The Bulldozer Review: AMD FX-8150 Tested
by Anand Lal Shimpi on October 12, 2011 1:27 AM ESTThe Pursuit of Clock Speed
Thus far I have pointed out that a number of resources in Bulldozer have gone down in number compared to their abundance in AMD's Phenom II architecture. Many of these tradeoffs were made in order to keep die size in check while adding new features (e.g. wider front end, larger queues/data structures, new instruction support). Everywhere from the Bulldozer front-end through the execution clusters, AMD's opportunity to increase performance depends on both efficiency and clock speed. Bulldozer has to make better use of its resources than Phenom II as well as run at higher frequencies to outperform its predecessor. As a result, a major target for Bulldozer was to be able to scale to higher clock speeds.
AMD's architects called this pursuit a low gate count per pipeline stage design. By reducing the number of gates per pipeline stage, you reduce the time spent in each stage and can increase the overall frequency of the processor. If this sounds familiar, it's because Intel used similar logic in the creation of the Pentium 4.
Where Bulldozer is different is AMD insists the design didn't aggressively pursue frequency like the P4, but rather aggressively pursued gate count reduction per stage. According to AMD, the former results in power problems while the latter is more manageable.
AMD's target for Bulldozer was a 30% higher frequency than the previous generation architecture. Unfortunately that's a fairly vague statement and I couldn't get AMD to commit to anything more pronounced, but if we look at the top-end Phenom II X6 at 3.3GHz a 30% increase in frequency would put Bulldozer at 4.3GHz.
Unfortunately 4.3GHz isn't what the top-end AMD FX CPU ships at. The best we'll get at launch is 3.6GHz, a meager 9% increase over the outgoing architecture. Turbo Core does get AMD close to those initial frequency targets, however the turbo frequencies are only typically seen for very short periods of time.
As you may remember from the Pentium 4 days, a significantly deeper pipeline can bring with it significant penalties. We have two prior examples of architectures that increased pipeline length over their predecessors: Willamette and Prescott.
Willamette doubled the pipeline length of the P6 and it was due to make up for it by the corresponding increase in clock frequency. If you do less per clock cycle, you need to throw more clock cycles at the problem to have a neutral impact on performance. Although Willamette ran at higher clock speeds than the outgoing P6 architecture, the increase in frequency was gated by process technology. It wasn't until Northwood arrived that Intel could hit the clock speeds required to truly put distance between its newest and older architectures.
Prescott lengthened the pipeline once more, this time quite significantly. Much to our surprise however, thanks to a lot of clever work on the architecture side Intel was able to keep average instructions executed per clock constant while increasing the length of the pipe. This enabled Prescott to hit higher frequencies and deliver more performance at the same time, without starting at an inherent disadvantage. Where Prescott did fall short however was in the power consumption department. Running at extremely high frequencies required very high voltages and as a result, power consumption skyrocketed.
AMD's goal with Bulldozer was to have IPC remain constant compared to its predecessor, while increasing frequency, similar to Prescott. If IPC can remain constant, any frequency increases will translate into performance advantages. AMD attempted to do this through a wider front end, larger data structures within the chip and a wider execution path through each core. In many senses it succeeded, however single threaded performance still took a hit compared to Phenom II:
At the same clock speed, Phenom II is almost 7% faster per core than Bulldozer according to our Cinebench results. This takes into account all of the aforementioned IPC improvements. Despite AMD's efforts, IPC went down.
A slight reduction in IPC however is easily made up for by an increase in operating frequency. Unfortunately, it doesn't appear that AMD was able to hit the clock targets it needed for Bulldozer this time around.
We've recently reported on Global Foundries' issues with 32nm yields. I can't help but wonder if the same type of issues that are impacting Llano today are also holding Bulldozer back.
Facebook
Twitter
RSS
430 Comments
View All Comments
vtohthree - Wednesday, October 12, 2011 - link
..but because Intel doesn't have to try hard to compete, here we were sitting as consumers waiting for a proper response from AMD so that Intel would be on their toes to unleash more potential in their chips or lower prices, but this is sort of sad...ET - Wednesday, October 12, 2011 - link
I recently bought a Phenom II X6 1090T to upgrade an X3 710. Looks like a better deal now. BD's power draw in particular is disappointing. No doubt Intel is what I'd recommend to others. I've have been using AMD CPU's for several years now (Athlon XP 2100+ was my first), and I still like AMD, but I'm disappointed by BD, especially after the long wait.Loki726 - Wednesday, October 12, 2011 - link
Thanks Anand for the compiler benchmarks.It seems like performance on bulldozer is highly application dependent, better at data-parallel and worse (even than Phenom) on irregular and sequential applications.
I'll probably skip this one.
I don't mind this tradeoff, but the problem is that AMD already has a good data-parallel architecture (their GPU). I'n my opinion they are moving their CPU in the wrong direction. No one wants an x86 throughput processor. They shouldn't be moving the CPU towards the GPU architecture.
AMD: Don't pay the OOO/complex decoder penalty for all of your cores. If your app is running on multiple cores, it is obviously parallel. Don't add hardware that wastes power trying to rediscover it. Then, throw all your power budget at a single or a few x86 cores.
Beat Intel at single threaded perf and then use your GPU for gaming, and throughput processing (video, encryption, etc).
I'm not a fan of Intel, but they got this right. If they get over their x86 obsession and get their data-parallel act together they are going to completely dominate the desktop and high end.
dubyadubya - Wednesday, October 12, 2011 - link
Care to share which tests are 64 bit? Each bench program used must specify if its 32 or 64 bit. Why do all review sites forget to includet this critical info? From the limited results I can find on the net AMD see's a large performance increase running 64 bit code over 32 bit code while Intel see's little if any increase.HangFire - Wednesday, October 12, 2011 - link
I've got an Asus board that promises to support BD, and I've holding off upgrading my unlocked/overclocked 550BE for literally months, and for this? I might as well just get a Phenom II quad or 6-core.I've said all along that AMD needs to address their clock versus instruction efficiency to be competitive. To do that they need to redesign their cores and stop dragging along their old K8 cores.
So here we are with Bulldozer, Wider front end, TurboCore now works, floating point decoupled, 8 (int) cores, and... still flogging the same instruction efficiency as the old K8 cores (at least, the integer portion of them).
Oh, yeah, I'm sure at the right price point some server farms will be happy with them, and priced low enough, they can hold on to the value portion of the marketplace. To do both they'll have to compete aggressively on price, and be prepared to lose money, both of which they seem to be good at.
Like Anand said, we need to see someone actually compete with Intel, but it appears that AMD has lost the ability to invent new processor cores, it can only manipulate existing designs. Instead of upgrading the CPU, it looks like I'll go for a full Intel upgrade, unless I can find an 1100T real cheap. Hmm, that's probably a real possibility. I'm sure a lot of AMD fans are going to be trading them in now that they see what their AM3 upgrade path is(n't).
alpha754293 - Wednesday, October 12, 2011 - link
I think that you should clarify the difference between what you call "server" workloads (i.e. OLTP/virtualization vs. HPC).I suspect that with one shared FP between two ALUs; HPC performance is going to suffer.
The somewhat-computationally intensive benchmarks that you've shown today already gives an early indication that the Bulldozer-based Opterons are going to suffer greatly with HPC benchmarks.
On that note: I would like to see you guys run the standard NCAC LS-DYNA benchmarks or the Fluent benchmarks if you can get them. They'll give me a really good idea as to how server processors perform in HPC applications (besides the ones that you guys have been running already). Johan has my email, so feel free to ask if you've got questions about them.
Ananke - Wednesday, October 12, 2011 - link
Bulldozer reminds me of Sun's (Oracle) Niagara architecture. It seems AMD aimed the server and professional market. It makes business sense. The profit margins there are net 50-60% (this is AFTER the marketing, support, etc overhead costs) and along with the high performance work stations is the only growing market as of now. Hence, the stock market lifted the stocks of AMD. Gaming and enthusiast market is around 0.7% of CPU revenue - yep, that is, I work with this kind of statistics data, guys.This is a promising architecture (despite the fact that is not good for home enthusiasts). AMD should focus on providing more I/O lanes through the CPU - aka PCI lanes on cheaper boards without requirement of additional chips. It will allow placing more GPUs using overall cheaper infrastructure - exactly the way HPC and server market is evolving. Then, they should really get a good software team and make/support/promote SDK for general GPU computing in line of what NVidia did with CUDA.
For anything mainstream / aka Best Buy, Walmart, etc./ Llano is good enough.
As I said, this Bulldozer chip apparently is not good for enthusiasts, and Anandtech is an enthusiast site, but unfortunately this is just a very small niche market. People should not bash a product, because it doesn't fit only their needs. It is OK for the vast market.
GatorLord - Wednesday, October 12, 2011 - link
Thanks for the VERY interesting stats. I had a hunch it made good sense, but since I don't work with these data it was just a hunch in the end. Now it's better...maybe a hunch plus. We should feel lucky that they even pay any attention to this segment...I suspect they do b/c a lot of decision influencers are also computer racers at home.alpha754293 - Thursday, October 13, 2011 - link
One thing that I will also say/add is that while people are perhaps grossly disappointed with the results; think about it this way:What you've really got is a quad-core (I don't count their ALUs as cores, just FPUs) processor doing a 6-core job.
So if they went to a 6-module chip, the benefits can actually be substantial.
And on the 8-module server processor, it can be bigger even still.
And yes, this is very much like the UltraSPARC T-series (which, originally was designed by UltraDense as a network switching chip), but even they eventually added a FPU per core, rather than just one FPU per chip.
The downside to the 8-module chip is a) it's going to be massive, and b) it won't be at the clock speeds it NEEDs to be to compete.
Icehawk - Wednesday, October 12, 2011 - link
I quickly ran the Rage vt_benchmark and got ~.64 @ 1thread and .25 for 2-6 threads which is what your Intel #s line up with - BUT I'm running a Q6600, 4gb, and a GTS 250... shouldn't I see much worse scores compared to a i7/current get video card? Is this something to do with Rage's *awesome* textures or?