NomNomNomNom...

[Biffa]

My main server at home which is a VMware server (DC/Exchange/SQL) has always led a sekret life, only masquerading as a good upstanding decent hard working grunt, but really flying the bigadv flag for OcUk as a Cpu only cruncher.

Its been a Q6600, an i7 920, and now it is about to get a transplant so that it can live up to its name (Gemini)

This thread is going to be the fast version of the build, not the full build log.

Full project rebuild log here

 

[Biffa]

was the new case really needed
in for a penny in for 2 grand

Unfortunately yes:

Its currently only in a little PC-A17 You need a case that can take the depth of the motherboard, currently there are only 2 or 3 stock cases that can do it, or you can get a custom MM mobo tray and hack it into a Corsair. The only other options are to buy a MountainMods case itself or a DangerDen Plexi case. Neither of which rock my world.

The PC-V2010 fits, but thats even more expensive, and not as good for watercooling I reckon.

Lucky the case isnt going to be on display anywhere as its an ugly mofo, this isn't going to be an elegant build, its going to be functional

 

[Biffa]

My good lord biffa, I'm guessing you didn't like speed beating you on the PPD front. I had my suspicions after you other thread. What's the planned gpu setup for Gemini or is it going to remain CPU only. Would be a waste of a 4 slot board.

Heh, no this was all ordered before this week started, speed and I swap PPD places quite often according to the vaguaries of bigadv allocation.

ChrisB had the big cahoonas last week, but no its not because someone might be catching up, its more to do with efficiency and my own curiosity taking advantage of an opportunity.

And no I won't be putting any GPU's in this one.

That looks seriously.....serious. Don't stop there man don't we get to know whats going in it, 24 threads 4 x ??? gpu, o/s damn this is more exciting than an nvidia launch party.

Nope, CPU only on this one.

One question: Does it smell good?

No but its huge, I thought the case box was massive, it was only when I got the motherboard box out that I realized how enormous that was!

The CPU boxes were a huge anticlimax, I initially thought they had forgotten to send them but then they were wrapped in bubble wrap at the bottom of the box.

Thats a i7 920 box, a i7 970 box (big cooler on those) and the X5650 box, now given the X5650 was around the same price as the i7 970 I was a bit dissapointed with the box size

To answer the GPU's question. OK this build is about folding efficiency. I've done the 6 core (12 thread) CPU and 4xGTX470 build with weebeastie and that pulls in around 90K PPD on average with a bigadv running on 10 threads and 2 threads dedicated to the GPUs. It pulls near 800W from the wall its good but not *that* good.

Problem is that GPU3 is a nightmare for getting in the way of the CPU and bigadv production does suffer badly.

According to results from other teams I should be able to push between 140-160K PPD from the one rig with decent bigadv WU's and for under 600W, well thats the theory

So with regular 160K PPD pushing out on Gemini I'll have more options to spread the load of projects to my other kit thats perhaps not optimised for folding. I've got some C2Q with 9800GT's in that seriously underperform for Folding that might be better used on Seti@home. Also my main home rig is not really best suited to run bigadv on folding.

I may switch weebeastie to bigadv only, and put the fermi's to work on Seti or another boinc project.

Plus if I feel like it I may consolidate my lead in Folding, and then I've let you boys pull ahead in S@H recently and I fancy getting my place back

 

[Biffa]

*sends a prayer up to the lottery gods*

Give me a million quid and trust me, I am kitting out an entire room with air-con and super foldy rigs!

As for the GPUs getting in the way of Bigadv, have you tried forcing CPU affinity to specific cores? You should be able to do it through Task Manager if the client doesn't offer an option. Should be less interference if the GPU3 clients can't physically encroach on the remaining cores.

I already do that on weebeastie. But for 4 GPU's I'm assigning 1 real core and one HT core on the i7 970 which drops its performance down to a i7 920 level.

So around 35-38K PPD for the i7 970 on bigadv and 13.5-14K per GPU = around 90K PPD, theoretically I might get 120K PPD with it like that, but I'd rather get 160K PPD

Insane amount of pc power there Mr Biffa, i'm allready tidying up my cell for when you come a calling....am fresh out of major funds to compete with that array, your going to have a ball building that lot sah!

Heh, not with what I know you are planning on buying.. are they £399 or £450+?

I'm pulling under 150w for 15k at the minute - loving the 912pts units
Not sure the efficiency stacks up when you inc. cost - but thoses bonus points make a massive differance.
Anyone know how long they will be around for?

To show you how much multi-gpu/cpu folding affects the performance I get 14K on 912 and 925 pointers on the GTX470's on weebeastie.

Not sure on the bonus points but A3 SMP WU's get them and bigadv get them so I think there will always be some

I might not get that, it depends on WU but I'm sure as hell gonna give it a try

 

[Biffa]

I've got a 460 in my home machine, OC'ed to 900 core (from 675), and currently the i7 920 is running a 2685 bigadv.

GTX460 - 15,670 PPD @ 51s/frame

Compare to weebeastie, where there are 4 GTX470's overclocked to 700 Core (originally 608) but the CPU is a 970 running a non bigadv A3

GTX470 - 13,800 - 14,000 PPD @ 47-48s/frame

Yes it all adds up to impressive performance, but I needed a hex core i7 to get the 4 GTX470's to work anywhere near their proper performance envelope.

 

[Biffa]

Its a combination of things, this might be part of it from one of the Pande Group, its about the 480 but applies to the 470 and 465 as well.

I haven't actually done any analysis into this question, so I can't say for sure, but it's important to know that there's a lot more different between the GTX 480 and GTX 460 than just 480 vs 336 shaders.

The internal organization of the GTX 480 (GF100) is actually as 15 cores, each containing 2 16-wide computational units ("CUDA cores" are a marketing name - they are not real cores, as they don't compute independently). The GTX 460 (GF104) has 7 cores, each with 3 16-wide units (48 CUDA cores). GF104 also has twice the number of special-function units (to evaluate exponentials, trigonometric functions, etc.). Perhaps the most relevant change, though, is that GF104 is able to use more of its execution hardware at the same time (e.g., both the shaders and special function units), because it is superscalar - it can run more than one instruction at the same time from each thread (http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king/2).

My suspicion (and again, I haven't done the testing to test these guesses) is that there are a few things behind GTX 460 being more efficient than expected:

• Fewer SMs may make it easier to load-balance the GPU
• More special function hardware may increase throughput in a few parts of the code
• Superscalar execution provides an efficiency boost in per-shader terms.

So yes I think the 460 may be more efficient at folding than the 470, certainly the increased core speed would help a lot with some WU's

But I also think its down to the inefficiencies of the GPU3 applications e.g. Core_15, they need a lot of the attention of the rest of the system, esp CPU interrupts, so when you have 3 or 4 cards in there they could actually be fighting against each other a bit

When I was building weebeastie I had a funny experience of if I increased the overclock on the cards some of them would just throttle back. I managed to find a good balance of voltage and overclock so that weebeastie has been stable wiith reliable output, but without a bluescreen or GPU crash or anomalie for a few weeks now. Thats the main thing, that I know I can get a reliable 90K out of it day in day out without it overheating or crashing. (it does do other things)

 

[Biffa]

Heh, you mad bugger, any time you need help just email, you have my address

 

[Biffa]

Damn blast and blitherations!!.. one supplier sends the wrong fans.. (120 instead of 140) the next sends the wrong cpu blocks (AMD instead of Intel fitting). How hard can it be!?

 

[Biffa]

Ooh thats happened to me you naught bumper you

I'll do some piccies tonight, I was doing accounts all night last night. Mind you I might be cream crackered tonight so I won't promise anything, can't even catch up on the weebeastie build log. (Think I have need to go and retake some pics of the final setup)

Anyhoo.. hopefully they will have the 1366 ones in stock. Supposed to be the dogs danglies as far as blocks go.

 

[Biffa]

AquaComputer Kuplex Kryos XT

Review here: http://skinneelabs.com/ac-kryos-xt.html

 

[Biffa]

EK HF was my next choice, if they don't have the one I want in stock then I'll probably get the nickel plexi EK HF

 

[Biffa]

Ooh suits you sir!

 

[Biffa]

Well we will see what we will see. There is a guy running a sr-2 with 8 Gtx480's but im not sure how legit it is

The 140mm Yates arrived yesterday and I spent half the day getting Spam to agree to stop dithering and just give me a full refund on the wrong blocks. Of course the ones i wanted are now out of stock so i have ordered the more purposefull looking black and nickel EK Supreme HF's instead. Which is what I should have done in the first place.

 

[Biffa]

think purposeful and mean

 

[Biffa]

Couple of piccies:
Remember I said the SR-2 was a big box?

SR-2 In the box/Out of the box:

EK SR2 Mobo block:
In the box/Out of the box:

Awww aren't they cute? reminds me of a twix ad.

The family all together at last

 

[Biffa]

Thats most kind of you sah, but my two should be here today

They are also sending a courier to pickup the ones they sent incorrectly.

Must say I forgot what a bugger putting EK blocks on is, I hate the EK way of using TIM to stick the standoffs on then trying to flip the board over and align everything without one of the standoffs slipping out of position. Most of the newer block makers use metal standoffs built into the block which is much more civilized and much less Heath Robinson

Its hard enough with their GPU blocks, but with a mobo block like this, well lets just say I was lucky this time and it went together first go, but I was frettin I tell ya!

 

[Biffa]

LOL

Caution: Please ensure the standoffs of you PC Case is aligned with Motherboard Mounting holes to prevent electrical shorting

jak, re L5640's yeah but I couldnt wait till I could toboggon on Satans front lawn to get one.

 

[Biffa]

You get extra plastic standoffs in the box for motherboard trays that don't have enough holes

 

[Biffa]

There was a heatsink under it.

On other news I got my CPU blocks, and I was worried because the pics from the supplier showed the old model but I got the latest model!

Thought I was gonna get these:

But got these

 

[Biffa]

Update:

So you guys are getting a preview of the highlights of the build really, I'm not going to go into every tiny step like a project page. So on with the show:

Board comes with pre-installed backplates to take Xeon CPU coolers. These have to go for the waterblocks or high end air coolers.

To do that you have to remove the CPU bracket

Then you can replace the backplates with standard ones, freeing up the holes to mount the block backplates:

Reflective much?

Current state: