._screen_16.jpg

IPC is lowered with Bulldozer.

There's more fake to this then that I believe AMD would not use some screen like that or make a presentation anywhere near to this. This is worse then 1st college class presentation

And yes I believe too IPC of a single BD "core" is maybe slightly lower then K10.5 core, but this doesnt mean these slides are not fake xD

[Mechanical Man]

Im just saying, even when some is clearly fake, it still does not rule out that there is slide deck from such event.

[spurious]

Even when Interlagos shows up, it still will not tell us the performance of Zambezi. I don't know if I read it here or somewhere else that Interlagos is going to use an earlier stepping. The problem with the earlier stepping does not effect server workloads as much. Can anyone confirm this?

[JF-AMD]

I don't believe that is true.

[BeepBeep2]

Shouldn't an hypothetical Phenom II X8 @ 4.8Ghz score between 10 a 10.5 points in CB11.5?

the old architecture gets about .3 pts per ghz per core, round down a little for inefficiencies

so up to 8.4pts at 3.6ghz for an octo PII
11.3 pts for 4.8ghz

keep in mind that in the CB11.5 thread weve seen a half point spread even at the same ghz. but we also see people getting mid 7pts in mid 4ghz. so these results look bad

You guys seem to have missed what chew said...
1. Bulldozer functions like a 4 core that is able to execute 8 threads. You get roughly 5x scaling in CB from one to eight cores. (Better than HT) I don't care what PR tells you, that's how it works. (Unless new charts have come out with magical 7.x scaling?)
2. You have two options with the "STARS" core.
...Option 1. Shrink X6 (probably with redesigned Llano "STARS" core, so 3% IPC increase lets say) and increase frequency a tad. Lets say that this arch. will do 4.5 Ghz.
...Option 2. Add more cores, most likely decrease IPC from Thuban or keep it the same using Llano's tweaked core...lets say this arch will do 4.1 Ghz.

Bulldozer. More IPC than X4 in 4 threads or less, "less" distributed per "core" in 8.

Also, IPC is so far behind in Phenom II, why would we use STARS again? We are verging on 25-30% slower than Intel CPC, with a 15% frequency difference and new intel chips are doing 3.8 Ghz turbo, OCing to 4.8-5.0. We need a change. (If 5% IPC hit means 15-20% more clocks, I'm all for it tbh.)

You guys are so focused on multi-threaded results it's rediculous. Trying to predict single thread performance with a multithread benchmark is rediculous.

It's kind of like comparing a 2600K that does 5 Ghz on air to a 980X that does 4.3. Do you really want that 980X because its Cinebench score is higher, or would you like the extra single thread perf?

With no real CLEAR results WITH explanations out yet, I am still firm with my belief that in layman's terms, CMT = AMD HT in physical form that can not be turned off, trumping Intel's HT.

[HTC]

Any word on stable underclock potential yet? I'm referring to power savings, when compared to stock.

[Manicdan]

You guys seem to have missed what chew said...
1. Bulldozer functions like a 4 core that is able to execute 8 threads. You get roughly 5x scaling in CB from one to eight cores. (Better than HT) I don't care what PR tells you, that's how it works.
2. You have two options with the "STARS" core.
...Option 1. Shrink X6 (probably with redesigned Llano "STARS" core, so 3% IPC increase lets say) and increase frequency a tad. Lets say that this arch. will do 4.5 Ghz.
...Option 2. Add more cores, most likely decrease IPC from Thuban or keep it the same using Llano's tweaked core...lets say this arch will do 4.1 Ghz.

Bulldozer. More IPC than X4 in 4 threads or less, "less" distributed per "core" in 8.

no one knows the the exact scaling because its hidden behind NDA.
and if they did stick with PII architecture it would be more than 3% because llano does quite nicely for not having and L3. it would also pack in more true cores. by omitting the extra space used for the second thread of the module, the core itself got bigger with BD, so IPC should be higher or they found ways to waste more space. (btw your option 2 makes no sense, why reduce IPC since they can easily pack in 8 old cores in BDs size, and why would that drop the clocks by 400mhz)

i honestly dont believe in just 6pts for BD at 3.6ghz, and if that is the case its probably on the very low end of the spectrum for how it handles otherwise it would be a step backwards.