SiSoft Leaks AMD FX "Zambezi" Scores: Worse Than Intel Core i7

[Edrick]

Sorry if this has been posted in the other BD benchmark thread....I gave up on that thread. So if this is a re-post, please remove.

http://www.brightsideofnews.com/new...-zambezi-scores-worse-than-intel-core-i7.aspx

 

[LondonBurning]

Take that story with a grain of salt. AMD have already stated any of these 'benchmarks' (obviously not using final product, which is barely hitting mass production now) will not be an accurate representation of the final product.

 

[notty22]

Here are the images of results.
all below a copy+paste.
This was the test setup:

And these were the results:

 

[Idontcare]

Lends credence to the view that the 4.2GHz 4170 SKU is intended to go toe-to-toe with the 3.3GHz 2500K. The clock-delta is supported by the Sisoft results.

 

[podspi]

Those results are so bad, that it is hard to believe they are representative of final shipping product.


As I said a while back, K10.5 (K10.6?) IPC has got to be the lower-bound. AMD would have to be insane to actually try and pull a Netburst w/ out the fab advantage that Intel has.


The amount of ridiculous information coming out about BD at this point is crazy. It's fast, its slow, its BOTH, another revision is coming in January 2012, its being released in a few weeks, its being released next Monday, it got cancelled, never existed, is actually Chuck Norris... I don't know what to believe anymore!

 

[StrangerGuy]

Lends credence to the view that the 4.2GHz 4170 SKU is intended to go toe-to-toe with the 3.3GHz 2500K. The clock-delta is supported by the Sisoft results.

"I feel a great disturbance in the Force, as if AM3+ owners suddenly cried out in terror, and were suddenly silenced."

 

[MrTransistorm]

What does SiSoft use for CPU video transcode testing? It doesn't add up. Based on the fact that the 6 core 1100T is only slightly behind the 2600K, I would expect an 8 core BD to beat it.

 

[nonameo]

Which sandy has a base clock of 3ghz? I dug through cpu-world for both xeons and i5/i7's and could not find one.

 

[frostedflakes]

Doesn't look great. D:

To be fair, though, that seems to be the lowest clocked 8-core Bulldozer not the flagship 8150, which is supposed to run at 3.6GHz non turbo 4.2GHz turbo I believe. Also what's the memory controller speed, is that CPU-NB frequency? If so shouldn't it be running at 2600MHz, not 2200? edit: Actually looks like BD CPU-NB is 3200MHz

^ And yeah I was wondering about Sandy clocks as well.

 

[sequoia464]

"I feel a great disturbance in the Force, as if AM3+ owners suddenly cried out in terror, and were suddenly silenced."

Just picked up an Asus Sabertooth. Rapidly looking like a mistake

 

[EightySix Four]

What does SiSoft use for CPU video transcode testing? It doesn't add up. Based on the fact that the 6 core 1100T is only slightly behind the 2600K, I would expect an 8 core BD to beat it.

This does not make sense.

 

[BlueBlazer]

Lends credence to the view that the 4.2GHz 4170 SKU is intended to go toe-to-toe with the 3.3GHz 2500K. The clock-delta is supported by the Sisoft results.

As for the 4 "core" version, there's also a SiSoft Sandra result for Processor Arithmetic Benchmark here >> AMD Eng Sample, ZD302046W4K43_36/30/20_2/8_A (2CU 4C 3GHz, 2GHz IMC, 2x 2MB L2, 8MB L3) (which I've posted with comparisons earlier). It does seem that 4GHz+ clock speed was the only way to make this SKU competitive. :\

 

[Riek]

Seriously guys.... Brazos is faster in those benchmarks then BD!!!

dual core brazos: 8.367 GOPS @ 1.6GHz

Quad Core BD: 28.695 GOPS @ 3GHz

Given Brazos has less integer execution resources, a much smaller front end etc, less cache even a half speed l2 cache....

 

[Ancalagon44]

Yeah at the moment I'm thinking these are complete crap.

We have a non standard SB clockspeed, what looks like a non standard BD speed, benchmark results that are worse than Phenom II or even Brazos (as above), and no other verification.

ALL of the benchmarks that we have seen about BD to date, I believe have been complete crap.

EDIT: Also note that there is no link to the original cached page, so we cant verify it ourselves. Plus, it wouldnt be SiSoft themselves doing the leaking, it would be whoever uploaded the results to SiSoft.

 

[psolord]

JF-AMD was perfectly clear that all benchmarks we will see until the launch will be fakes. The guy is in the know, so to say that publicly, he knows something important that we don't. Why do you guys get so negatively excited?

AMD is probably about to pull another rv770 only better.

 

[Ancalagon44]

JF-AMD was perfectly clear that all benchmarks we will see until the launch will be fakes. The guy is in the know, so to say that publicly, he knows something important that we don't. Why do you guys get so negatively excited?

AMD is probably about to pull another rv770 only better.

To be fair, he would rather that we believe they are fake, even if they were real. If we believe that they are fake, then they wont affect our purchasing decision one way or another.

Its too late for it to be another RV770 - that was launched on time if I recall correctly.

 

[frostedflakes]

JF-AMD was perfectly clear that all benchmarks we will see until the launch will be fakes. The guy is in the know, so to say that publicly, he knows something important that we don't. Why do you guys get so negatively excited?

AMD is probably about to pull another rv770 only better.

That's nonsense. JF-AMD has said benchmarks using engineering samples and pre-release silicon will not necessarily be indicative of performance of final release products. That's just common sense, unfinished products will tend to have lower performance. That doesn't mean benchmarks from pre-release products are fake, though, just that you shouldn't make assumptions about finished products based on them. That was the point I think he was trying to get across. Calling benches based on pre-release hardware "fake" because they may not represent performance of shipping products is just PR spin IMO, though. And I'm not necessarily making any claims as to the veracity of these benches, they may be 100% made up, but the idea that all leaks before the official launch must be fake is just ridiculous. Finished parts start shipping out weeks before launch and sometimes NDA is broken with these parts that will end up in consumer's hands very soon and legitimate leaks occur.

 

[CHADBOGA]

Just picked up an Asus Sabertooth. Rapidly looking like a mistake

Why do people buy the motherboard before the CPU performance is definitively known?

Did you get duped by the hype the AMDroids are always furiously spinning?

 

[psolord]

To be fair, he would rather that we believe they are fake, even if they were real. If we believe that they are fake, then they wont affect our purchasing decision one way or another.

Its too late for it to be another RV770 - that was launched on time if I recall correctly.

There's no point into claiming fakeness, if the benchmarks were valid. What for? So to make people wait another month and the go to Intel anyway? It's one thing getting the better product and a quite different thing being also angry to the rival company for making you wait. JF-AMD wouldn't want that, so he was telling the truth. The guys above (brazos example) already showed how they can't be true anyway.

As for the rv770, I meant its surprising performance. Yes it was not delayed.

 

[psolord]

That's nonsense. JF-AMD has said benchmarks using engineering samples and pre-release silicon will not necessarily be indicative of performance of final release products. That's just common sense, unfinished products will tend to have lower performance. That doesn't mean benchmarks from pre-release products are fake, though, just that you shouldn't make assumptions about finished products based on them. That was the point I think he was trying to get across. Calling benches based on pre-release hardware "fake" because they may not represent performance of shipping products is just PR spin IMO, though. And I'm not necessarily making any claims as to the veracity of these benches, they may be 100% made up, but the idea that all leaks before the official launch must be fake is just ridiculous. Finished parts start shipping out weeks before launch and sometimes NDA is broken with these parts that will end up in consumer's hands very soon and legitimate leaks occur.

I agree, but this is the case here. Pre release hardware, ergo fake (or not representative if you prefer) benchmarks.

When we'll hear that the retail versions of Zambesi are shipping and we get some benchmark after that, then yes they may be true. At this point in time though, it's more sensible to debunk them, than to accept them, especially since we received a very targeted warning beforehand.

Again, AMD does not gain anything if they spin things. Delaying everyone's upgrades, if the product plain sucks, will only make people not only go to their rivals, but also getting angry at them as well.

In any case JF-AMD does not speak for AMD in these forums. I take his postings as advice only. I hope the rest do the same and not misunderstand it, by getting angry in the first place.

 

[Arkadrel]

Sorry if this has been posted in the other BD benchmark thread....I gave up on that thread. So if this is a re-post, please remove.

http://www.brightsideofnews.com/new...-zambezi-scores-worse-than-intel-core-i7.aspx

 

[BlueBlazer]

Seriously guys.... Brazos is faster in those benchmarks then BD!!!

dual core brazos: 8.367 GOPS @ 1.6GHz

Quad Core BD: 28.695 GOPS @ 3GHz

Given Brazos has less integer execution resources, a much smaller front end etc, less cache even a half speed l2 cache....

If you consider that Bulldozer "quad core" as a dual core (with 4 threads), then its a different story. Quote from chew* sometime ago....

Actually it's not. For whatever reason amd chose to market it as cores.

In reality however BD is more akin to 4 cores / 8 threads as they share resources.

I have said this since the beginning.

This is similar to my conclusions in another post :hmm:

 

[inf64]

If you consider that Bulldozer "quad core" as a dual core (with 4 threads), then its a different story. Quote from chew* sometime ago....This is similar to my conclusions in another post :hmm:

QC Bulldozer actually has 4 cores,as opposed to dual core Brazos. Who cares if the front end is shared?You still have 4 actual integer cores in the device. And no,there is not a way each integer Bulldozer core with all the improvements over F10h can end up being slower than each Brazos core.Not only you have much beefier units,you have L2 cache that's on another level compared to L2 of Bobcat(4x the size and double the frequency),you have prefetching and branch prediction that is on another level,unified scheduler for 4 execution units etc. For this thing to end up slower than Bobcat at the same clock,the very uarchitecure has to be fundamentally broken in some way.Which won't happen. I won't even mention FP resources since it's not even funny how big the difference is.

 

[Dresdenboy]

I'd like to hear the points of those who think that these ES results with cores/caches/etc performing worse than in 10h or Llano are representative of systems (note: not stepping) going on sale later this year.

And to repeat (since these Sandra results have been discussed multiple times already):

2. Look at those memory latency results, up to 73% higher latency using the same memory as Sandy Bridge.
If true, presumably this is a bios or stepping issue and once it is fixed overall performance should increase significantly.

2. Look at those memory latency results, up to 73% higher latency using the same memory as Sandy Bridge.
If true, presumably this is a bios or stepping issue and once it is fixed overall performance should increase significantly.

http://www.realworldtech.com/page.cf...2610181333&p=9

If prefetching isn't even enabled for linear data access in these ES....

both: http://www.semiaccurate.com/forums/showpost.php?p=132211&postcount=788

@riek:
Unfortunately there are too many possible influences on these ES results. But one thing I like about scores out of a Sandra result DB is that they are likely to be real. OTOH this doesn't tell us anything about the performance to be expected if anything is working and switched on as expected on these chips (e.g. the prefetchers as Hans de Vries pointed out were disabled in that leaked Sandra article).

Without prefetching performance might be around 40-90% for many workloads. Someone looked at this effect on an Opteron processor for his master's thesis.

More SiSoft result links can be found in my blog.

 

[sm625]

I dont know why someone would even bother to post all that crap when it is obviously slower than a phenom X6. Why not compare directly to phenom? Why even muddy the waters by putting intel in there? If it cant even beat previous generation amd, then it is doa. So these guys need to stop with the intel clusterFUD.