AMD Releases FX-Series Bulldozer Desktop CPUs

[psycho_driver]

Nevertheless, Bulldozer just doesn't seem worthwhile (not saying anything about their gfx, though) though it is still faster in every way than the previous gen, I think.
Even in multithreaded apps its not really any better than SB especially when you consider the number of cores in use (realising that amd is obfuscating this metric by using the term module).

It is pretty debatable as to whether or not the two integer cores within a bulldozer module should be considered true cores or not. The Windows 7 scheduler issues that AMD is claiming is costing them some performance in these benchmarks? Microsoft says they're working on fixing it in Windows 8 by having the scheduler treat a Bulldozer module the same way it treats an i* core with hyperthreading.

I think AMD would have been a lot better off just calling their Bulldozer module a core. Everybody would be complaining a lot less about today's benchmarks if the 8150 was marketed as a 4 core processor rather than an 8 core which gets trounced by Intel's 4 core offering in a few areas.

[liam]

It is pretty debatable as to whether or not the two integer cores within a bulldozer module should be considered true cores or not. The Windows 7 scheduler issues that AMD is claiming is costing them some performance in these benchmarks? Microsoft says they're working on fixing it in Windows 8 by having the scheduler treat a Bulldozer module the same way it treats an i* core with hyperthreading.

I think AMD would have been a lot better off just calling their Bulldozer module a core. Everybody would be complaining a lot less about today's benchmarks if the 8150 was marketed as a 4 core processor rather than an 8 core which gets trounced by Intel's 4 core offering in a few areas.

That's certainly true but then people would be wondering, 1)why does a 4 core AMD chip cost so damn much, 2) why does a 4 core chip need 2 BILLION transistors, 3)it's still slower than Intel's quad cores (I don't recall a single benchmark where it beat SB).
However, it is an interesting architecture (I think the 4 issue per module is actually a cool way to save space) but that cache latency is pretty damn high, but that's been an issue with AMD for awhile.

[Fenrin]

That's certainly true but then people would be wondering, 1)why does a 4 core AMD chip cost so damn much, 2) why does a 4 core chip need 2 BILLION transistors, 3)it's still slower than Intel's quad cores (I don't recall a single benchmark where it beat SB).
However, it is an interesting architecture (I think the 4 issue per module is actually a cool way to save space) but that cache latency is pretty damn high, but that's been an issue with AMD for awhile.

Accoding to a big IT news publisher (heise.de) the FX-8150 is in many multithreading benchmarks faster than a Intel Core i5-2500K (4 core, about 19% cheaper than a FX-8150). Examples: 7-zip compression almost 50% faster, WinRar 38% faster, Cinebench R11.5 11%, Linux kernel compiling 20% faster. At data compression it can also beat Intels Core i7-2600K.

To 2) FX-8150 has a huge cache, that's why it has so much transistors. Don't ask me why it uses so much cache.

[devius]

Even with the P4 it did out perform the P3 in most areas, the same cannot be said about BD. The P4 just did not match the competition.

That is just wrong. The P4 didn't outperform the P3 (let alone Athlon) in most areas when it came out. When it did outperform the P3 (1GHz) it was by a small margin, and not what would be expected by a chip with 50% higher clock rates and a new architecture. Ok, in synthetic memory bandwidth tests it beat the crap out of everything else at the time, but that was it. The P4 was exactly in the same situation as Bulldozer is in right now. In some very few tests it manages to barely beat the competition, but in most cases it falls far behind.

[deanjo]

That is just wrong. The P4 didn't outperform the P3 (let alone Athlon) in most areas when it came out. When it did outperform the P3 (1GHz) it was by a small margin, and not what would be expected by a chip with 50% higher clock rates and a new architecture. Ok, in synthetic memory bandwidth tests it beat the crap out of everything else at the time, but that was it. The P4 was exactly in the same situation as Bulldozer is in right now. In some very few tests it manages to barely beat the competition, but in most cases it falls far behind.

OK we will draw a parallel between P4 and Bulldozer and it is still the exception to the rule rather then the rule. The question is can AMD really afford to have a P4? When do they start considering not killing the x6 line and dropping it down to 32nm and adding functionality to it? This is what intel had to do with their line after the P4 didn't pan out, the difference being is that intel could afford it, I'm not so sure AMD can.

[liam]

Accoding to a big IT news publisher (heise.de) the FX-8150 is in many multithreading benchmarks faster than a Intel Core i5-2500K (4 core, about 19% cheaper than a FX-8150). Examples: 7-zip compression almost 50% faster, WinRar 38% faster, Cinebench R11.5 11%, Linux kernel compiling 20% faster. At data compression it can also beat Intels Core i7-2600K.

To 2) FX-8150 has a huge cache, that's why it has so much transistors. Don't ask me why it uses so much cache.

Fascinating. Anandtech has the opposite (this is the i7-2600K so it's clocked a bit lower than the FX but it's pretty close).These are all multithreaded, BTW.

3dsmax 9 SP1: SB:20.1 FX:16.1
Cinebench 10: SB: 22875 FX:20254
Cinebench 11.5: SB:6.86 FX:5.99
7 zip (32MB): SB:19744 FX:21041 (finallly a win!)
par2 (deals with archive parity): SB:17.3 FX:17.6 (lower is better)
x264(pass 1): SB: 94.9 FX:75.5
x264(pass 2): SB:36 FX: 35.8 (basically a tie)
x264(pass 1 w/AVX): SB:150.3 FX:121.6
x264(pass 2 w/AVX): SB:38 FX:38.6 (another win!)
Adobe artist retouch: SB:11.3 FX:14.8 (lower is better)
chromium compile: SB:18.6 FX:28.8 (lower is better and this is compied with visual studio; does anyone know what compiler that uses?)
stupid excel monte carlo: SB: 11.1 FX:14.2 (lower is better)

So there you go. Two wins for BD (one of which was basically a tie, though). What is interesting are the FP numbers in x264. That goes to show that they made a good choice with their arch, IMHO.

BTW, since these are all facts, I don't think reporting them in this detail should be an issue.

One more thing, those three things I mentioned were more jokes than anything else. I was pretending to be a semi-knowledgeable chip enthusiast who sees the transistor count for a part that AMD is calling (in our hypothetical) quad core. Of course since it is closer to 8 cores in function the transistor count makes sense (also has 16MB of cache, including L2+L3). Make sense?

[duby229]

Check this out guys...http://www.xtremesystems.org/forums/...-Threaded-Perf

OK, so if 1cu = 1 module and 1c = 1 int processor then 4cu/4c would be 4 modules with 1 int processor activated per module.... 2cu/4c would be 2 modules with 2 int processors activated per module....

Very interesting results... I think we will see alot of performance advantage by treating the 2nd int processor the same way as how hyperthreading treats the logical core. Basically it comes down to scheduling threads so that each thread will be able to run on its own module. We may need a patch to the kernel so that it is capable of scheduling threads for the best performance.

That is of course only if your running 4 or less threads... If your running 5 or more threads then some modules will have to run 2 threads...

[Kivada]

The new cpus seem to have got even more inefficiant single cores than the phenom series. That means every apps that prefers to use mainly 1 core is even slower with a new cpu. even the cheapest pentium g620 (with only 2.6 ghz) beats the new chips in that discipline (cinebench 11.5). As there are not that many games that run faster with more than 4 cores it is really weird that the chips should be for GAMERS, thats absolutely not logical.

The gamers interested in this are the same that run video transcoding and F@H while they game, which you see allot more of then you'd think at the top end of the market, simply because they have the power to spare and because their box is always on always at least running some sort of distributed computing project.

[Kivada]

If I remember correctly, AMD told about a year ago that one Bulldozer core would offer at the same frequency about 90% throughput performance of an Phenom K10 core. This CPU is clearly not for people who want single core performance.

Exactly, because the desktop market doesn't matter anymore performance wise, you aren't going to see much, if any real world difference between a bottom of the line chip and a top of the line chip in any of the things 95% of the people on earth use their comp for. So why waste time targeting the 5% when almost everything in the server market loves increases in cores and parallelism and is far more profitable then that %5 of "high end" desktop users could ever be?

So what ends up happening? You start designing for the server market and take scaled down versions of those chips and sell them as desktop parts.

What does matter to the desktop market these days? Competent graphics, no, not top end gamer stuff, just stuff better then what Intel sticks on as an afterthought. Why? Because its the only place where your 95%er will notice extra performance when his porno doesn't drop frames or tear with a little 3d for those "Oooo, shiny!" window effects.

Will this change? No, only GPUs will be more important if we end up with 4K res monitors and video, the CPU requirements will still be handled perfectly by an old P3, since the GPU will do all of the actual heavy lifting, even if transcoding.

Don't bother bringing gaming into it, 99% of titles out there all flatline across any not bottom of the line CPU at a res of 1920x1080 or higher because the GPU runs out of steam long before the CPU does. You only see pronounced differences at 1024x768 and under, which nobody not running a netbook or craptacular laptop ever run.

[duby229]

Exactly, because the desktop market doesn't matter anymore performance wise, you aren't going to see much, if any real world difference between a bottom of the line chip and a top of the line chip in any of the things 95% of the people on earth use their comp for. So why waste time targeting the 5% when almost everything in the server market loves increases in cores and parallelism and is far more profitable then that %5 of "high end" desktop users could ever be?
..............

I disagree heartily.... Despite your sentiment there -IS- such a thing as mindshare. It is clear that AMD has lost alot of it. The other important thing is, dont discount 5% when your marketshare is less the 20%

Point being that every single time that AMD gained mindshare, they also gained marketshare in every single market that they competed in. AMD lost mindshare with Barcelona and look what happened to there marketshare. They gained mindshare with Thuban, and look what happened to there marketshare... Its an unavoidable fact and AMD seems to be completely oblivious to it...

I believe AMD's single biggest problem is that they are allowing several key marketing guys to make engineering decisions, and then when the engineers cant make it work, they fire the engineers... Its just plain retarded.