Bulldozer Review! Legit?

[Black96ws6]

From a Dutch site. Translated URL: http://translate.google.com/transla...b.nl/hardware/componenten/processoren/fx-8150

Thoughts?

 

[formulav8]

I don't think i've ever heard of that site? :\

5.0 ghz oc looks good

Looks right between 2500 and 2600 series cpus in a few areas.

 

[ThatsABigOne]

Sounds good to me. I really want true, neck and neck competition, even if I exclusively buy Intel.

 

[classy]

Based on all the comments we have seen, I would say that its legit.

 

[pcslookout]

Sounds good to me. I really want true, neck and neck competition, even if I exclusively buy Intel.

Same here. We don't know if it is real though. Taking it with a grain of salt as I always do until we get official word.

 

[StrangerGuy]

i7-965 is beating up BD so hard in gaming that its not even funny.

 

[Despoiler]

It seems legit since they aren't linking off to fake OBR and DH benchmarks. I'm not sure I can accept them w/o seeing the full testing setup and methodology. They are not consistent with what rigs they use in each test. It's not upto the quality I'd expect from say Anandtech.

I like the 5GHZ @ < 1.5v overclock though.

 

[LOL_Wut_Axel]

Not really a good review; not enough benchmarks.

But I'll say it again: FX-8120 will be ~10&#37; faster than i5-2500K in multi-threaded, ~25-30% slower in single-threaded.

Reviews will be here in very little time, so you'll see.

 

[Tomashoch]

It's real. The source is the website of a physical magazine in the Netherlands, part of VNU Media/The Nielsen Company. PCM stands for Personal Computer Magazine.

The review is offline, so Google's Cache can be deleted within a few hours. I've saved an back-up at this URL: http://bouweenpc.nl/upload/pcm2/AMD FX-8150 Personal Computer Magazine.htm. Google Translate:
http://translate.google.com/transla...50%20%20%20Personal%20Computer%20Magazine.htm

 

[nyker96]

looks legit, but it seems IPC still below Intel's chips, thus performance/watt is lower. still seem at in games, amd has made some advancements compare to k10.5 architecture. they are close enough to intel now. however, for the price of 220Euro or close to 300 bucks, i think 2600k is better deal.

 

[Venom20]

i7-965 is beating up BD so hard in gaming that its not even funny.

Lol, nice comparison? Check the price tags again.....

 

[StrangerGuy]

Lol, nice comparison? Check the price tags again.....

There is something called "Sandy Bridge". What are you, living in 2008?

 

[LOL_Wut_Axel]

Lol, nice comparison? Check the price tags again.....

Comparing to the now-irrelevant Quad-Core Nehalem is pretty useless, except for looking at architectural comparisons and IPC (which, BTW, will still be significantly lower than Nehalem).

Sandy Bridge has ~11&#37; higher IPC than Nehalem, and the vast majority of games take advantage of two-four cores. Because of that, IPC is an important thing in them unless you're in a very GPU-bound game. We'll see in more in a bit when more in-depth reviews come out, but for gaming Sandy Bridge will be faster.

What you want to look at Bulldozer for is multi-threaded. THAT is where it'll be competitive, especially the FX-8120. Given these should reach at least 300MHz higher clocks at moderate voltage than Sandy Bridge, I'd expect the FX-8120 to match the 2600K both OCed in multi-threaded. For a lot less money.

 

[Black96ws6]

If this review is legit (we know the press kit is out there now), then I'm happy that AMD has created a better processor than the previous generation, but sad that it still lags behind in gaming\IPC.

Some choice comments from the article, slightly paraphrased by me due to the rough translation:

But how does the processor perform at stock speed? Test setup is an Asus Crosshair V w/ 4Gigs RAM and dual Radeon HD5970 video cards. HD is a WD Velociraptor.

Cinebench 10 single-threaded score: 4074
Nehalem i7 965: ~4900
2500\2600: ~5800

Multi-threaded: 20615
2500k: ~18615
i7 965: ~19600
2600k: ~22615

Cinebench 11.5 scores: 6.01
2500k: 5.37
i7 965: 5.73
2600k: 6.73

3DMark Vantage CPU: 19119
2600k: 22500

Total Score: 21949
2600k: 25500

3DMark 2011: 6616
2600k\i7 965: ~7385

I'll post Gaming scores next as this post is getting long

 

[StrangerGuy]

If this review is legit (we know the press kit is out there now), then I'm happy that AMD has created a better processor than the previous generation, but sad that it still lags behind in gaming\IPC.

Some choice comments from the article, slightly paraphrased by me due to the rough translation:



I'll post Gaming scores next as this post is getting long

CB10:
(5800 / 3800MHz) / (4024 / 4200MHz) = SB has 1.6x IPC per clock than BD.

Ouch.

 

[Black96ws6]

Here's the gaming benchmarks:

Gaming

Dirt3 1920 x 1080 All details on Highest setting
FX-8150: 105fps Avg, 75fps Min
i7 965: 93fps Avg, 71fps Min

Mafia II
FX-8150: 68.3fps Avg
i7 965: 76fps Avg

Far Cry 2 Max Settings DX10
FX-8150: 111fps Avg, 23fps Min
i7 965: 126fps Avg, 75.2fps Min

Interesting how in some it does better, in others, less...perhaps due to multi-core\threading support?

 

[Arkaign]

Comparing to the now-irrelevant Quad-Core Nehalem is pretty useless, except for looking at architectural comparisons and IPC (which, BTW, will still be significantly lower than Nehalem).

Sandy Bridge has ~11% higher IPC than Nehalem, and the vast majority of games take advantage of two-four cores. Because of that, IPC is an important thing in them unless you're in a very GPU-bound game. We'll see in more in a bit when more in-depth reviews come out, but for gaming Sandy Bridge will be faster.

What you want to look at Bulldozer for is multi-threaded. THAT is where it'll be competitive, especially the FX-8120. Given these should reach at least 300MHz higher clocks at moderate voltage than Sandy Bridge, I'd expect the FX-8120 to match the 2600K both OCed in multi-threaded. For a lot less money.

This is the best summary I've seen yet, both of what we should have been expecting, and now with the first credible-looking benches/review. Anyone expecting BD to be king of gaming performance was silly.

That said, this opens the door to feasible competition, depending on where prices go after this, and what eventually becomes the choice at the $150 price point where Intel isn't very exciting at this time (locked chips, dual cores, bleh). I'd also expect mainboards to be available for AMD that will support decent OC, have decent onboard graphics options, etc, all for less than intel Z68, P67, etc. The H61 and H67 aren't impressive in those areas at all.

 

[sangyup81]

Given that I expect it to cover the Nehalem gap but not quite leap over Sandy Bridge, this article meets my expectations but I still have to consider that the numbers might not be real

 

[Arkaign]

Here's the gaming benchmarks:



Interesting how in some it does better, in others, less...perhaps due to multi-core\threading support?

Dirt3 looks good, and we know AMD has been focusing on that title intently for a while now, but the FC scores are scary. Bumping up against cache limit or temporarily causing resource allocation issues with the FPU units? Those minimums are terrible.

Of course, this is the area where patches help dramatically.

 

[RussianSensation]

Not really a good review; not enough benchmarks.

But I'll say it again: FX-8120 will be ~10&#37; faster than i5-2500K in multi-threaded, ~25-30% slower in single-threaded.

Reviews will be here in very little time, so you'll see.

Maybe even worse than that.

Take a look at Cinebench R10. They reported FX-8150 scores 4,074 in single threaded scenario. Under such a case, full 4.2ghz Turbo is enabled.

Even the older X6 1100T performs as well. That means 1 BD core is slower than 1 Phenom core and IPC didn't improve at all. It actually has gotten worse. As suspected all these months by Xbitlabs and many many websites, AMD needed higher Turbo because per core performance was too weak. Hence so many re-spins and delays to get those clocks higher. :hmm:

It looks like this will continue the legacy of the X6 of performing well in multi-threaded apps, but still not be enough to close the gap in per core performance. Problem is unlike X6 1090/1100T go for $170-190; and in light of those supposed benchmarks, BD at $230-250 seems way too expensive. Most programs don't use more than 4 threads and 2500k is still cheaper. Not sure how in the world AMD is trying to claim the FX legend is back. FX obliterated Pentium 4.

 

[formulav8]

Yeah that 23fsb minumum looks WAY out of place. I'm wondering if he read the scores properly when he did the timedemo...

 

[Black96ws6]

Maybe even worse than that.

Take a look at Cinebench R10. They reported FX-8150 scores 4,074 in single threaded scenario. Under such a case, full 4.2ghz Turbo is enabled.

Even the older X6 1100T performs as well. That means 1 BD core is slower than 1 Phenom core and IPC didn't improve at all. It actually has gotten worse. As suspected all these months by Xbitlabs and many many websites, AMD needed higher Turbo because per core performance was too weak. Hence so many re-spins and delays to get those clocks higher. :hmm:

It looks like this will continue the legacy of the X6 of performing well in multi-threaded apps, but still not be enough to close the gap in per core performance. Problem is unlike X6 which go for $160-180, at $230-250 that's way too expensive since most programs don't use more than 4 threads. 2500k is still cheaper than that.

Yeah IPC = OUCH. Stock clock is 3600. Let's assume Turbo is not working here and needs a patch, otherwise then those numbers @4.2Ghz look bad compared to the 1100T@3.3Ghz (turbo'd to 3.7). Why not just do a die-shrink on the existing Phenom X6?

It looks like I'll be picking up a 2600k the next time I see one for ~$279 between now and Black Friday...

 

[RussianSensation]

Sandy Bridge has ~11&#37; higher IPC than Nehalem,

At least 2 reviews say 14% average increase in IPC:

Review 1
Review 2

Also, if you look at Computerbase's and Digit-Life's analyses, the average is dragged down by the gaming average. If you just look at IPC increase in applications, it's actually much closer to 20%.

Based on this and the fact that BD turbos to 3.9ghz/4.2ghz, SB appears to be 45-50% faster in IPC.

Why not just do a die-shrink on the existing Phenom X6?

1. The module design is far more forward looking. Early P4s also didn't really outperform P3. In 3-5 years from now, if programs scale to 8-16 threads, an 8-16 core BD would fare well. AMD has previously indicated that the module design allows them to double the core count by only increasing die space by around 12% iirc. So from a cost/die size perspective to manufacture the CPU, BD is better for AMD.

2. Theoretical architecture/design of BD may have put it in a far better light than what turned out to be the case in the real world. But after spending $$$ and 5 years developing it, it was too late to consider shrinking Phenom II.

3. This CPU architecture may scale farther with clock speeds than perhaps even Phenom II could do on 32nm. We will have to see. Once process matures and AMD shrinks this design to BD Next to 22/28nm, I can see 5.5-6.0ghz clocks in the cards.

4. The CPU's power envelopes and the entire design around dynamic power consumption gives it a huge advantage in server space. Even if you shrank the Phenom II, I doubt it would have been as power efficient.

CB10:
(5800 / 3800MHz) / (4024 / 4200MHz) = SB has 1.6x IPC per clock than BD.

Ouch.

That would be really devastating if true. If Haswell increases IPC another 15% by 2013, AMD may need a 12 core CPU to compete with a 6-core HT Haswell in multi-threaded apps.

 

[Martimus]

Why not just do a die-shrink on the existing Phenom X6?

Even a die-shrink at 32nm may be worse than the mature and functioning X6 on the 45nm process. The upside is that since the 32nm process has shown its warts are quite large, that there is a lot of room for overall improvement just from process improvements. (Although I don't remember seeing a lot of improvement on the 65nm process from AMD, but I didn't really follow it then.)

 

[MentalIlness]

Make you wonder how Bulldozer will do in Battlefield 3. Knowing it loves more cores.