... is anyone running a pair of 4870s (non-crossfire) or a 4870X2 with the GPU App?

If so - any benefits?

P.

... is anyone running a pair of 4870s (non-crossfire) or a 4870X2 with the GPU App?

If so - any benefits?

P.

I've heard of one person using a 4870x2, but so far the optimized app is only making use of one side. I think it's likely that someone, somewhere is testing an app that will make use of both sides, and maybe even crossfire. If so, I hope they tell me when they do and get it working so that I can add it to zslip ;)

Not heard of anyone using crossfire yet, or even two cards (non-crossfire) in the same box.

---

0.19d: 26th Feb 2009
- multi GPU support (still experimental, can be turned off by exclusion of all but one GPUs in
app_info.xml)
- release of the CPU during GPU calculations, CPU load down to ~10% of one core or less

0.19c: 25th Feb 2009
- multi GPU and CPU release test, limited publication
- added support for command line options (GPU choice, CPU release, max number of WUs in flight)

0.19d: 26th Feb 2009
- multi GPU support (still experimental, can be turned off by exclusion of all but one GPUs in
app_info.xml)
- release of the CPU during GPU calculations, CPU load down to ~10% of one core or less

0.19c: 25th Feb 2009
- multi GPU and CPU release test, limited publication
- added support for command line options (GPU choice, CPU release, max number of WUs in flight)

For what it's worth, I ran with 0.19d for a few days on an HD4850 and a HD4870 with no problems, not that I needed the multi GPU support, but reverted back to 0.19b yesterday after which I managed to get MW crunching alongside CPDN again.

[edit]Now that 0.19d has been announced, I can tell you that the sub-one minute WU in my profile was crunched with 0.19d, and I will put the "d" back in it accordingly.

---

I've heard of one person using a 4870x2

That would be me. Status quo is that both GPUs are used by version 0.19d, but it seems unable to get the combined load past 50% (putting the x2 on par with a single GPU card - at best).

I'm just guessing now, but I suspect that two singles in crossfire ought to behave in a similar fashion to one x2. (Two singles would have the better cooling, of course.)

A little patience, and I think we'll see more of Cluster Physik's genius :)

---

I've heard of one person using a 4870x2

That would be me.

Yes I know. I'm forever behind you chasing you up the top 20 now ;)

---

I have just been speaking with someone on the PC Pro (magazine) overclocking forums about mixing NV and ATI graphics cards.

It seems that if you follow Ice's advice on ZSlip - get the drivers, power down, install the ATI card, reboot and load ATI drivers - the normal graphics will continue to work with the NV card (in my case). That leaves the ATI card for crunching, like Ice is doing so successfully.

I then asked about installing 2 x HD4850s (non-crossfire) in the same box, and the consensus was that should be OK, assuming the PSU was up to the load demand.

I would assume that PCs with 3 graphics cards - 1 NV or ATI for normal graphics, and 2 ATI for MW would need a PSU rated at least 750W.

Anyway, when I get my sick quad back, and the wallet can stand it I will try some experimentation.

I have a second HD3850 512Mb that'll arrive tomorrow to go into my Q6600 box. At $64 USD it was an easy call. I'll report back on how that works out in a few days..

-jim

I have a second HD3850 512Mb that'll arrive tomorrow to go into my Q6600 box. At $64 USD it was an easy call. I'll report back on how that works out in a few days..

-jim

That will be interesting, and show a 3 graphics card set up will work.

NOTE: I am assuming you are adding the 2 x HD3850 to a PC with an existing graphics card?

I have a second HD3850 512Mb that'll arrive tomorrow to go into my Q6600 box. At $64 USD it was an easy call. I'll report back on how that works out in a few days..

-jim

That will be interesting, and show a 3 graphics card set up will work.

NOTE: I am assuming you are adding the 2 x HD3850 to a PC with an existing graphics card?

The motherboard is a Gigabyte GA-X38T-DQ6 with no discrete onboard graphics, so no. Since the box only gets used sporadically for load testing I couldn't give a whit about response or performance when unattended. That being said, it makes a PII with a #9 imagine 32 look like an F1 car if I try and do anything while it's crunching, mostly from a video refresh standpoint.

As it crunches 8 wu's at an indicated 25% cpu per with one card, I'll be looking for the same number with (hopefully) 50% per wu cpu.

I'll report back, though probably not before late Monday.

-jim

I have a second HD3850 512Mb that'll arrive tomorrow to go into my Q6600 box. At $64 USD it was an easy call. I'll report back on how that works out in a few days..

-jim

That will be interesting, and show a 3 graphics card set up will work.

NOTE: I am assuming you are adding the 2 x HD3850 to a PC with an existing graphics card?

The motherboard is a Gigabyte GA-X38T-DQ6 with no discrete onboard graphics, so no. Since the box only gets used sporadically for load testing I couldn't give a whit about response or performance when unattended. That being said, it makes a PII with a #9 imagine 32 look like an F1 car if I try and do anything while it's crunching, mostly from a video refresh standpoint.

As it crunches 8 wu's at an indicated 25% cpu per with one card, I'll be looking for the same number with (hopefully) 50% per wu cpu.

I'll report back, though probably not before late Monday.

-jim

Do you have any thoughts or feel for the likely output of your HD3850?

The Gas Giant suggested his HD4850 would eventually give an RAC of 50,000. So, an indication of these different cards' (HD 3800 and 4800) potential would help fellow crunchers planning GPU work and on a budget.

I.E. Where does the value for money i crunching output lie? Especially if 2 x HD3850s give more output than 1 x HD4870.

Do you have any thoughts or feel for the likely output of your HD3850?

The Gas Giant suggested his HD4850 would eventually give an RAC of 50,000. So, an indication of these different cards' (HD 3800 and 4800) potential would help fellow crunchers planning GPU work and on a budget.

I.E. Where does the value for money i crunching output lie? Especially if 2 x HD3850s give more output than 1 x HD4870.

I do have a feel for what a HD3850 can do, which is why I bought an identical one on the cheap and consider it a sound BOINC investment.



The Q6600 was running (on .19b optimised) at ~8000 WU per day up until 2/26 I switched to the ATI GPU app. It took a few days to settle, but it's now putting out ~26k per day. The gain is impressive, hence giving it a shot with a second GPU..

This is the video card I've been using: http://www.newegg.com/Product/Product.aspx?Item=N82E16814102715

-jim

NOTE: These figures are all with stock clocks, both CPU and GPU..

I don't know if this helps you much, but my HD4870 PC is on an average credit of 43,760 - the HD4850 is 37,960

They are not identical PCs running for identical times etc, but the HD4850 card was fitted first and the output from the HD4870 is higher even though it started later.

---

So, an indication of these different cards' (HD 3800 and 4800) potential would help fellow crunchers planning GPU work and on a budget.

I.E. Where does the value for money i crunching output lie? Especially if 2 x HD3850s give more output than 1 x HD4870.

I've posted something about the relative speeds some time ago.

To quote it

When neglecting the 1-2 seconds CPU calculation for every WU (which is justified if you have two or more concurrent WUs running), one can break it down to the following performance relations (taking the HD3850 as base):

3850 : 3870 : 4830 : 4850 : 4870
  1  : 1.16 : 1.72 : 2.33 : 2.80

Of course the ratios only applies for stock clocks (670MHz for 3850, 775MHz for 3870, 575MHz for 4830, 625MHz for 4850 and 750MHz for 4870 with the memory speed only a very minor factor).

I guess that a 4850 is giving most bang for the buck (don't know how the latest prie cuts influence this).

My biggest problem is keeping the GPU working. :(

And sadly, UCB seems to be slow rolling the issue of the ATI GPU ...

My biggest problem is keeping the GPU working. :(

And sadly, UCB seems to be slow rolling the issue of the ATI GPU ...

I don't think I care much about what UCB smoke slowly rolled.

I'm just glad that the people who worked hard to bring optimization to MilkyWay, GPU or otherwise, didn't smoke that stuff :P

---

My biggest problem is keeping the GPU working. :(

And sadly, UCB seems to be slow rolling the issue of the ATI GPU ...

I don't think I care much about what UCB smoke slowly rolled.

I'm just glad that the people who worked hard to bring optimization to MilkyWay, GPU or otherwise, didn't smoke that stuff :P

Seems to be the story to this point, sadly, the projects are way ahead of the curve. Either through internal development (GPU Grid) or volunteer work (SaH and Milky Way (Yea CP! and whomever else is helping)) ...

... is anyone running a pair of 4870s (non-crossfire) or a 4870X2 with the GPU App?

If so - any benefits?

P.

I just immigrated here yesterday, I've bought an 4870x2 after my gtx 280 burned up while running gpugrid for 6 months non stop.

Windows 7, latest catalyst driver in Visa compatibility mode, running like a charm.

GPU-Z as well as catylst control center reports 90% GPU usage while running 4 simoultanous tasks. I might bump it up to 8 later today but the problem is the server's new tasks queue is not generous at the moment.

Currently my tasks history shows a 5 minute interval between each 6 new tasks assignments. 4 tasks will complete between 3 to 10 seconds.

I need more time to settle in here and figure out this project and hopefully I wont burn this card like I did with my GTX 280. :P

... is anyone running a pair of 4870s (non-crossfire) or a 4870X2 with the GPU App?

If so - any benefits?

P.

UL1, the current top cruncher, has 3 way 4870x2.

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=23348757

... is anyone running a pair of 4870s (non-crossfire) or a 4870X2 with the GPU App?

If so - any benefits?

P.

UL1, the current top cruncher, has 3 way 4870x2.

http://milkyway.cs.rpi.edu/milkyway/result.php?resultid=23348757