Observations on FGRBP1 1.18 for Windows

archae86
archae86
Joined: 6 Dec 05
Posts: 3145
Credit: 7023314931
RAC: 1824054

Gary Roberts wrote:I know

Gary Roberts wrote:
I know it's cheaper for me to buy an RX 480 than a GTX 1060 and I believe it will give a greater output as well.  Once I've checked that, I'll probably buy one to measure output against power consumption.

Gary, I'd be really interested in both the productivity and the power consumption comparisons, specifically on Einstein work.

I had an RX 480 order queued up back in the early days, and cancelled it.  I'm a bit foggy on the details, but am pretty sure I had the impression at the time that the power consumption per unit Einstein productivity was likely enough worse than the relevant Pascal cards to discourage me.  But that was at least two Einstein applications ago, and I don't think I had really good numbers.  I look forward to seeing them from you.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109386046813
RAC: 35920001

wes. wrote:Gary Roberts

wes. wrote:
Gary Roberts wrote:

Of course, if you were to spend a little more you could get something like an RX 480 - a lot cheaper in your country than in mine. I saw an RX 480 on newegg for just 160 bucks after rebate.

 

Just bought one...Will post timings...It's all your fault Gary. Wink

You're most welcome!! :-).   I was just reading a comparison of the 1060 vs the 480 just updated in December which seems to imply significant driver improvements by AMD over the previous months.  I know this is all about gaming but maybe it flows over into crunching as well.  Looking at the prices quoted in the review, the newegg price seems pretty decent.

I look forward to hearing what sort of performance you get.  If you happen to have a killawatt style power meter, it would be nice to know how much power the system uses, both at idle and full load, and with and without the card in place.  Not running any CPU tasks for all the measurements should allow the performance of the card itself to be worked out.

 

Cheers,
Gary.

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109386046813
RAC: 35920001

archae86 wrote:Gary, I'd be

archae86 wrote:
Gary, I'd be really interested in both the productivity and the power consumption comparisons, specifically on Einstein work.

You can rest assured that I'll do just that when I get a card and get it properly set up.  I know how important it is to test precisely with a particular app version and with different mixes (including zero) of CPU tasks and GPU tasks.

There are a couple of things I need to do first.  I don't think my distro yet supports the amdgpu-pro (the proprietary driver I would like to use for OpenCL support) out of the box.  I will need to investigate this.  I could just install Ubuntu on a test machine simply for the purpose of gathering the data but I'll probably try to get things running on my distro of choice first.

 

Cheers,
Gary.

Phil
Phil
Joined: 8 Jun 14
Posts: 579
Credit: 228493502
RAC: 0

@GaryIf memory serves (it's

@Gary

If memory serves (it's been awhile) you are using PCLinux? Or was it Arch? I'm considering moving my Windows machines back to Linux. I didn't have time to wrap my head around it last time, but I may have enough knowledge to get away with it now.

I was using Linux Mint in the past but I really want a leaner distro with no GUI. I use Nvidia and want to swap my cards around and use a Linux distro to see if I can squeeze out some more performance.

Phil

 

Gary Roberts
Gary Roberts
Moderator
Joined: 9 Feb 05
Posts: 5842
Credit: 109386046813
RAC: 35920001

Phil_58 wrote:If memory

Phil_58 wrote:
If memory serves (it's been awhile) you are using PCLinux?

Yes, its proper name is PCLinuxOS.  I wouldn't call it all that lean - I use a full KDE desktop - but in a couple of cases I've run CPU tasks without loading the GUI and the crunch times are essentially the same.  If you have enough RAM, the convenience of a GUI completely outweighs any small saving of crunch time.

Phil wrote:
I was using Linux Mint in the past but I really want a leaner distro with no GUI. I use Nvidia and want to swap my cards around and use a Linux distro to see if I can squeeze out some more performance.

If you have enough physical RAM to avoid swapping, you wont really sacrifice performance by using a GUI.

Good luck with whatever distro you choose to use.  The most important part is to choose one that you really feel comfortable with.

 

Cheers,
Gary.

Phil
Phil
Joined: 8 Jun 14
Posts: 579
Credit: 228493502
RAC: 0

@Gary Ok, thanks for the

@Gary

Ok, thanks for the info. I figured the GUI was dragging it down a bit, so I'll drop that part for now.

I'll swap the cards and see what happens to completion times.

Phil

 

Zalster
Zalster
Joined: 26 Nov 13
Posts: 3117
Credit: 4050672230
RAC: 0

wes. wrote:Gary Roberts

wes. wrote:
Gary Roberts wrote:

Of course, if you were to spend a little more you could get something like an RX 480 - a lot cheaper in your country than in mine. I saw an RX 480 on newegg for just 160 bucks after rebate.

 

Just bought one...Will post timings...It's all your fault Gary. Wink

 

Not to rain on anyone's parade but Gary totally ignored the machine at positions 2,3,4....

And that fact that the #1 machine is running 8 AMD cards compared to the 4 nvidia in those other machines.  Those 3 machines each are producing 2 million a day. But I can understand why he choose to ignore them as they have 980Tis in them. Since cost is an issue, looking at machine #5, it has 6 AMD cards and produces less than the 3 above it. 

So figuring an average cost in USD of $218 that means 6 AMD cards would be $1308. If he added 1 more card, it would probably bring that machine into the 2 million a day mark and cost $1526.

The 8 AMD card could cost you $1744 (if you had enough slot for them, my guess..risers)

But again, you could say it's still cheaper than 4 Nvidia 980Tis except, there is a 4th machine I have that has 4 1070s in it and also is a 2 million a day producer (it just hasn't made it's way up the list of RAC top cruchers yet)

An EVGA 1070 cost $380 USD.   4 of them cost $1520 total for a 2 million a day cruncher. That places it almost exactly the same as a 7 AMD machine.

4 cards vs 7 or 8  

True you have to deal with the work units using 1 CPU per work unit, but then..I don't run CPU task so it really isn't an issue and by the way, how much is a CPU task really going to add to a 2 million a day GPU cruncher anyway?

Just showing the other side of the coin

 

Zalster

 

Edit..

Sorry, looks like that 6 AMD machine is now slot #7. Didn't check it's position and realized it was replaced by 2 other machines when I posted.

emoga
emoga
Joined: 19 May 14
Posts: 15
Credit: 649338082
RAC: 0

Zalster wrote: looking at

Zalster wrote:
looking at machine #5, it has 6 AMD cards and produces less than the 3 above it.

Correct. (4) GTX 980 Ti's outproduce (6) R9 290x's (AMD Hawaii).  I should hope so since the 290x's were release on October 25, 2013.

Quote:
So figuring an average cost in USD of $218 that means 6 AMD cards...

Where are you getting this number for 290(x)'s?  You can't buy them brand new from any major retailer that I looked at.  Amazon and ebay have them listed around @300 for used ones which is a terrible deal.  

I remember buying mine years ago at a much higher cost though Money Mouth @500 CAD?  and being disappointed that they weren't better at DP than my R9 280x's. Cry ( I was crunching Milkyway at the time)

Quote:
The 8 AMD card could cost you $1744...

Those are R9 Fury's (AMD Fiji ).  The cheapest ones on newegg are $265 w/rebate = $2120

RAC: 2,896,620.89 / 8 cards = 362,078 per card. Not the best at crunching Einstein per USD. It also has a TDP of 275 Watts Surprised

Quote:
An EVGA 1070 cost $380 USD.   4 of them cost $1520 total for a 2 million a day cruncher.

Cool. Cool

500,000 points / $370 USD w/rebate on newegg = 1351 points per USD

What Gary is talking about is a RX 480. (AMD Radeon (TM) RX 480 Graphics)

stoneageman's 480's (on average between 4 hosts) are producing 427,012 RAC per card.(at this time)

The cheapest RX 480 on newegg is $170 (427,012  ppd)= 2512 points per USD

2512 (RX 480) / 1351 (GTX 1070) = 1.86 times 

Of course this could all change next app or even when the RAC finalizes.  But at this time, the RX 480's are 1.86 times better at crunching einstein (point per $) than GTX 1070's.

Zalster
Zalster
Joined: 26 Nov 13
Posts: 3117
Credit: 4050672230
RAC: 0

Quote:Of course this could

Quote:
Of course this could all change next app or even when the RAC finalizes.  But at this time, the RX 480's are 1.86 times better at crunching einstein (point per $) than GTX 1070's.

Yes, if my machines were static and not moving from project to project.  But I like to use them for other ones as well. For that reason, they are much better deal for the $ I spend on them than AMD.   AMD are fine for Einstein but not so great for other projects.

Stoneageman's RAC for each machine will probably close to 1.2 million for each machine.

The #1 machine produces 4.2-4.4 Million a day or exactly twice what the #2 machine does. So in any case, he would still require 3 machines with 4 GPUS of either design to reach his 6 million a day mark.

DanNeely
DanNeely
Joined: 4 Sep 05
Posts: 1364
Credit: 3562358667
RAC: 1580

Gary Roberts wrote:DanNeely

Gary Roberts wrote:
DanNeely wrote:
... on the nVidia side they're still eating a full CPU core each.

Yes, exactly.  On the older ones I've tried - 550Ti, 650 and 750Ti, the CPU component is virtually equal to the full elapsed time.  I don't know if that's also true for the latest and greatest in the NVIDIA range - I don't own any and I haven't bothered to check because they're too expensive for me.  I'm more interested in something a bit less costly like the RX 480 (or 470 or 460) where an elapsed time of around 1340s only requires 96s of CPU support.  I haven't done any real research yet but I know it's cheaper for me to by an RX 480 than a GTX 1060 and I believe it will give a greater output as well.  Once I've checked that, I'll probably buy one to measure output against power consumption.

 

I just finished a 1-4 at a time test on my W10 4790K + GTX1080 system (results to be posted soon).  It takes most of a core; and the difference between what it takes and 100% could easily be dominated by the usage of other applications on the system (it's my primary PC).  I'm planning to do 2 vs 3x at a time on a slightly slower system with a 980; and will be bumping the clocks by ~10% on my 1080 box (forgot to bring them back up after fixing a watercooling problem that had me on air for a while).  Those results should help shed a bit of light on it.

 

Gary Roberts wrote:
I think it's quite fortunate that AMD cards that need to compete on price just happen to have the better OpenCL performance (or so it seems).  It seems to be a win/win for non-gamers who just want to crunch :-).

Great for you anyway. Just make sure you're factoring power into the total cost of ownership, AMDs lcards are price  competitive with nVidia in the low and middle end, at a cost of running a lot hotter.  For something that runs 24/7/365 the rule of thumb for estimating power cost is that each cent/kilowatthour you pay costs 10c for each watt what you're running uses (ignoring any free space heater in winter or extra AC in summer factors).  Depending on exact models and what you're paying that cheap AMD card might cost an extra $50-200/year in electricity.

OTOH I'm a gamer who crunches on my current and older hardware.  The last few generations have seen nVidia competitive to significantly faster than AMD; and with nVidia having a decent perf/watt advantage even when gaming perf is a bit of a wash.  I hope AMD gets something competitive in the high end segment of the gaming market soon; the 480's done better for them than the previous generations have; but the high end cards are where all the profit is.  They need that to fund their R&D.

 

 

Comment viewing options

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