AMD Zambezi news, info, fans !

[Spectrobozo]

if priced right the FX-4170 can be interesting against the i3 2100, looks to be a bit faster on cinebench, about the same on 3dmv, but while the i3 2100 is locked the FX-4170 can probably be overclocked... but still... IPC seems very low, single threaded performance to... I don"t know, but considering how a PII X4 955 perform and that it costs only 120 usd... this looks very uninteresting for the most part....

[mav2000]

Does that say ES on the CPUZ?

[rog]

Edit: The guy's still on 08XX BIOS, so far rather mediocre results on most benchmarks, but...

08xx ist final review Bios version for the tester. 0813 to be exact. Performance wise no difference to the older 0705 Bios.

Need to see the NB speed on that, cause that is TERRIBLE memory latency
Also, wasnt B2 the not good stepping?

What stepping do you expect? B2 is release step.

[EniGmA1987]

What stepping do you expect? B2 is release step.

Is it? I must have been thinking B1 then

[rog]

Is it? I must have been thinking B1 then

Why do you think B1? B2 is known since ages.

[Daveburt714]

I could be wrong, but it seems like there were a number of B2 steppings. B2f, B2g etc...
Not sure how that would be shown with software, CPU-z or otherwise.

[rog]

I could be wrong, but it seems like there were a number of B2 steppings. B2f, B2g etc...
Not sure how that would be shown with software, CPU-z or otherwise.

It's not wrong but it is still B2. In CPUz we see only the "2".

[Pt1t]

only cpuz 1.58.6 and 7 can read correctly the CPU name

[dess]

WTF!!!!!! Why CB shows 4.22GHz FX4170 faster than 4.22GHz 8150??????

1. Probably the all-cores-on or the half-the-modules-on turbo could be activated on 4170. On the 8150, those were perhaps disabled or could not be activated.
2. Both have 8 MB L3 cache -> 1 MB/core for 8150, 2 MB/core for 4150.
3. BIOS (perhaps).

does this superpi mean anything? my 5 year old core 2 duo with ddr2 does 1M in 25s.

It uses a very ancient x87 code from the 90's, optimized for Intel.

TDP is thermal design power, has nothing to do with power the chip pulls from the board but the heat it gives off.

Nothing to do? 99% of energy pulled will turn into heat in the CPU. (The remaining 1% will leave on the I/O lines.) Elementary physics, boy!

Therefore, a 125w chip should run about as hot as any other 125w chip assuming they are same architecture with same thermal sensor...(or you could simply use an external probe, like k-type bead or flat diode)

At stock/original turbo clocks and voltages... With overclocking and over-voltaging the power draw/heat production can increase significantly.
The FX CPU's, being Black Edition, is meant for serious O.C.

BTW, TDP is not at all average power draw/heat production. It's assigned in relation to the maximal power draw/heat production of the chip (*), at stock/original turbo clocks and voltages. Moreover, there are TDP classes. If a given value doesn't fit one class, a higher TDP class will be assigned to the chip. Fx.ex. 95W -> 95 W TDP; 96W -> 125W TDP, for AMD.

(* - For AMD, it's the absolute maximum with the most demanding task. For Intel, it's the average of maximums with a selection of everyday tasks.)

[informal]

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.

[Oese]

I still think there might be something wrong with the frontend or with FlexFP utilization.

[liberato87]

AIDA 64




it seems something is wrong in l1 cache bandwith (write and copy)..

fx 4170
L1 WRITE 23477 mb/s
L1 COPY 47025 mb/s

x4 980
L1 WRITE 65535 mb/s
L1 COPY 87317 mb/s

[Apokalipse]

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 that figure is not compared to Deneb, it's compared to a single thread in the module.

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?

I'm gonna say this isn't final performance.

[Oese]

old AIDA

this one is from the review kit online since yesterday (newer version no.):

[informal]

But that figure is not compared to Deneb, it's compared to a single thread in the module.
I'm gonna say this isn't final performance.

Yeah of course it's not compared to deneb,it was meant versus a hypothetical "other" bulldozer. But now we have deneb to compare with since that "other" BD is never going to happen. And comparing it with Deneb it is slower per "core" by about 25% in SIMD workloads. It is also slower by the same amount in single thread SIMD when supposedly whole flexfp is on one core's disposal. We don't see this behavior anywhere. One core has roughly one fmac to deal with,it never gets the second one or there is a problem that stops the first core from utilizing the whole flexfp. We see the whole flexfp used ONLY when we have second thread from second core working,which goes against what AMD engineers and marketeers talked about for 1.5+ years now.
Untitled.jpg
As for this not being final performance,this might be true,but remember these results are coming from press kits AMD sent out to reviewers. And FX4170 is behaving exactly like the press kit 8150 with half the cores disabled. So it is pretty much the final performance,which is kinda sad. We didn't expect to see 20+% slower per "core" results,at least not with all the talk about IPC going up,single thread integer IPC going up,BD being "HPC monster design". None of these are true.

[xdan]

But that figure is not compared to Deneb, it's compared to a single thread in the module.
I'm gonna say this isn't final performance.

Quite interesting.
So when i said that BD is not at 8 core it's an quad 8 threads, many guys just don't stop saying is an 8core with cores that share resources.
But you say that we shouldn't compare 1core/1thread of Deneb with 1"core"/ 1thread of BD? but with an module?
Than what's FX 4170 a quad or a dual?
Final performance it how it is, bad.

[Oese]

it is bad, single-thread.

and it is bad, marketing-wise. If this would be regarded as a dual-core, 4-thread chip, it might not be as bad.

Its still a modern, multithread design, even for this 2-module-variants, thats clear.

I dont expect nice single-thread numbers, but maybe multithread it might compete.. Still, Power consumption is one of the most important figures now.

[informal]

Summary of FX4110 @ 4.2Ghz vs Phenom II X4 980 @ 3.7Ghz:
super pi :meh I don't want to bother finding results in this "benchmark"
fritz chess: FX4110 gets 7332pts, X4 980 gets 9067pts. FX is 24% slower at stock(vs stock) and 40% slower per "core" at the same clock. Multithreaded benchmark.
3d mark vantage CPU test : FX4110 gets 10660pts, X4 980 gets 12780pts. FX is 20% slower at stock and 36% slower per "core" at the same clock.Multithreaded benchmark.
wprime 32m: FX4110 gets 17.9s, X4 980 gets 11.45s. FX is 56% slower at stock and 77% slower per "core" at the same clock.Multithreaded benchmark.
c11.5 : FX4110 gets 3.42pts, X4 980 gets 4.34pts. FX is 27% slower at stock and 44% slower per "core" at the same clock.Multithreaded benchmark.

I have skipped over 7zip since I can't find comparable benchmarks . Aida cache and memory shows somewhat better memory read/write and L2/L3 cache BW for reads. The rest of cache performance is on par or slower than Deneb.

Conclusion: overall FX4110 is 32% slower than Deneb X4 @ 3.7Ghz stock vs stock and 49% slower when both are at 4.2GHz. Either all these tests are failure of the platform bugs (or something else) or Bulldozer is much slower than Deneb with the same "thread" count. All above tests utilize the "world's first 256bit FPU" and it fails hard versus "old 128bit" Deneb FPU,even in single thread mode... Imagine the OC you have to reach to just match Deneb,it has to be sky high (think 5.5-6Ghz on air to match 4Ghz Deneb). How is AMD going to charge 140$ for this chip is beyond me.

[Oese]

indeed.

however, this doesnt fit the marketing slides... except u take 2-module BD as 2-core chip, 4-threaded..

[radier]

More in this thread: http://forum.overclockers.ua/viewtopic.php?f=2&t=42451

[informal]

Thanks raider!

So 4.5Ghz on 8150 and it scores 40.5Gflops in Linpack. For comparison this is what Deneb X4 @ 3.8Ghz gets : 34.7Gflops (or @ 4.5Ghz 41Gflops). So 8C Bulldozer is slower than 4C Deneb in pure FP benchmark that is used to measure HPC performance in Top500 supercomputer list... I have no idea what is going on with the FPU in Orochi.

[radier]

I have only one saved test at 4,5GHz on my 2500K from IntelBurnTest which is using the same linpack library. It was over 62GFlops.

[informal]

Forget SandyBridge,it's another level. This thing can't match Deneb X4,let alone X6 Thuban or SB...

BTW I just noticed something strange with these tests. That Russian dude definitely has the chip and the board,so results are not "fake" (but real indeed). How come at 4.5Ghz he scores in super pi 1m a score of 1 minute and 42s ??? Is he running at 500Mhz in reality or what? Either his chip is throttling or his platform is bugged. Zambezi seems really slow,but no way Zambezi is THAT slow . It should score at around 18-19s,instead of 1minute 42s.


EDIT: ahh man,you tricked me . The thread is 2 months old and it's about 8130P that is ES and doesn't exist in retail form... So all those results are irrelevant ,even if they are close to retail(like I believe linpack is).

[Oese]

http://i.piccy.info/i7/dacae6ac61908...35/bulaida.jpg

AIDA bench with newest version. Seems same copy/write L1.. So there definitely we have an issue..

[informal]

Well that guy's 8130P performs exactly like 8150 so all those leaks from china with 8130P ES results were true... The Russian dude posted results @ 4Ghz in cinebench,linpack,winrar and all match 8150 from press kit.