EINSTEIN: Power/Production Ratio

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 770
Credit: 573,640,270
RAC: 180,885

RE: Bare in mind that the

Quote:
Bare in mind that the GPUs are actually quite different, with the Bonaire GPU being a new own design and having much more extended power saving features included into the GCN 1.1 architecture.


Bear in mind that the GPUs are actually quite similar ;)
Provocation aside, you are right in that the Bonaire chip is more efficient than the first wave of GCN cards, as the comparison between HD7770 and HD7790 quickly reveals. And there have been a few computing-oriented tweaks in GCN 1.1, but from what I read I'm not convinced these are of any benefit if one doesn't program explicitly with them in mind.

What I find amazing, though, is the extremly high voltage (for 28 nm) AMD can get away with on this chip. 1.26 V are unheard of in the 28 nm generation. This means the chip should easily reach 1.2 GHz, possibly even 1.3. Or could have its efficiency improved by a signifiantly lower voltage at stock clock.

Quote:
And other sites already confirmed that it indeed draws much less power (consistent with its lower TDP) than the older HD7850 using GCN 1.0 and a different design.


Anand measured savings of 22 W at the wall under Battlefield 3, but this includes an increased CPU load due to the higher frame rate of the HD7850. And as I linked to before 3D center gathered 15 W less for the HD7790, averaging over all test on launch day.

Yes, the HD7790 does draw less power. But not as much as the TDPs suggest.. and thereby I challenge your statement that the HD7790 would have a better performance/Watt at Einstein. The data derived from gaming doesn't look like that (IMO), but we need actual measurements to know for sure.

If the HD7790 actually requires less cooing effort depends on how much less power it will actually draw. Fans at moderate speeds typically consume 1 - 2 W, so that's not really a factor here (and both need active cooling).

Quote:
For projects with lower VRAM bandwidth requirements (which should be most, Einstein seems to be among the most demanding in bandwidth requirements from what I've read), it is ... the preferrable choice over the HD7850 for these reasons (additional plus : it's much cheaper as well).


Here I agree with you :)

I'd be interested in the voltages your cards run at. Both should have plenty of reserves, either for higher clocks or lower voltages. On the HD7790 running Einstein I'd rather OC the memory (relieve the bandwidth bottleneck) and lower the voltage of the chip (no point in pushing a bottlenecked chip higher, and with a high stock voltage significant power savings may be realized). On the HD7850 you could do either.. I'd likely push GPU clocks higher.

MrS

Scanning for our furry friends since Jan 2002

FalconFly
FalconFly
Joined: 16 Feb 05
Posts: 191
Credit: 15,650,710
RAC: 0

Before I got into the power

Before I got into the power measurements, I have to look into something odd that occured :

As I set only Perseus WorkUnits to run for a test, I'm now getting roughly these runtimes :

HD7790 PCIe Slot 1 : 16400s
HD7850 PCIe Slot 2 : 22300s
HD8670D : 49500s

*scratches head*

Both PCIe Cards run slower now, with the HD7850 taking the biggest hit (??)
Both slots are true PCIe 16x 2.0 slots. I've disabled the Windows PCIe power saving scheme for now, although I have doubts this does anything.

Since Catalyst only lets me see the primary Card data (temps etc. are okay), I'll have to see if GPU-Z or similar programs lets me see the other Card's temperature (should be okay though, they both get pounded with fresh air from 7 120mm Fans :p )

Weird....

-- edit --
I've taken a look at all GPUs with GPU-Z, everything looks okay.

Vcores as requested :
HD7790 : 1.150V
HD7850 : 1.038V

FalconFly
FalconFly
Joined: 16 Feb 05
Posts: 191
Credit: 15,650,710
RAC: 0

*lol* found the

*lol* found the error....

Turns out my Motherboard runs only 1 Slot at PCIe 16x, when in use the other PCIe Slot is limited to 4x. That eats away tons of performance :p

Einstein GPU computing really is pushing all I/O bandwidth limits, all around.

That means I'll have to find a new motherboard and will likely be unable to build an AMD platform with 3 discrete GPUs (can't find any Board that would support 3x full-up 16x PCIe Slots, not even 16/16/8)

ROBtheLIONHEART
ROBtheLIONHEART
Joined: 16 Aug 12
Posts: 47
Credit: 58,199,880
RAC: 0

I don't know if this helps

I don't know if this helps but the last time I was checking out MB's I remember Gigabyte has one with 2x16 2x8 and 2x4.If I remember correct its GA-990FXA-UD7 ? Although I don't think it will crossfire 3.

FalconFly
FalconFly
Joined: 16 Feb 05
Posts: 191
Credit: 15,650,710
RAC: 0

That's a Socket AM3+ Board,

That's a Socket AM3+ Board, I'm looking for an FM2 Board :/

I guess the current AMD FM2 Chipsets just don't support so many PCIe lanes.

Sid
Sid
Joined: 17 Oct 10
Posts: 164
Credit: 961,394,362
RAC: 360,826

RE: I don't know if this

Quote:
I don't know if this helps but the last time I was checking out MB's I remember Gigabyte has one with 2x16 2x8 and 2x4.If I remember correct its GA-990FXA-UD7 ? Although I don't think it will crossfire 3.


You may try to think about motherboard and processor that support PCI 3.0 specification. Roundly speaking PCI X16 2.0 = PCI X8 3.0.

FalconFly
FalconFly
Joined: 16 Feb 05
Posts: 191
Credit: 15,650,710
RAC: 0

I'll keep a lookout. For

I'll keep a lookout.

For now it looks like I found an acceptable compromise to run what I got.

HD7970 @ PCIe 2.0 x16 (wishes for PCIe 3.0 x16)
HD7850 @ PCIe 2.0 x8 (seems okay-ish)
HD7790 @ PCIe 2.0 x16 (definitely good)
HD7750 @ PCIe 2.0 x4 (surprisingly happy)

Seems only intel can offer the ideal PCIe 3.0 boards, but I'm in no mood to totally revamp all my systems and certainly don't want to spend all that cash.
After all, I assume Einstein is an extreme case concerning these bandwidth requirements, other GPU projects may run far more unrestricted (at least I hope so).

ExtraTerrestrial Apes
ExtraTerrestria...
Joined: 10 Nov 04
Posts: 770
Credit: 573,640,270
RAC: 180,885

For POEM not even a high end

For POEM not even a high end i7 with fast main memory and PCIe 3 16x could max out your HD7970, but others are most probably less demanding than Einstein.

MrS

Scanning for our furry friends since Jan 2002

Jeearr
Jeearr
Joined: 23 Feb 06
Posts: 14
Credit: 107,155,091
RAC: 0

Hi, thought I would chime

Hi,

thought I would chime in. I have a standard (non-oc variant) sapphire branded 7790 running on an intel i3-2120T processor (on a biostar B75MU3+ board). The system draws a total of 76 watts at the plug/cord/wall running BRP4G and BRP5 tasks. My average RAC is 29615 (average from boincstats), which puts it at 10,000 stones per day at 25.32 Watts.

Cheers!

FalconFly
FalconFly
Joined: 16 Feb 05
Posts: 191
Credit: 15,650,710
RAC: 0

My setup has changed quite a

My setup has changed quite a bit several times, the current setup is like this now :

AMD A10-6700 (internal GPU disabled)
HD7850 @ 8x PCIe 2.0
HD7850 @ 8x PCIe 2.0
HD7750 @ 4x PCIe 2.0

I basically tried to assemble a bit of an AMD "Stompinator", a powerhouse that still yields some efficiency, while avoiding single massive bottlenecks (not done yet due to PCIe limits).

It roughly chunks out something like 110000-120000 Cr/day, but also consumes 310W.
That yields about 371Cr/W and might be the presently fastest AMD based Host on Einstein.

As it is severely limited by the PCIe lanes and also by CPU performance, next month I'll modify it one last time to afford PCIe 2.0 16/16/8 with a different board (AM3+) and CPU (likely 6core FX-6300).
Wattage will increase a tad but performance should increase by ~40% as is.

In the end, it may run 2x HD7850 and 1x HD7790 or 3x HD7850.
That should place the rig in the Top 5 I hope, without busting my personal Wattage limits (absolute max. I want to invest = 500W, less is better and looks like I'll be good on that).

Comment viewing options

Select your preferred way to display the comments and click "Save settings" to activate your changes.