In light of bigadv changes, new best ppd/$ recommendations

[KGH]

I think still early to jump into conclusion, especially based on such benchmark.
And also keep in mind, PPD is non-linear with BA bonus system, thus few minutes in TPF can make a huge difference in PPD ( 2 x 2P SR2s' PPD << 1 x 4P MC's PPD)

you can claim for desktop SB-E at high OC =~ 2 BD. When it comes to 2P SB-E vs 4P IL, you can not have the same projection: much harder to sustain high OC 24/7.

SB-E would consume more power@OC, so high maintenance.

 

[Psykoikonov]

Project: p6901
Average time/frame: 00:06:13 {in hh:mm:ss} ~317,863 ppd

Project: p6903
Average time/frame: 00:13:56 {in hh:mm:ss} ~509,764 ppd

CPU: Opteron 6176se @ 2.3 GHz
# of CPU sockets: 4
# of Physical cores: 48

# of FAH CPU processes: 48
# of FAH GPU Clients: 0

RAM GB installed: 32
RAM Type: DDR3
RAM Speed: DDR3-1333 @ 9-9-9-24, quad-channel
NUMA enabled/disabled/not used: ? never checked
OS name/kernel version: Ubuntu 10.10
Client: Linux 6.35
Running in VM: No

edit: added in rough ppd's from bonus calculator

 

[Macaholic]

I'm waiting until after Thanksgiving to build a 4P machine based on the 6272. I think numbers will start showing up with greater frequency in the next three weeks on many forums. Some preliminary numbers on a 2P 6274 16-core 2.2GHz machine that was a 2P 6128 8-core 2.0GHz;

Ok, got them installed and running. 6903 times went from 46 min/frame to 31 min/frame. PPD went from 88k to 150k. They've been up and running for 12 hours now with no problems....

 

[Zerix01]

due to their shared nature and simplicity a BD core is about as efficient as 90% Stars Core(PhenomII)

Well no, the 90% figure that AMD has thrown around is not 90% the speed of a PII. BD is faster single threaded than a PII. The 90% comes from them saying if they never bothered to share any CPU resources then current BD chips would be 90% the speed of that non existent chip. So they are admitting to a 10% speed penalty to share chip resources. But They were throwing around numbers of 20 - 30% faster per core over the prior generation. I need to see more benches for this last part to know if they met that goal.

I belive that AMD did not take into account Folding(F@H) when creating BullDozer. they wanted to pack as many threads into a single processor as possible, and that is awesome for server duties(runing VMs SQL) but its not so good for F@H.

Don't look at the unified FPU as an issue for F@H. The FPU can handle 1x 256-bit, 2x 128-bit, or 4x 64-bit instructions at a time. I thought F@H uses 64-bit SSE instructions but I know for damn sure they are not using 256-bit. So the shared FPU will not be a bottleneck for F@H. The FPU itself may not be as fast as Intel's, but the shared part of it should be of no concern.

 

[ChasR]

Almost all FAH projects run with single precision, which uses 32-bit FPU operations. Rarely, FAH projects use double precision operations which are 64-bit calculations.

 

[Zerix01]

Almost all FAH projects run with single precision, which uses 32-bit FPU operations. Rarely, FAH projects use double precision operations which are 64-bit calculations.

Ah ok. I never did get a good answer on how Bulldozer handles 32-bit FP instructions. So could be eight instructions at a time or still four, either way the module will still act as two cores vs if 256-bit instructions were used. I think each 128-bit half of the FPU is still based on the prior gen as far as instruction capabilities just to give some level of comparison.

 

[]-[itman]

Ah ok. I never did get a good answer on how Bulldozer handles 32-bit FP instructions. So could be eight instructions at a time or still four, either way the module will still act as two cores vs if 256-bit instructions were used. I think each 128-bit half of the FPU is still based on the prior gen as far as instruction capabilities just to give some level of comparison.

For an 8 core bulldozer cpu, it can handle up to 8, 128-bit fpu instructions at a time.

Here are the 4 likely scenarios for each cycle:

Core 1 - 2x128b AVX or 2x128b un-recompiled SSE

Core 2 - No FP command
------------------------------

Core 1 - No FP command

Core 2 - 2x128b AVX or 2x128b un-recompiled SSE
------------------------------

Core 1 - 128-bit FP command

Core 2 - 128-bit FP command
-------------------------------


Core 1 - No FP command

Core 2 - No FP command
------------------------------


As you can see, the flexibility of the FPU really gives total flexibility to the system, designed to deliver optimized performance per core per cycle.

http://blogs.amd.com/work/2010/10/25/the-new-flex-fp/

How well this is actually executed, that's a different story, but this is what the architecture is designed to do.

 

[AMD_OCER]

It would be a Supermicro H8QGi-F. Nice board. May get another Real soon. Yes it has OC potential. Dynatrons work fine, wee bit noisy, but they do a decent job for the price and size. Noctuas would be better, but I went inexpensive for the first G34.

Hi Mac, can you explain what OC potential the H8QGi-F has? I can't find anything searching the interweb or looking through the BIOS manual. I have heard that the Tyan's don't support OC at all.

 

[CJ145]

There is not a single dual / quad socket AMD system that supports overclocking right now as far as I know. The exception is if you have engineering sample chips which have unlocked multipliers.

 

[Macaholic]

There is not a single dual / quad socket AMD system that supports overclocking right now as far as I know. The exception is if you have engineering sample chips which have unlocked multipliers.

You got it. Kraken and Langouste author tear and sfield would be the two to contact in order to find out more. Definitely not for the faint of heart.

 

[KGH]

There is not a single dual / quad socket AMD system that supports overclocking right now as far as I know. The exception is if you have engineering sample chips which have unlocked multipliers.

not true, 12% OC on MC 4P is doable on most Supermicro boards (don't have to be ES chip). you just need to contact the right folks.

 

[CJ145]

I suppose that is better than nothing, if I built one I would still want to be able to hit the chips limit though, after spending 2k+ on a system.

 

[AMD_OCER]

I suppose that is better than nothing, if I built one I would still want to be able to hit the chips limit though, after spending 2k+ on the CPU's.

Fixed it for you...

Yeah better than nothing. I am not scared looking at the details... Just wish there was more to be done.

Pad mods maybe?

 

[Macaholic]

Fixed it for you...

Yeah better than nothing. I am not scared looking at the details... Just wish there was more to be done.

Pad mods maybe?

More information found here regarding optimal memory choice and configuration.

 

[GIXXERGUY6]

OMG I've been away far too long!!!

I don't like AMD but this picture just made me all happy inside!!! I think I'm going to sell my wife and get 2 of them(better make it 5 so she doesn't think she isn't worth much)

600k PPD vs my 920 and 930 I don't even have my other guys folding with me anymore

 

[Norcalsteve]

GIXX! Long time bud! That is a nutty rig... I just upgraded myself

 

[GIXXERGUY6]

GIXX! Long time bud! That is a nutty rig... I just upgraded myself

What is going on my friend?? I stopped folding at the beginning of last summer when I was jobless and we got hit with 2 separate months of electrical. Ills of over 500 due to the ac being in( that is t that efficient anyways)

If I get time this weekend I'm going to fire up my main and the other machine I have here. I'm going to talk to my buddy down south to see of he's still game(that is if I can figure it out again)