AMD 'Bulldozer' FX CPU Reviews Arrive
I.M.O.G. writes "Today AMD lifted the embargo on their most recent desktop AMD FX architecture, code named Bulldozer, whose CPU frequency record Slashdot recently covered. The fruition of 6 years of AMD R&D, this new chip architecture is the most significant news out of AMD since the Phenom II made its debut. The chips are available now in all major retail outlets, and top tier hardware sites have published the first Bulldozer reviews already."
Here are reviews from a few different sites — pick your favorite: Tom's Hardware, PC Perspective, Hot Hardware, [H]ardOCP, or TechSpot. They don't agree on everything, but the consensus seems to be that the new chips aren't blowing anyone's socks off, and that they struggle to compete with Intel's comparable offerings. The architecture shows promise, but performance gains will take time to materialize, making it difficult to leapfrog Intel to any significant degree.
And the comparisons seemed pretty much benchmarked performance based, with a side of price comparison. Fair enough, as these are pitched as 'enthusiast' parts; but left me wondering about one thing:
Of late, intel's somewhat confusing set of model numbers has been distinguished, in addition to differences in speed, by various features being lasered off of certain parts, but not others, mostly virtualization-related stuff. AMD generally left those on at all times and distinguished primarily by speed.
Does anybody have an idea how the price/performance comparisons change(if in fact they do) from the pure-benchmark ones given in TFAs, if the buyer requires that all the relevant virtualization features be enabled?
buy a 6 core Phenom II, overclock it, and pray that AMD can stay around long enough to fix this mess.
Go check the techreport review and look at the price/performance chart: The 2500K has slightly higher performance, lower price, and *much* better energy efficiency.
Go look at the LKML where you'll see Linus & Ingo Molnar calling out AMD for design flaws in Bulldozer's cache that AMD wants to paper-over with kludgy software workarounds in the kernel: http://us.generation-nt.com/answer/patch-x86-amd-correct-f15h-ic-aliasing-issue-help-204200361.html
I feel bad for AMD's engineers. I *don't* feel bad for the marketing hype machine that has been relying on "geek-cred" from sites like Slashdot and the usual David vs. Goliath myth to get unearned praise. If Intel had come out with Bulldozer instead of AMD, we'd be calling this Prescott version 2.0.
AntiFA: An abbreviation for Anti First Amendment.
You have to hope that whatever sacrifice AMD made in this design was made to better enable the CPU and GPU to be fabricated on the same process in Trinity and beyond.
Sorry, but even though the reviews try not to kill the underdog this chip is huge, hot, performs crap in anything but extremely well threaded applications and even there it barely competes with Intel's 2500K/2600K. Anandtech passed 300W trying to overclock this beast. It's like AMD implemented every bad idea Intel had with the PIV.
Live today, because you never know what tomorrow brings
Anand has always done a better job than Tom IMHO
http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested
Well now I know for sure why SB-E (sandybridge-e) is not arriving until Q1 2012... Intel is just going to continue to milk SB parts for the time being. Sad because I really wanted to get an Ivy Xeon rig to replace my current dual proc mobo, but I'm not sure I can wait until 2013!
I buy AMD, mostly when I build myself and when I'm on a budget. Not because I like weak chips. I love CPU speed, but I also love to keep my wallet as full as possible. This is what AMD offers: "bang for your buck". AMD is interesting for anyone who wants to balance between spending money and reasonable performance. Want pure performance and it doesn't matter what it costs? Go Intel... No questions asked. (This wasn't so in the Athlon XP/64 days)
Also keep in mind that we are now really on a computing-power plateau. At least from the point of view of the average user. I've got a frigging Atom D525 as my primary desktop, which isn't exactly a powerhouse. It has some quirks, notably because Firefox sometimes seems to slow down (Installing the 64-bit Linux flash player made the situation better, for some odd reason... It still is not really ok though) Most of the time it works just perfectly fine. I'd wager that adding 50% more power to that chip would nullify my troubles.
I can't say much about these new Bulldozer machines, but I just ordered an AMD A6-3560 + Gigabyte motherboard + 16GB RAM for less than 250€. That's a lot of power for not much money. (Moms computer died, building a new one for her... She doesn't need much, this should be more than overkill) I'm very keen on trying it out.
Ahhh...the great dumpster continuum. Many a free computer will be found there. -- sowth (748135)
Similar to Intel Insider? In 2007 there was talk of disallowing users access to the framebuffer, did any of this ever materialise?
This is what is keeping me away from buying core i5 cpus, even if the AMD ones might be a bit slower
The architecture shows promise, but performance gains will take time to materialize, making it difficult to leapfrog Intel to any significant degree.
Hasn't that been the case for 20 years now?
The three laws of thermodynamics:(1) You can't win. (2) You can't break even. (3) You can't even quit.
As usual, I feel somewhat obligated to post up AnandTech's review, which always seems much more in-depth and polished than almost all the sites out there:
http://www.anandtech.com/show/4955/the-bulldozer-review-amd-fx8150-tested
uKernel and Virtualization folks will buy. I really don't care about performance. I care about documentation and standards compliance. My next PC will be AMD.
Intel has limited pci-e in the i3 i5 and low end i7 boards that makes USB 3.0 and other on board stuff eat in to the X16 for video.
With amd you can get a board with lot's of pci-e lanes with out the need for the high end cpu and or get a high end cpu and not need to buy a super high end MB.
In the available reviews i found no benchmark about compilation. Is it unusual these days to compile the linux kernel? The FX-8150 could show here if it indeed is capable of decent real-world multi-threading performance and i guess many slashdotters compile software on a regular basis. So please post a link to such a benchmark if you have one.
It seems this new generation of AMD will run Dwarf Fortress more slowly than Phenom II :(
When new computers are slower, there's something seriously wrong!
- First, there is the huge delay intel caused by engaging in fraud by paying pc makers to not use amd chips, right at the time amd was at an advantage.
- Then there is the fact that these synthetic benchmarks use intel's proprietary libraries, which were proven to work ineffectively when 'non genuine intel' architecture was detected.
- Then there is the fact that this is a new platform, and its just out, and the main deal with this is being easily increasable in cores. so amd will just add more cores without any research being needed. expect 32 core cpus in a year or so. 16 cores already out.
- As you can understand these cpus are geared more for server environment, and will take that environment over.
- Amd is moving to trinity in one year or so. Trinity is the APU format that all amd cpus will take from then on. Llano apus have been quite successful in gaming fro example 50-80 fps in starcraft 2 (crossfired and not) -> you dont need to buy an external card anymore, and if you do you can crossfire it with the cpu contained one. http://www.anandtech.com/show/4476/amd-a83850-review/6 http://techreport.com/articles.x/21730/8 intel is worlds behind in this one.
and then there is the ultimate question of what the fuck i am going to do if i grab a powerful processor. really. i bought an overclockable board, and an unlocked cpu. and when i played games, i found out that it was mostly the video card i added that did most of the thing. the cpu i had was way, way over any potential requirements and needs of these games. i didnt need to buy a powerful one at all.
i went about hardware/software forums asking what i could do with a powerful computer. answers have been 'video encoding', 'benchmark', 'seti'. as it seems, any daily usage for cpus are WAY behind the power of modern cpus. to utilize your cpu power at all, you need to do unorthodox, unnecessary shit, or be in a profession that works on these.
so i think all this performance talk is bullshit. there is no way in hell you will use that performance, even in hardcore gaming with an eyefinity 3 monitor setup in 5000x resolution, with 2x antialiasing and full detial. (and i just have 2x 5670 cards).
future is in the heterogeneous chips i think. llano already has been a success, and its possible to save 30% on the cost of cpu + mobo + graphics card if you go the llano way over anything intel, and gaming performance is incomparable. when trinity comes, i think there will be a big change in computing. especially when amd puts out a computing platform like cuda.
Read radical news here
If you were to sum up the Bulldozer concept with just one word, it’d have to be scalability. AMD put the bulk of its effort into designing a building block that’s small enough to be duplicated over and over in silicon, and yet capable enough to handle both integer- and floating-point-based workloads as deftly as possible. Indeed, the company confirms this was a from-scratch project started several years ago after considering some of the target markets its next-gen architecture would end up addressing: everything from mainstream clients to the top of the server range.
this.
now amd can just go adding cores/modules, all the way up to 30 and beyond. this would be unbeatable in multi threated applications.
Read radical news here
Lots of negative posts here about BD but remember, it's all the code being tested coming from compilers that have no clue about BD. In this case, the BD code is probably coming out as generic X86. To really test BD fairly, the compilers need updated to optimize for BD like the Intel processors get optimized.
Pretty much intel has won. I have a hard time believing AMD will pull this one out on the desktop market.
Intel can now charge whatever they want. The good thing is that now intel's primary competition is samsung, qualcomm, TI and apple. Low powered ARM processors will change the market and I for one am tired of space heaters. Future systems are 10W or less, not 300W or less.
Great... now we can expect Intel to rest on their Laurels and never upgrade the Core i5 2500k and Core i7 2600k series. If it weren't for AMD, we'd still be using Pentium 4's at 1.6Ghz
The point of Bulldozer and pairing it with AMD GPGPUs is to leverage all the wealth of work AMD has put into OpenCL 1.1 with OpenGL 4.x.
When more and more apps leverage OpenCL 1.1 [and the list is growing rapidly] using the likes of LLVM/Clang where AMD has worked hard at leveraging you'll begin to see a lot of these ``benchmarks'' being truly useless and tuned specifically for Intel.
The work AMD is putting in with that marriage should be obvious: http://developer.amd.com/pages/default.aspx
Until applications truly leverage the architecture in conjunction with the AMD 6000/7000 GPGPUs talking about game benchmarks is truly juvenile. I mean seriously folks. Grow up, sit back and watch this platform advance and those same games become tuned for that CPU/GPGPU marriage.
...Bulldozer is massively underwhelming. Low IPC, lower clocks than expected, deeper pipelines, absurd power consumption... Sound familiar? So much for 2 cores in one.
AMD predicts 10-15% improvement per year... This means, in the best case twice the power in... 5 years !!! Moore was more optimistic than AMD...
If Bulldozer is supposed to be big arrays of cores doing symphony of carpet-bombing every big computing problem, then why the main work of core multiplication is done by their MARKETING instead of ENGINEERING department ?
Why are they rying to sell us one module as two cores ? Furthermore, why is there so much cache taking the useful die area ?
If you say that you did some optimisations and that now you can get a module for only extra 18% die area of your K-10 core, the why the heck wouldn't you put that 32nm to a good use and pepper on 100% more MODULES than you had K-10 CORES in last, 45nm generation ?
( 45^2 / 32^2 ~2) means you can cram 2x as much logic on the same area as before, so by pure logic one would expect to see 12-moduled Bulldozer, that could do some pretty nifty tricks and in some situations double the number of actual cores, with each module performing as two cores.
And that is without stinging on the cache, giving it the same proportion of the die as on the Phenom II.
Thinning of the cache could byu you quite a few extra modules.
Yes, you would be TDP limited, so the frequency with all modues fully active would have to go down. But the aggregate CPU power would still be staggering.
And, you could heavilly turbo when using just smaller subset of modules.
This is just AMD cutting corners and expecting everyone else to pay for this as some kind of extra value...
Does Bulldozer really have to be the fastest chip ever? Or simply fast enough to run everything that you want to run at the best price/power available? While some people will insist that Too Fast is simply not possible, I am the other 99% that says too expensive isn't worth it.
"It's the height of ridiculousness to say for those 9 lines you get hundreds of millions."
This is the comparison of a8 as a standalone cpu versus 1100T. In this test, the onboard die is not used. this is not how this platform should be run. It should be run as its onboard gpu in hybrid crossfire with a 6670 external gpu :
http://www.anandtech.com/show/4476/amd-a83850-review/6
http://techreport.com/articles.x/21730/8
Read radical news here
True, you have to check in advance which motherboards support ECC. It seems that many vendors don't want the effort of testing it or whatever, so the just kill the feature. But with a bit of looking around, you can usually find a board that supports ECC. Asus for instance supports it in many (but not all) AM3 boards.
My own current upgrade (parts bought, but not installed yet) consists of an Asus board and a Phenom II X4 because of ECC. Call me paranoid if you like, but the error protection seemed more important than the extra performance of a Sandy Bridge quad core.
C - the footgun of programming languages
Ignore AMD vs Intel for a second. There are several single and threaded benchmarks showing the Phenom II x6 3.4ghz out performing the BD 3.6ghz.... WTF?! Even the older AMD chips use less power and are faster.
Intel spanked the BD in almost every benchmark, both single and multi-threaded.
Intel Gulftown 6core(12 HT) 1.17 billion transistors
AMD BD 4module(8 core) 2 billion transistors
Intel's chip is not only 1/2 the transistors, but it's also faster per thread. AMD needs to fix something, and fast
Couple of 10 bucks, more likely. I've recently bought parts for an AMD-based PC upgrade (in Germany) and the price differences were
- about 7 euros more for 2x2GB ECC Ram, compared to the same amount of non-ECC. Both Kingston Value RAM BTW.
- maybe 10-20 euros more for a board that supports ECC. That one is not as clear-cut BTW, it is more a case of having less choices if you want ECC, and the cheapest boards tend to not support ECC.
In dollars, that's maybe 40 bucks difference total. Or 50 bucks if you want 2x4GB and assume a similar price difference.
Of course, if you buy Intel, it will be a couple of hundred bucks because their desktop CPUs don't support ECC at all. That means getting a Xeon, and those are expensive.
C - the footgun of programming languages
it is a better performing chip then thuban because the playfield is limited. that's it. the hybrid crossfire there, as you can see from benchmarks, gives the user all it needs for current gaming. further than that becomes performance enthusiasm. this is not about the performance of the unit as a CPU. its about the performance of the unit as a unit for gaming.
not to mention in roughly a year, they will be replaced by trinity.
Read radical news here
> and if you do you can crossfire it with the cpu contained one
Yes, if you want to make your games run *slower* -as you see if you read the article that you linked to ;)
They test WinZip and make a big deal about how limited the new AMD chip is under it. Yet they show it deals much better with 7-Zip, which is the same sort of program and free (as in beer and speech, LGPL) to boot. Then they make a big deal about WinZip performance, as if it matters.
Um, hello? When are they going to stop punishing the chip for a poorly implemented application? Sure, if a popular game title or the editing suite you absolutely must use at work performs poorly because it is coded poorly and a processor can make the difference, buy the processor you need. The professional workstation should be specified around the high-end software needs anyway.
If, OTOH, you have an option to pay thirty bucks for something easily replaceable with a free program that forms a small part of your use, why not take the free one that offers good performance on both chips?
why ? ?
Read radical news here
I think testing AMD's FX chip and Intel's by pairing them with powerful graphics cards kind of misses the point of Bulldozer, which would be how well it does without the additional graphics card. I think the best comparison would be $235 for the AMD chip and $225 for the Intel chip but add to the Intel price the price of a mediocre graphics card that you would need to add to it to do the stuff the AMD chip does by itself.
The Llano is one of these cases, as the CPU has to share the memory bandwidth with the built-in GPU. A review I've read (can't remember where) showed it to be really bandwidth limited, an Athlon II X4 with discrete GPU (that was not much stronger on paper) performed much better.
So I guess the Llano could really use quad channel memory. Probably not going to happen, but at least you want to utilize those two available memory channels.
C - the footgun of programming languages