AMD FX-8150 Bulldozer finally tested

[BababooeyHTJ]

So what? For CPU testing, canned benchmarks are a MUST in order to achieve accurate repeatability. If not, the results are all over the place. In GPU testing, walkthroughs can be done since the entire point is finding where the GPU bottleneck is.

If you can prove to me that another method is as accurate, I'm all ears.

I don't know my results were pretty repeatable by just walking around in a cpu limited game like Fallout New Vegas. For each test I just loaded up the same area three times, five if there were any inconsistencies. I just loaded up the area outside of Vault 3 and shot up some baddies. For GTA4 I just drove around for around the same amount of time for the same amount of time for each test. In Serious Sam HD I just loaded up the beginning of Oasis and killed a few baddies until I got to a certain point. In Crysis I just loaded up First light and played until I hit a loading point. All of the results were with in a few fps of each other and I saw some scaling. Sadly I did not save my fraps log.

Actually here is what I saw when I ran through Oasis with a stock i7 860 and a 6870

SSHD Oasis

2010-12-28 15:32:24 - SamHD
Frames: 14888 - Time: 214669ms - Avg: 69.353 - Min: 43 - Max: 116

2010-12-28 15:36:08 - SamHD
Frames: 14306 - Time: 204958ms - Avg: 69.800 - Min: 37 - Max: 118

2010-12-28 15:39:43 - SamHD
Frames: 14809 - Time: 212689ms - Avg: 69.627 - Min: 46 - Max: 122

Here is the same system but with the i7 860 set to just 3.6ghz.

2010-12-31 13:37:21 - SamHD Oasis
Frames: 21079 - Time: 210863ms - Avg: 99.965 - Min: 60 - Max: 158

Sadly these are the only runs that I saved from the fraps log. I can do a few stock vs overclocked benchmarks on my current cpu and post the fraps log. Generally when there are discrepancies like with the minimum frame rate they translate to the other tests as well so the difference between your best and worst runs for each cpu will be about the same. Obviously trying to do that in a game or area thats not cpu limited in anyway is just going to make it look like the results are all over the place.

Something like that would be nice to supplement the canned benchmarks. Plus actually using known cpu limited games would be nice. A high end video card would also be nice to see. I mean at least a GTX570.

[eric66]

how can 2 billions transistors be so SLOW?
sad, very sad day

I really hope that piledriver is not that bad...

Still don't understand why does movieman said that AMD have a winner. This is not a winner in any way you look at it

I'm sad.

lol simply add 10 more percents and you have junkdriver performance

[Buckeye]

Noticed that I am quiet too ;-)

I think everybody needs to cool down a little. JF-AMD is like me, passionate about what he does, if you don't believe into it, you can't put so much energy into making those Processors.
Green or Blue, we are all Geeks excited and just never forget, those Processors are the most complex machines build by human kind, at the smallest scale.

Not one man can claim to make any of this by himself, those are massive undertaking, the size of those transistors are mind blowing.

so, yes, there is competition, and yes, this architecture is not the happiest , but at least show some respect to the men who dedicated their life to build such complex machines.
It is ok to make a little fun of it, but please respect the men behind it, and get some perceptive, see how large is the achievement of putting something like Bulldozer together.

So, I am respectfully asking all of you to stay gentlemen ... No personal attack is ever acceptable.

Thanks all , stay on the high road!

Francois Piednoel

Much respect Francois for making this post.

Yes everyone please keep the flaming down and be respect full here.

I am sure then when some of us start running these chips over 8.0ghz it will be fun. After all that is what many of us here live for.
There were many demos run not all that long ago with BD CPU's doing some amazing things so please put stuff in perspective.

[solofly]

Ok folks it's over now and we can go back to our lives...

[G0ldBr1ck]

AMD is expecting at least 10-15% increase at core level alone. Thats not including the higher clock speeds so if they can hit the goal. Piledriver might at least get them back on track.

[Brice MJ]

AMD is expecting at least 10-15% increase at core level alone. Thats not including the higher clock speeds so if they can hit the goal. Piledriver might at least get them back on track.

"How will Piledriver get its projected 10 to 15 percent speed up? AMD says one-third will come from IPC improvements like structure size increases (so, three to five percent) and two-thirds will come from power optimizations that reduce consumption, enabling higher frequencies at a constant TDP (another six to 10 percent)."

http://www.tomshardware.com/reviews/...fx,3043-9.html

[[XC] Lead Head]

More performance using higher clock speed at a constant TDP? Sounds like Intel's original goals with NetBurst. Not the way to go.

[BeepBeep2]

Disabling every other "core" (cluster, part of the module internally) gives about 25-35% IPC increase it seems.

You know what that means? I believe it means "IPC increases"
Too bad though, that makes it a quad core and slower than the 8 core in applications that can use 8 threads.

AMD should have sorted out how to make each compute unit (module as PR likes to say) work as one thread when 4 or less threads are executed. They don't, hence ty CPU.

I hope the Windows 8 scheduler fixes this, or at least AMD could come up with a driver...or something.

Still, strong quad at 5 Ghz will rock.

I think I'll buy an 8120 now. I already have the damned motherboard, so it's too late to sell off and turn to sandy.

[InCredible]

Disabling every other "core" (cluster, part of the module internally) gives about 25-35% IPC increase it seems.

there benchmarks of that? i just would like to see

[haylui]

Disabling every other "core" (cluster, part of the module internally) gives about 25-35% IPC increase it seems.

You know what that means? I believe it means "IPC increases"
Too bad though, that makes it a quad core and slower than the 8 core in applications that can use 8 threads.

AMD should have sorted out how to make each compute unit (module as PR likes to say) work as one thread when 4 or less threads are executed. They don't, hence ty CPU.

I hope the Windows 8 scheduler fixes this, or at least AMD could come up with a driver...or something.

Still, strong quad at 5 Ghz will rock.

I think I'll buy an 8120 now. I already have the damned motherboard, so it's too late to sell off and turn to sandy.

where do you have information on disabling core in a module will give 25% IPC improve?

[BeepBeep2]

there benchmarks of that? i just would like to see

http://www.xtremesystems.org/forums/...-Threaded-Perf.
It's pretty interesting.

As low as 7% in some benches, but up to 30% like in wPrime and 25% in Fritz Chess, 20% in Vantage CPU.

Keep in mind that the CPU will produce both less heat and use less power in this config...but theoretically you should also gain OC headroom.

I bet BD can pull sub 10s in 1M under LN2 easy when configured properly.

Hopefully that CPU is running at 3.6, if it's at 3.9 then ST IPC is still pretty horrendus.

1100T limited to 4 cores at 3.6, 3000 NB and DDR3-2000 CL6 = 10.5 seconds wPrime, FX at that speed is around 12.5 on 4 comp units (PR modules)

But will OC to 5 Ghz instead of 4.2.

I think it is a decent upgrade in ST performance in that config...rough calculations I got at 5 Ghz it will do about 5.1 pts in cinebench over 4 cores.

[stangracin3]

its in the amd section under "single thread" thread

a nice thought out review. so technically ipc may have increased... it just could be a bug amd couldn't figure out in time.

edit: beep beat me to it

[Sparky]

It may be as someone else mentioned where there is a lot of L1 cross-referencing/corrupting/whatever the heck it was, basically it was having to repeat certain things several times on certain data loads. A design flaw if you ask me but maybe that is one major part of it all. Maybe they'll get that all sorted out with PD? No idea IMO they should have had whatever it is sorted out before the BD launch, but what's done is done

[Viking1962]

I droped the plans to buy a Bulldozer (for now) and get a Phenom II x6 1100T. I hope Piledriver will be much better when it come out (if they ever will be available)

[G0ldBr1ck]

Stangracin3, for some reason I canot find the thread. has the name changed?

[stangracin3]

Stangracin3, for some reason I canot find the thread. has the name changed?

oops gave you the wrong name lol...

http://www.xtremesystems.org/forums/...-Threaded-Perf.

[G0ldBr1ck]

Excellent. I would like to see more testing done on this. It does appear to be an IPC increase of PH2.

[stangracin3]

Excellent. I would like to see more testing done on this. It does appear to be an IPC increase of PH2.

something tells be there will be a B3 with this fixed in Q1/12 if they don't have PD ready soon.

[Andi64]

Excellent. I would like to see more testing done on this. It does appear to be an IPC increase of PH2.

To me it seems BD without a cluster active gains IPC to be on the "not so bad" league, but it's still behind PH2... (just compared those test with my X6 running 4-threads and the BD loses every time).
It will have a difficult time matching a Phenom II at 4Ghz.

[stangracin3]

To me it seems BD without a cluster active gains IPC to be on the "not so bad" league, but it's still behind PH2... (just compared those test with my X6 running 4-threads and the BD loses every time).
It will have a difficult time matching a Phenom II at 4Ghz.

did you remember to set your x6 to near 3.3-3.6 and the nb to 2000-2200 ish to match the bd?

[Wrench]

What I would like to know is how well do these perform under a virtual environment like Vmware.
The VMkernel is a good scheduler this might work in AMD favour, if you can stick more VMs on one core that be great. I would love to see a 16 core CPU

[Brice MJ]

To me it seems BD without a cluster active gains IPC to be on the "not so bad" league, but it's still behind PH2... (just compared those test with my X6 running 4-threads and the BD loses every time).
It will have a difficult time matching a Phenom II at 4Ghz.

4M/4C is about 5% faster than 4M/8C in games

http://img696.imageshack.us/img696/7746/img0033832.gif

so yes, probably phenom II is still faster

[AzureSky]

Are you blind? Have you even looked at the benches? Even in heavily multi-threaded scenarios, it loses to a quad core 2600k. It won't even perform anywhere near hexa core Intel CPUs like the i7 970 or the upcoming SB-E Hexas.

Attachment 121199

Attachment 121201

You're right about one thing, BD plain sucks in single threaded scenarios.

I think you need to watch the video about maxon and intel, maxon(who make the bench your images refer to) worked directly with intel using all intel tools to create the bench, intels compiler(icc) is known to be horribly bias against non "genuine intel" processors, sending them a far slower code path then what they send intel parts, this means that no non-intel chip has a shot at beating intels products without either patching the program to remove the bias cpu dispatcher or change the cpuid of the chip in question.

http://www.agner.org/optimize/blog/read.php?i=49

http://youtu.be/-M1eX5Vvrg8

so, yeah, cinbench is yet another bias canned bench tool that really should only be used to compare chips of the same brand against other chips of the same brand, and even then, its turned to make better use of HT then more real cores im sure, after all, it was developed with intel tools and intel help.

its also why I dont trust sandra results, sandra has and likely always will be an intel bias bench, even back in the a64 days sandra showed the p4 being far better then it was in real life use....

[RussC]

Oh great, so now were back to complier issues. Jeff Goldblum in Into the Night, "now you don't know what to believe".

RussC

[Drwho?]

No no no ... They are no CPU dispatch code in cine bench, just verified, and AMD is working with Sandra, as intel does, and they have their own optimized code path in Sandra.

So, please do not start Fuding the all thing, be accurate, and only speak with verified facts.

As I just did.

If you look for the CPUID opcode in cinebench, you will find it only in the part of the code that identify your CPU to report it on the screen.

Thanks!

Ps: forgot, if you want to verify yourself, count how many retired instructions are done on intel and AMD and compare the numbers, they land very very very close to each other on cinebench

Francois