[OFFICIAL] Bulldozer Reviews Thread - AnandTech Review Posted

[RussianSensation]

 

[dma0991]

The good news is AMD has a very aggressive roadmap ahead of itself; here's hoping it will be able to execute against it. We all need AMD to succeed. We've seen what happens without a strong AMD as a competitor. We get processors that are artificially limited and severe restrictions on overclocking, particularly at the value end of the segment. We're denied choice simply because there's no other alternative. I don't believe Bulldozer is a strong enough alternative to force Intel back into an ultra competitive mode, but we absolutely need it to be that. I have faith that AMD can pull it off, but there's still a lot of progress that needs to be made. AMD can't simply rely on its GPU architecture superiority to sell APUs; it needs to ramp on the x86 side as well—more specifically, AMD needs better single threaded performance. Bulldozer didn't deliver that, and I'm worried that Piledriver alone won't be enough. But if AMD can stick to a yearly cadence and execute well with each iteration, there's hope. It's no longer a question of whether AMD will return to the days of the Athlon 64, it simply must. Otherwise you can kiss choice goodbye.

I like Anand's final words on BD. :'(

 

[pman6]

Wasn't everyone hating on Donanimhaber and OBR over the last few weeks?

I take it those people will take their words back... eh?

 

[Tikerz]

I am ashamed to display my sig on every one of my posts here at AnandTech. I'm going to take it down now and crawl into a hole. :'(

 

[Deanodarlo]

What I can't work out is WHY Windows 7 scheduler isn't working correctly for AMD's module design, thus wasting performance. For example, currently two threads share resources on one module, instead of using one module each.

The WHY comes from the fact that surely in 5 YEARS of design, AMD would have made sure a patch was ready for the current Windows version to communicate correctly with their new CPU on release. It appears AMD still haven't learnt that working with software developers is just as important as churning out new hardware.

Having said that, using two modules for two threads would probably use even more power under load! I would just be interested to see if it helped push it up 10% or so.

The biggest laugh however is Bulldozer was conceived at a time when P4/Pentium-D was made to look like a bad design - low IPC resulting in high GHz and high power usage to be competitive. Intel switched to high IPC and lower power usage, the same thing AMD targeted for years.

So why the f**k did AMD decide to go for low IPC and high GHz pushing the watts to insane levels just to get some decent performance out of the thing? Not only that, but it costs them more to produce as the die size is much bigger than the better performing Sandybridge, WHICH HAS INTEGRATED GRAPHICS! Honestly, it beggars belief. Don't forget these engineers are clever guys, but it seems they COULDN'T SEE THE FOREST FOR THE TREES.

I hope for AMD this architecture does indeed come into its own in the future as they tweak it, because I would love to eat my words and say they did indeed see the forest. For now though, it's Intel for desktop use.

 

[theevilsharpie]

So why the f**k did AMD decide to go for low IPC and high GHz pushing the watts to insane levels just to get some decent performance out of the thing?

My guess is that AMD's primary motivation for Bulldozer was to cram in as as many cores onto a single die as they possible could, and the lowered IPC was simply a side effect of the design choices they had to make to achieve that goal and maintain a reasonable die size.

Bulldozer's power utilization and thermals, and the resulting stock clock speeds, may simply be a design defect in the chip or a problem with GloFo's manufacturing process, so I'll reserve judgement on the architecture as a whole for now.

 

[LOL_Wut_Axel]

JF-AMD will now get a lot of crap for saying IPC increased. IPC according to Anandtech is actually ~7% lower, and even with its extremely high clock speeds the FX-8150 is outpaced noticeably in most multi-threaded applications by the Core i7-2600K. Given the 16% higher clock speed than the FX-8120, we should see the Core i5-2500K being comparable to the FX-8120 in multi-threaded while giving you a whooping ~47% higher IPC, translating to about 40-45% better single-threaded performance. What this boils down to is that Bulldozer is comparable to Sandy Bridge in video encoding, 3D rendering, content creation, and file compression/decompression. It's significantly slower in gaming, audio encoding, compiling (with Visual Studio), and Photoshop.

Overclocking wasn't what we were told at the CPU frequency record breaking event by AMD and in fact it's no better than Sandy Bridge, a testament that Bulldozer's much deeper pipeline approach didn't work in its favor in frequency scaling.

Then we come to the efficiency, measured in performance/watt. It's typically faster than the Phenom II X6 1100T in multi-threaded workloads, but it also consumes more power to get there. The design decisions AMD took, going with a CMT architecture, didn't pay off what it was supposed to in power efficiency. Even more interesting is that going down a full process node to 32nm didn't help matters.

In the end, consumers should skip Bulldozer altogether, including people that already have an AMD Phenom II X6 system. As a matter of fact, that CPU is better than the FX Six-Cores, and the FX Eight-Cores are not enough of an improvement to be worth it. For people making new systems Sandy Bridge should be a clear and easy choice given the Core i5-2500K will be comparable to the FX-8120 in multi-threaded and 45% better in single-threaded while only costing $15 more, money you'll probably recoup anyway given the 2500K's much lower power consumption. Platform costs are comparable for both, as is overclocking. AMD really has nothing compelling for enthusiasts with Bulldozer.

 

[BFG10K]

This chip is a failure in desktop space. It can’t even convincingly beat the top X4 and X6 models. They should’ve just die-shrunk the X6 and released it with higher clocks/cache.

 

[jmarti445]

I am ashamed to display my sig on every one of my posts here at AnandTech. I'm going to take it down now and crawl into a hole. :'(

Umm, Its not like you are Repping a Pentium D 840 at 3.8 Ghz. Seriously. Bulldozer isn't a great processor but to compare it to the worst processors ever isn't really fair. It isn't like it is a K6, nor is it like a Cyrix 6x86 processor. This processor would have worked if one thing hadn't occurred, if the Thuban was never released.

Honestly, though I'm not sure why AMD didn't just add the SSSE3, SSE4, and AVX instructions to the Thuban microprocessor like Intel does with their microprocessors and do a dieshrink. Heck they could have added 2 extra cores to a Thuban processor and had a much more powerful design. AMD got the bulldozer just plain wrong. Most accurate release I could compare it to isn't in the CPU world but in the GPU world with the HD 2900XT. Its Hot and not as fast the competing product. Recommendation to AMD is just shink the Thuban core to .32nm and give it an extra two cores at least it would be able to trade blows on a threaded level. Bulldozer is one step forward and two steps back.

 

[cebalrai]

What I can't work out is WHY Windows 7 scheduler isn't working correctly for AMD's module design, thus wasting performance. For example, currently two threads share resources on one module, instead of using one module each.

The WHY comes from the fact that surely in 5 YEARS of design, AMD would have made sure a patch was ready for the current Windows version to communicate correctly with their new CPU on release. It appears AMD still haven't learnt that working with software developers is just as important as churning out new hardware.

Having said that, using two modules for two threads would probably use even more power under load! I would just be interested to see if it helped push it up 10% or so.

The biggest laugh however is Bulldozer was conceived at a time when P4/Pentium-D was made to look like a bad design - low IPC resulting in high GHz and high power usage to be competitive. Intel switched to high IPC and lower power usage, the same thing AMD targeted for years.

So why the f**k did AMD decide to go for low IPC and high GHz pushing the watts to insane levels just to get some decent performance out of the thing? Not only that, but it costs them more to produce as the die size is much bigger than the better performing Sandybridge, WHICH HAS INTEGRATED GRAPHICS! Honestly, it beggars belief. Don't forget these engineers are clever guys, but it seems they COULDN'T SEE THE FOREST FOR THE TREES.

I hope for AMD this architecture does indeed come into its own in the future as they tweak it, because I would love to eat my words and say they did indeed see the forest. For now though, it's Intel for desktop use.

Well, AMD is a $6 billion company and Intel is a $43 billion company. Larger companies can devote more resources to their products. Still, AMD made a processor with a new design that competes with a company eight times their size in heavily threaded tasks. Lightly threaded tasks aren't so hot on BD, but what did you all expect? Did anyone really think that little AMD could take on a giant just by getting lucky or something? Was a staggering difference in R&D budget not supposed to matter?

 

[theevilsharpie]

Well, AMD is a $6 billion company and Intel is a $43 billion company. Larger companies can devote more resources to their products. Still, AMD made a processor with a new design that competes with a company eight times their size in heavily threaded tasks. Lightly threaded tasks aren't so hot on BD, but what did you all expect? Did anyone really think that little AMD could take on a giant just by getting lucky or something? Was a staggering difference in R&D budget not supposed to matter?

Your argument would be more compelling if Bulldozer wasn't being outperformed by AMD's previous product.

Phenom II gets a lot of shit on these forums, but it's still a competitive product, with the 1100T trading blows with Intel's best as the thread count increases. AMD may not have the R&D budget that Intel does, but the fact that AMD released a new flagship product that is worse than their existing top-end offering in many respects tells me that their engineering resources are not being managed well.

 

[wlee15]

In the server room, AMD is squaring off against Intel's 8- and 10-core Xeons which are substantially faster in multi-threaded applications than Intel's quad-core desktop chips while still retaining a 130W TDP. AMD hasn't been competitive in the server space for a while, and I don't see that changing with Bulldozer.

I disagree 16-core Interlagos should be pretty competetive with 8-core Sandy Bridge-EP and 10-Core Westmere-EX at least until Ivy Bridge-EP and EX arrives in 2013

with the 1100T trading blows with Intel's best as the thread count increases.

Well BD can actually win.

 

[Ancalagon44]

Umm, Its not like you are Repping a Pentium D 840 at 3.8 Ghz. Seriously. Bulldozer isn't a great processor but to compare it to the worst processors ever isn't really fair. It isn't like it is a K6, nor is it like a Cyrix 6x86 processor. This processor would have worked if one thing hadn't occurred, if the Thuban was never released.

Honestly, though I'm not sure why AMD didn't just add the SSSE3, SSE4, and AVX instructions to the Thuban microprocessor like Intel does with their microprocessors and do a dieshrink. Heck they could have added 2 extra cores to a Thuban processor and had a much more powerful design. AMD got the bulldozer just plain wrong. Most accurate release I could compare it to isn't in the CPU world but in the GPU world with the HD 2900XT. Its Hot and not as fast the competing product. Recommendation to AMD is just shink the Thuban core to .32nm and give it an extra two cores at least it would be able to trade blows on a threaded level. Bulldozer is one step forward and two steps back.

When it gets beaten by the previous generation processors, yeah I'm gonna say its up there with the worst.

Where's Fumbles the AMD engineer?

 

[frostedflakes]

Really makes you wonder if it all comes down to process issues. As Anand mentioned they were shooting for 30% higher clock speeds with Bulldozer. 30% is pretty vague like he said, but if you use the fastest X6 with 3.3GHz base and 3.7GHz turbo as the baseline, that would put you at about 4.3GHz base and 4.8GHz turbo for the flagship Bulldozer. Would probably be quite competitive at those speeds, and considering those clocks are high (but attainable) for 32nm Sandy Bridge chips with a far shallower pipeline, you'd think a 32nm process from GloFo that was performing as intended could reach those clocks with BD and do so without sucking down hundreds of watts.

Would be really interesting to see how clocks scale with future steppings. For example the 8170 is supposed to be a new stepping, isn't it? Supposedly base clock will be 3.9 and turbo is 4.5. Still not 30%, but would be getting closer to that goal.

 

[Deanodarlo]

Well, AMD is a $6 billion company and Intel is a $43 billion company. Larger companies can devote more resources to their products. Still, AMD made a processor with a new design that competes with a company eight times their size in heavily threaded tasks. Lightly threaded tasks aren't so hot on BD, but what did you all expect? Did anyone really think that little AMD could take on a giant just by getting lucky or something? Was a staggering difference in R&D budget not supposed to matter?

Which is a good point, but they are selling this at a price point that is similar to Intel while costing them more to produce (larger die size). Also, they have been bigging up the FX line for a long time and should perhaps have let us know they are bowing out of the dying desktop market in terms of the performance crown. It's all about promises and expectations over the 5 years we've been waiting.

They would have been better sticking to budget CPU's and APU's for the desktop, shrinking a Phenom II to 32nm and improving IPC and performance per watt if they can't compete in most usage scenarios. That would have saved them a lot of money and since the desktop market is dying probably brought them a lot of sales for those wanting cheap upgrades.

If this makes them a nice profit in the server and niche market then great, it's achieved its goals of catching the monster Intel by going for a different design philosophy; as you said they can never out perform a company with a 2 year lead on fab and a huge design budget so had to try and attack from a different angle.

We are however talking from the standpoint of desktop CPU use for most of us, and it just isn't a good CPU design for that currently and doesn't even have a Windows 7 patch ready to make better use of its design. So I do believe the flack is warranted regardless of budget as it was pushed as a top of the line desktop CPU also, so will be compared to Intel on price/perfromance and performance/watt for general use. In that area it fails.

 

[theevilsharpie]

I disagree 16-core Interlagos should be pretty competetive with 8-core Sandy Bridge-EP and 10-Core Westmere-EX at least until Ivy Bridge-EP and EX arrives in 2013

Bulldozer is already a very large die with very high power requirements. We'll see how AMD can balance power and performance with a 16-core chip. Unless there's something wrong with Bulldozer that they intend to correct with Interlagos, I have my doubts that they will produce a chip that server admins will find acceptable.

Well BD can actually win.

So can the 1100T.

BD has two more cores than the 1100T, runs at a higher frequency, and uses more power. The fact that it's performance superiority over the 1100T isn't certain doesn't bode well for the chip's future.

 

[Dribble]

Wow, that's bad. I remember all the early enthusiasm and getting panned for saying it probably wouldn't equal SB as the leap was too big, but even a cynic like me didn't expect performance to be that bad.

Remember the 8150 is really a reviewers special - you can tell by the fact it is clocked so much higher then anything else and the fact power consumption is off the chart. The 8120 is the true top end part imo, and that's even worse.

 

[Medu]

Really makes you wonder if it all comes down to process issues. As Anand mentioned they were shooting for 30% higher clock speeds with Bulldozer. 30% is pretty vague like he said, but if you use the fastest X6 with 3.3GHz base and 3.7GHz turbo as the baseline, that would put you at about 4.3GHz base and 4.8GHz turbo for the flagship Bulldozer. Would probably be quite competitive at those speeds, and considering those clocks are high (but attainable) for 32nm Sandy Bridge chips with a far shallower pipeline, you'd think a 32nm process from GloFo that was performing as intended could reach those clocks with BD and do so without sucking down hundreds of watts.

Would be really interesting to see how clocks scale with future steppings. For example the 8170 is supposed to be a new stepping, isn't it? Supposedly base clock will be 3.9 and turbo is 4.5. Still not 30%, but would be getting closer to that goal.

It is probably a part of the problem, but not all of it. As was said above, JF-AMD has been saying that IPC would be go up and that doesn't seem to be the case. So it does seem like AMD have ran into problems on both the design and process sides which isn't a good combo!

 

[Blue Shift]

I'm surprised that none of the reviewers have manually assigned tasks to different corse and tinkered with workload balancing that way... In theory, you could assign threads that are part of the same process to the same module to save on cache, or assign each FP-intensive thread to a different module to give them each a nearly-dedicated wide FP unit.

It probably wouldn't be enough to push it past the 2600k in most anything, but still... It would be fun to see the "I can do these 5 things at once best with THIS core assignment" tests.

 

[jmarti445]

When it gets beaten by the previous generation processors, yeah I'm gonna say its up there with the worst.

Where's Fumbles the AMD engineer?

Again, I disagree, it isn't like we are talking a straight across loss vs the previous Generation. One thing everyone has to remember with this, It only has 4 floating point units, just like a Denab processor. Thuban has 6 floating point units which is why it is extremely competitive in benchmarks like Cinebench and other suites that require a lot of Floating point operations. If you look at the Integer based suites like Office and adobe, the Bulldozer is a better processor. However it doesn't help us gamers being that Floating point units are significant to things like gaming and scientific operations.

 

[wlee15]

Bulldozer is already a very large die with very high power requirements. We'll see how AMD can balance power and performance with a 16-core chip. Unless there's something wrong with Bulldozer that they intend to correct with Interlagos, I have my doubts that they will produce a chip that server admins will find acceptable.

16-core Opteron 6282 SE 140 TDP 2.6 Ghz
16-core Opteron 6276 115 TDP 2.3 Ghz

8-core Xeon E2-2687W 150 TDP 3.1 Ghz (Sandy Bridge-EP)
8-core Xeon E2-2670 115 TDP 2.6 Ghz (Sandy Bridge-EP)
10-core Xeon E7-4870 130 TDP 2.4 Ghz (Westmere-EX)
10-core Xeon E7-4820 105 TDP 2 Ghz (Westmere-EX)

Looks pretty competitive to me.

 

[theevilsharpie]

16-core Opteron 6282 SE 140 TDP 2.6 Ghz
16-core Opteron 6276 115 TDP 2.3 Ghz

8-core Xeon E2-2687W 150 TDP 3.1 Ghz (Sandy Bridge-EP)
8-core Xeon E2-2670 115 TDP 2.6 Ghz (Sandy Bridge-EP)
10-core Xeon E7-4870 130 TDP 2.4 Ghz (Westmere-EX)
10-core Xeon E7-4820 105 TDP 2 Ghz (Westmere-EX)

Looks pretty competitive to me.

Bulldozer looked competitive too, on paper.

We'll see what Interlagos brings to the table. I doubt it will outperform the Xeons.

 

[cebalrai]

Which is a good point, but they are selling this at a price point that is similar to Intel while costing them more to produce (larger die size). Also, they have been bigging up the FX line for a long time and should perhaps have let us know they are bowing out of the dying desktop market in terms of the performance crown. It's all about promises and expectations over the 5 years we've been waiting.

They would have been better sticking to budget CPU's and APU's for the desktop, shrinking a Phenom II to 32nm and improving IPC and performance per watt if they can't compete in most usage scenarios. That would have saved them a lot of money and since the desktop market is dying probably brought them a lot of sales for those wanting cheap upgrades.

If this makes them a nice profit in the sever and niche market then great, it's achieved its goals of caching the monster Intel by going for a different design philosophy; as you said they can never out perform a company with a 2 year lead on fab and a huge design budget so had to try and attack from a different angle.

We are however talking from the standpoint of desktop CPU use for most of us, and it just isn't a good CPU design for that currently and doesn't even have a Windows 7 patch ready to make better use of its design. So I do believe the flack is warranted regardless of budget as it was pushed as a top of the line desktop CPU also, so will be compared to Intel on price/perfromance and performance/watt for general use. In that area it fails.

Well... Remember BD was meant to compete with Nehalem. Design issues, delays, and setbacks plagued comparatively tiny AMD and therefore it gets compared to SB. AMD couldn't just abandon their investment when they saw this dynamic coming.

It sucks for them, but that's what happens when small companies have to compete against semi-monopolistic giants.

 

[Deanodarlo]

Well... Remember BD was meant to compete with Nehalem. Design issues, delays, and setbacks plagued comparatively tiny AMD and therefore it gets compared to SB. AMD couldn't just abandon their investment when they saw this dynamic coming.

It sucks for them, but that's what happens when small companies have to compete against semi-monopolistic giants.

Very true - I think we've all had the bump back down to earth and we realize the facts we thought were coming are now confirmed. Intel can't be beaten in the top line desktop, but the market is shifting towards portable low power machines anyway.

Who knows, it may pay off somewehere down the line as there will be a lot of optimizing to do, but as of now AMD and Intel are in very different markets as far as us desktop enthusiasts go and probably will be for a long time, perhaps forever!

 

[Ancalagon44]

Isnt it amazing that BD is 2 billion transistors and cannot compete with a 1.16 billion transistor Intel CPU?

Thats how bad it is. They should abort the whole thing right now.