AMD FX-8120P benchmark from Coolaler

[lol123]

The BD is not just a CPU. Isn't it a given that BD would be slower than SB since AMD dedicated a lot more die area for graphics, unlike SB?

Bulldozer does not have an IGP.

 

[pantsaregood]

If these are real, then this is actually worse than Phenom was. The K10 architecture was ultimately competitive with Conroe and Penryn, the original Phenom just had a pretty significant bug and failed to clock well. It was, however, an improvement over K8.

Bulldozer looks to actually be significantly slower than K10 if a six-core K10 can outperform an eight-core BD. At the very least, K10 matched K8 with equal core count and clock speed.

This is absurd.

 

[BD231]

You can have all the collective brilliance of Nobel laureates in physics, chemistry and mathematics at your command, but if you suck at leadership and vision then they will surely flounder under your misguided directives all the same.

What made MS different from IBM? Or Intel different from AMD in the late 1970's and early 1980's? What made Nvidia different from Matrox?

Leadership.

And in Jan of this year, for some very good reason that's never been made all that clear to its shareholders, AMD's Board of Directors elected to dismiss the leadership that gave rise to Bulldozer.

lol hogwash, a bad idea is a bad idea. You can't blame business folk for being convinced accredited engineers are leading them in the wrong direction.

 

[BlueBlazer]

Bring out AMD's Secret Weapon - Laser Cats!

http://www.hulu.com/watch/121061/saturday-night-live-snl-digital-short-laser-cats-5#s-p1-sr-i2

Off topic, that's hillarious!

 

[GaiaHunter]

Some dumb ass engineer convinced AMD sticking with the same number of FPU's and increasing the integer workload would do them any good. JFAMD stated way back that AMD believed 90% of desktop user workloads were integer based.

No - the FPU unit is much more powerful on paper than 2 FPU units in Phenom II.

The problem is probably somewhere else - rumors point to cache performing as crap.

 

[choliscott]

Can I hope that the person who produced these results is really the president of the Intel Fanboy club & the real results are going to be a lot better then this?

 

[BlueBlazer]

Can I hope that the person who produced these results is really the president of the Intel Fanboy club & the real results are going to be a lot better then this?

Coolaler is one of a few very well known for leaking engineering samples (including the recent Sandy Bridge-E ones, before Toms Hardware article), possibly has insider/close ties with motherboard manufacturers in Taiwan.

 

[Eeqmcsq]

Since we don't yet know how well that 2nd core performs, I see two general possibilities. First, that 2nd core performs extremely poorly, adding very little multithreaded performance. This means that Bulldozer is similar to Nehalem with hyperthreading, that FX-8120 DID improve its IPC over Phenom II, that it's acting more like a 4 discrete core processor, that it SHOULDN'T beat Thuban's 6 discrete cores, and AMD is overmarketing the 2nd core as a "core". Mu gut tells me this isn't the case, though.

Second, that 2nd core performs somewhat like a discrete core, let's say at least 50%, which means the FX-8120 functions at least iike 6 discrete cores, and that it has worse IPC than Phenom II, so the starting price of $220 is a major facepalm next to the 1100T.

AMD's only hope is that their 2nd gen Bulldozer cleans up whatever's wrong with this 1st gen Bulldozer, similar to Phenom II's improvement over Phenom I. Rory Read has a tough task ahead of him.

 

[MegaWorks]

Who's the engineer in charge if this abomination?

I really hope these result are fake because if you can't beat your competitor there's only one thing left to do.

 

[Vesku]

Who's the engineer in charge if this abomination?

I really hope these result are fake because if you can't beat your competitor there's only one thing left to do.

No shame in not beating your competitor if they are 5+ times bigger than you. There would be shame in not beating your own 2 year old+ product though. Just as Intel received deserved enthusiast scorn for first P4s that were not beating top end P3s.

 

[Idontcare]

lol hogwash, a bad idea is a bad idea. You can't blame business folk for being convinced accredited engineers are leading them in the wrong direction.

You do know Dirk's background, right? What he did before going to AMD, what he did when he got to AMD, and why he was promoted up the chain to eventually become the CEO, yes?

 

[pantsaregood]

Since we don't yet know how well that 2nd core performs, I see two general possibilities. First, that 2nd core performs extremely poorly, adding very little multithreaded performance. This means that Bulldozer is similar to Nehalem with hyperthreading, that FX-8120 DID improve its IPC over Phenom II, that it's acting more like a 4 discrete core processor, that it SHOULDN'T beat Thuban's 6 discrete cores, and AMD is overmarketing the 2nd core as a "core". Mu gut tells me this isn't the case, though.

Second, that 2nd core performs at least 50% of a discrete core, let's say at least 50%, which means the FX-8120 functions at least iike 6 discrete cores, and that it has worse IPC than Phenom II, so the starting price of $220 is a major facepalm next to the 1100T.

AMD's only hope is that their 2nd gen Bulldozer cleans up whatever's wrong with this 1st gen Bulldozer, similar to Phenom II's improvement over Phenom I. Rory Read has a tough task ahead of him.

It isn't something that will be easy to clean up.

Phenom and Phenom II are very much the same. Phenom II overcame the clocking issues that the original Phenom faced and slapped on an extra 4 MB of L3 cache. Benches that don't take advantage of L3 cache generally perform identically (or near identically) on Phenom and Phenom II models of equal clock speeds.

Phenom II ultimately clocked around 40% higher than Phenom. Phenoms generally topped out around 3.0 GHz (my 9750 actually wouldn't overclock beyond 2.8 GHz), while higher-end Phenom IIs are pushing 4.2 GHz. At stock speeds, you're looking at 2.6 GHz vs. 3.7 GHz, which is a bit over 40%.

Boosting BD's clocks by a similar amount would land it in the 5.5 GHz range, which I simply don't see happening. If something is seriously screwed up with the cache, maybe they can clean it up. Otherwise, the architecture in general may be a flop.

 

[RussianSensation]

I wonder if these benchmarks are only with 1 module enabled, just like their extreme overclocking event days ago?

Something is off here. There is no way they'll be able to sell higher end FX processors for $225-270 with this performance.

 

[drizek]

So it's pretty obvious that these are real. The question is how gimped are the CPUs that were released to the benchmarkers, and how much has AMD been able to fix since the delay?

 

[drizek]

AMD's only hope is that their 2nd gen Bulldozer cleans up whatever's wrong with this 1st gen Bulldozer, similar to Phenom II's improvement over Phenom I. Rory Read has a tough task ahead of him.

Phenom II was a die shrink. Bulldozer isn't getting shrunk any time soon. 32nm is it for the next 2 years, and that's being optimistic.

 

[dac7nco]

You do know Dirk's background, right? What he did before going to AMD, what he did when he got to AMD, and why he was promoted up the chain to eventually become the CEO, yes?

No, at least I don't.

Daimon

 

[drizek]

for some very good reason that's never been made all that clear to its shareholders, AMD's Board of Directors elected to dismiss the leadership that gave rise to Bulldozer.

That good reason could simply have been that they wanted somebody to blame and take the fall for it.

Let's say bad bulldozer performance is nobody's fault in particular. It would still be better for the board to point at the CEO and blame him than it would be to just put their hands in the air and say "we did our best, but it turns out we just can't keep up with Intel".

And no, I don't know anything about Dirks history either.

 

[sm625]

The ignorance just amazes me. All AMD needed to do is take 2 llano dies, strip out the GPU, and tie together what is left, and they would have an 8 core chip that is faster AND smaller than this BD. They would have known this for at least 6 months, but they did not do what I just suggested (as far as we know). The only logical conclusion is that BD has to be much faster than what is being shown here. Obviously. Billion dollar companies do not shovel out garbage when it would be much cheaper and faster ttm for them to produce a cookie cutter 8 core llano. You people who cant understand that need to get over yourselves.

 

[Accord99]

Since we don't yet know how well that 2nd core performs, I see two general possibilities.

Since the two cores are equal in capabilities, they should be identical in performance. With these results, the problem would seem to be that the BD cores have gotten much weaker, weaker than a single Thuban core and possibly even weaker than half a SB core while Hyperthreading.

So at a high level view, with 8 threads BD and HT-enabled SB are pretty comparable. Each thread gets a narrow core (of roughly comparable power) to run. However, with fewer threads SB can combine both narrow cores to form a very powerful core to run a single thread. However, for BD even with one thread, it's still using only the same narrow core with a modest boost from extra cache and higher clock speeds.

AMD's only hope is that their 2nd gen Bulldozer cleans up whatever's wrong with this 1st gen Bulldozer, similar to Phenom II's improvement over Phenom I. Rory Read has a tough task ahead of him.

I think the only real way to improve BD is to improve the performance of each "core" or massively increase the clock speed.

 

[Mem]

So it's pretty obvious that these are real. The question is how gimped are the CPUs that were released to the benchmarkers, and how much has AMD been able to fix since the delay?

I'm guessing AMD won't be able to do much until they release BD MK2 version down the road,I'm sure they know where they can get more performance out of BD and also the main bottlenecks etc...however how much time and resources will it take for a redesign and improved version for next gen?


I'm hoping official benchmarks are better but don't have high hopes at the moment.

 

[Ancalagon44]

No - the FPU unit is much more powerful on paper than 2 FPU units in Phenom II.

The problem is probably somewhere else - rumors point to cache performing as crap.

No its half as powerful. One BD module now shares what a Phenom II core had to itself.

See diagram on this page

 

[Edrick]

Everyone is asking how AMD could release this CPU if the performance is this bad. The answer is quite simple, they spent lots of $$ in development and they need to get it back somehow. I am sure they did not plan on this and have bugs to work out. There is no question that the second gen BD will be much improved.

Remember, Intel released Itanium back in 2001? when it was already years late and performance was well behind their competitors at the time. They had no choice, they invested billions into it and just throwing it away was not an option. AMD can not just trash BD at this late in the game.

Then again, these benchmarks may just be fooling us.

 

[BlueBlazer]

So it's pretty obvious that these are real. The question is how gimped are the CPUs that were released to the benchmarkers, and how much has AMD been able to fix since the delay?

These should be the latest B2.F steppings, though the final shipping one should be B2.G stepping. How much they have fixed is unknown (possibly memory and clock issues comes to mind), it does seem there's a few revisions since B0 stepping. Every silicon re-spin takes time, thus the inevitable delays. :hmm:

No, at least I don't.

Daimon

Dirk Meyer was an engineer/architect with DEC on the Alpha CPU team. After Intel bought out DEC, he moved to AMD and became the head of the Athlon CPU team.

 

[Elixer]

Alright AMD, the joke is over. Where is the REAL Bulldozer chip?

It is fairly obvious that this is a server chip trying to be a desktop part, when it is not.
This also falls in line with what JFAMD has been saying.
Cray and all the other big players LOVE the chip, and that is by no accident.

I suppose the next refresh of this CPU is when the 'real' desktop part will appear.

In order to make the desktop crowd a bit more happy, it seems you can o/c this chip pretty nicely, and you will need that o/c to beat the current generation of their CPUs.

 

[Natfly]

I wonder if these benchmarks are only with 1 module enabled, just like their extreme overclocking event days ago?

Something is off here. There is no way they'll be able to sell higher end FX processors for $225-270 with this performance.

I heard it's using only 1 thread on the 1 module with L2, L3, and prefetch disabled and using only half a memory channel.