David,

As much as I want to and it is very tempting, I don't think I can justify it. Thanks for looking into it though.

I'll watch with great interest how the 5970s perform though!

Paul.

Typical, just as I was going to crunch some Zzzzzzz.

Finally, I get some WU's to test.
00:08:12 100% Collatz device 0
00:08:06 100% Collatz device 1

Hmmm... strange times....

on my stock HD5870 I have 6:39-6:53....

On MW I see that you make MW task in 105-110s.

My stock HD5870 make MW in 82-84s. (3 task runing at the same time - one task takes real: 4m:09s

---

A proud member of the Polish National Team

COME VISIT US at Polish National Team FORUM

Well that makes perfect sense. A 5970 is, essentially, two 5870's stuck together. Some resources are shared so I would expect a single 5870 to be faster. In the same way my GTX 260's are faster, on a single WU, than my GTX 295's.

I have also got the 5970's running full speed on all cores now by adding the crossfire bridge and enabling crossfire. So the throttling problem has gone away.

Interestingly, this is the exact opposite of what I do with my Nvidia cards, where they are NOT connected with SLI enabled.

---

David

What setting do you have in the app_info.xml ?

I just put 2x 5970 in a machine and am only able to run with <coproc>1</coproc>
as when I set it to 0.5 it stalls and sloooooooooows right down. Is this the throttling you were mentioning in your post?

I'll try 0.5 with the bridge on and see if that frees mine up but I am still interesting in your app_info settings.

Thanks ! :)

EDIT:

So it seems I cannot run more than 1 MW task per GPU. This is down from 2 per GPU on the 4850x2 that were in there. So i have gone from 8 concurrent tasks to 4. Now each task should be alot faster so I guess I will have to see after 24 hours how its playing out but right now my hourly RAC is the same as it was with the 2x 4850x2 with 2x 5970.

David

What setting do you have in the app_info.xml ?

I just put 2x 5970 in a machine and am only able to run with <coproc>1</coproc>
as when I set it to 0.5 it stalls and sloooooooooows right down. Is this the throttling you were mentioning in your post?

I'll try 0.5 with the bridge on and see if that frees mine up but I am still interesting in your app_info settings.

Thanks ! :)

EDIT:

So it seems I cannot run more than 1 MW task per GPU. This is down from 2 per GPU on the 4850x2 that were in there. So i have gone from 8 concurrent tasks to 4. Now each task should be alot faster so I guess I will have to see after 24 hours how its playing out but right now my hourly RAC is the same as it was with the 2x 4850x2 with 2x 5970.

I have 1 ATI 5970 crunching MW for about 20 hrs so far. only 2 WUs @ time in 1.40mins Temp only 78 degree C
Also have 1 4870x2 doing MW 2WUs per 3.13 mins.
so the ATI 5970 is more than twice the speed of a 4870x2.
May double up the 5970 in the future as that box has a 1000 watt PSU, Speedo case.
Ross

---

I lined the two with a crossfire cable and enabled, and off they went! The throttling was caused by GPU 0 being used for display only, and GPU 1 running two tasks at once.

Currently I am running only one task at a time, but was running two with the <count>.5</count> command, It was only a little faster but 4 degrees hotter running two at a time. I am now in the low 80's per card.

---

I'm currently running two HD 4870s which are toggling between Collatz and MilkyWay. Because of all of the changes I've had lately with my machine, as well as with MW, I can't give an exact figure for how much they would do in an average day but if I were to roughly guess it's between 130-180K credits depending on how much of the day I am not running BOINC. However, I have one of these on the way to replace those two space heaters. I should both improve performance and decrease load on the system as well as lower heat output significantly, and certainly noise too. I'm curious as to exactly how much that additional 25MHz will bring, likely not much but should prove to be an even better replacement for my 4870s. These are truly incredible cards when it comes to raw power and it will be interesting to see just how Fermi from Nvidia will behave when it comes to computation seeing as that's (seemingly) where they have focused their attention with it based on their announcements on the architecture and growing touting of CUDA.

The extra 25MHz won't bring anything to the party. The extra shader units, on the other hand, will bring heaps of grunt, and make a significant difference to your times.

---

....... TBH, it'll do just as well as th 'vanilla' 5870s - they clock quite well, certainly more than 25MHz. Of course, if you overclock at all your warrenty is technically void, but if I was overclocking, then I wouldn't be stopping at a 25MHz uplift. Personally, I wouldn't pay any extra for a card with this amount of 'OC' .... but that is just my $0.02 worth. You should see 200K/day from this card.

The way I figure it, the card was the only in stock at the time and with XFX's stance on overclocking (that is to say, they cover it under warranty) I figured why not? If nothing else the additional cost was negated through the inclusion of a game.

They ship Tetris with video cards nowadays? :p

---

Cheers,

PeterV

.

Little tip: If you take the memory clock down on the 5870 to 300mhz you can overclock the core to around 970mhz without increasing the vcore and still keep the temps at a decent level

Hi, it seams that my HD5970 makes WU in 97s. (1 WU per GPU), but it has 86C and fan goes to 44% witch is extrem loud.... OMG at 725Mhz.... so loud!

Hi, it seams that my HD5970 makes WU in 97s. (1 WU per GPU), but it has 86C and fan goes to 44% witch is extrem loud.... OMG at 725Mhz.... so loud!

---

A proud member of the Polish National Team

COME VISIT US at Polish National Team FORUM

Little tip: If you take the memory clock down on the 5870 to 300mhz you can overclock the core to around 970mhz without increasing the vcore and still keep the temps at a decent level

I tried something similar over at the Collatz Project with a Couple of 5850's, set them to just 400 Memory & 925 Core. The Wu's ran twice as slow doing that ...

I haven't run collatz, but on milkyway it doesnt slow the WU's down one bit. They take around 76 seconds to complete. They'd propably complete even faster than that but I run folding@home in vmware on my CPU

I haven't run collatz, but on milkyway it doesnt slow the WU's down one bit. They take around 76 seconds to complete. They'd propably complete even faster than that but I run folding@home in vmware on my CPU

Does it actually speed them up though doing that ???

Well the great and wonderful Catalyst 9.12 have landed.

And the difference? Nada. Still can only use 3 gpus. Greeeat.

Okay Sapphire sent me new BIOS files and I flashed both.

SUCCESS on MW but not Collatz so I'm staying here for sure :)

I have been running Milkyway@home on 5970 at stock frequencies. Within the optimized app, also everything at stock except the w factor which is 0.8. Both GPU loads are %99 and each wu take 95.5 seconds to complete. After my liquid cooling kits arrive, I'll post the oc values.