I expect this must be the next way of getting more number crunching. I've been reading around and it seems I can push my HD4850 up from 625MHz to maybe as much as 690MHz. Has anyone tried this?

---

Your HD4850 only runs at 625MHz? Wow, my HD3850 does 669MHz with 829MHz on the memory.

---

Jord.

The BOINC FAQ Service.

I used the CCC "Auto Tune" button. It reported everything passed. Then about 1 in 4 WUs started errored out. I set it back to stock for now. Maybe if I have nothing to do for a whole weekend, I'll slowly increase the gpu speed and keep monitoring closely.

Since the app isn't memory intensive, increasing the memory speed probably won't help much here.

I used the CCC "Auto Tune" button. It reported everything passed. Then about 1 in 4 WUs started errored out. I set it back to stock for now. Maybe if I have nothing to do for a whole weekend, I'll slowly increase the gpu speed and keep monitoring closely.

Yes, in a game one or two pixels with a wrong color won't kill you, but for GPGPU computing that is nothing you want. I would not rely on that short autotune test either.

Your HD4850 only runs at 625MHz? Wow, my HD3850 does 669MHz with 829MHz on the memory.

Now I feel inadequate :( But then the HD3850 has 320 Stream Processing Units as oppossed to 800 for the HD4850 ;)

---

I expect this must be the next way of getting more number crunching. I've been reading around and it seems I can push my HD4850 up from 625MHz to maybe as much as 690MHz. Has anyone tried this?

I've overclocked my HD4870 just slightly, and it seems to be running fine. What I did was use ATITool's scan for artifacts tester, and set the sensitivity to the highest level. Then I gradually pushed up the core and tested until an error was reported. Then I did the same with the memory. I reached a final overclock of 780/990, up from 750/900 stock.

---

Your HD4850 only runs at 625MHz? Wow, my HD3850 does 669MHz with 829MHz on the memory.

I'm clocking my HD4850 at 690MHz and 1183MHz on memory.

[Edit] No I'm not. It just crashed. Just a little tweak needed I expect...

---

Your HD4850 only runs at 625MHz? Wow, my HD3850 does 669MHz with 829MHz on the memory.

I'm clocking my HD4850 at 690MHz and 1183MHz on memory.

[Edit] No I'm not. It just crashed. Just a little tweak needed I expect...

You tweakers and your crashes...

---

me@rescam.org

Your HD4850 only runs at 625MHz? Wow, my HD3850 does 669MHz with 829MHz on the memory.

I'm clocking my HD4850 at 690MHz and 1183MHz on memory.

[Edit] No I'm not. It just crashed. Just a little tweak needed I expect...

You tweakers and your crashes...

It's tweaked down to 685MHz and 1050Mhz and behaving itself right now. But even if it runs OK at these settings I will probably notch it down a little more. Why stress it when the tasks it is churning out are more than enough? It's been interesting trying to clock it, and I'm interested to hear if others have tried it.

---

It's been interesting trying to clock it, and I'm interested to hear if others have tried it.

NO ... :)

And aside from one set of 8 tasks that died during a restart I have had no failures either ...

One GPU, out-of-the-box and into-the-box and 6-14 seconds per task ... geeze, what more is needed?

That one card is making about 50K something a day ... it might go for more when the work issue and detection/management issues are fixed so that it always does work ...

Hmmm ... just rebooted and trashed all the ATI MW tasks I had on hand ... Houston, we may have a problem ...

[One GPU, out-of-the-box and into-the-box and 6-14 seconds per task ... geeze, what more is needed?

Ha ha, yes I was thinking that, wondering if the 12 seconds might come down to 11 or 10.... then, what the heck - 12 seconds is fast enough! ;)

But then where would we be without a challenge? I'm looking forward to seeing what else Cluster Physik and speedimic can do with their magic :)

---

Ha ha, yes I was thinking that, wondering if the 12 seconds might come down to 11 or 10.... then, what the heck - 12 seconds is fast enough! ;)

But then where would we be without a challenge?

You want a challenge?

I owe a beer to that nvidia developer who produces a CUDA version that runs faster on a brand new dual GPU GTX295 card than my version on a single GPU HD4870.
I first thought about offering the same for a comparison to a mainstream HD4850, but in that case there would be a (slim) chance ;)

Ha ha, yes I was thinking that, wondering if the 12 seconds might come down to 11 or 10.... then, what the heck - 12 seconds is fast enough! ;)

But then where would we be without a challenge?

You want a challenge?

I owe a beer to that nvidia developer who produces a CUDA version that runs faster on a brand new dual GPU GTX295 card than my version on a single GPU HD4870.
I first thought about offering the same for a comparison to a mainstream HD4850, but in that case there would be a (slim) chance ;)

Cluster Physik, I don't know much about nvidia and CUDA, except that nvidia is slowly giving me some credits on a laptop. I've also heard about some who have some nvidia going spare now that they have taken them out and replaced with ATI. If I could take out my on-board nvidia and replace with ATI I certainly would, but I just don't have any solder for my soldering iron right now :P

---

Ha ha, yes I was thinking that, wondering if the 12 seconds might come down to 11 or 10.... then, what the heck - 12 seconds is fast enough! ;)

But then where would we be without a challenge?

You want a challenge?

I owe a beer to that nvidia developer who produces a CUDA version that runs faster on a brand new dual GPU GTX295 card than my version on a single GPU HD4870.
I first thought about offering the same for a comparison to a mainstream HD4850, but in that case there would be a (slim) chance ;)

So overall which do you think is the faster card - the HD4870x2 or the GTX295?

---

me@rescam.org

I owe a beer to that nvidia developer who produces a CUDA version that runs faster on a brand new dual GPU GTX295 card than my version on a single GPU HD4870.
I first thought about offering the same for a comparison to a mainstream HD4850, but in that case there would be a (slim) chance ;)

So overall which do you think is the faster card - the HD4870x2 or the GTX295?

At MW? The Radeon wins hands down.
A single HD4870 already does practically the same DP flops here at MW as a GTX295 has as theoretical peak performance.

So overall which do you think is the faster card - the HD4870x2 or the GTX295?

At GPU Grid: GTX 295
At MW: HD4870

Something about no programs for the alternative cards ... :)

Credit award wise, one HD4870 card I have is returning about the same CS per hour as GTX 295s, one GTX 280 and one 9800GT ... which one can argue that either MW is over paying, or that GPU Grid is underpaying as they admit that they are ...

I owe a beer to that nvidia developer who produces a CUDA version that runs faster on a brand new dual GPU GTX295 card than my version on a single GPU HD4870.
I first thought about offering the same for a comparison to a mainstream HD4850, but in that case there would be a (slim) chance ;)

So overall which do you think is the faster card - the HD4870x2 or the GTX295?

At MW? The Radeon wins hands down.
A single HD4870 already does practically the same DP flops here at MW as a GTX295 has as theoretical peak performance.

At GPU Grid: GTX 295
At MW: HD4870

Something about no programs for the alternative cards ... :)

Credit award wise, one HD4870 card I have is returning about the same CS per hour as GTX 295s, one GTX 280 and one 9800GT ... which one can argue that either MW is over paying, or that GPU Grid is underpaying as they admit that they are ...

Actually I meant for overall performance, like gaming, for my new system build. I seem to be finding mixed results on the intraweb thingy.

btw, I was adjusting my system memory timings to lower the latencies. Somewhere amongst 4 BSOD and 2 lockups I lost MW from my BOINC projects tab. A reattach (and a new DL from zslip) got me back on board.

---

me@rescam.org

Actually I meant for overall performance, like gaming, for my new system build. I seem to be finding mixed results on the intraweb thingy.

Ah, well, I can't be any help as I had to give up that kind of gaming a few years back. The kind of games I do play I have gotten adequate performance out of a 9800GT ... :)

I would guess in my case the GTX 295 cards would come out on top for the reason that there is more memory on the card. The ATI I got has only 512 where the 295s have 1700 something ...

As for GPU Processing I suspect that the debate is going to rage for quite a bit until we can start to see more than on application that can be spanned over the cards ... something that only can be done with Folding@Home right now.

But I suspect that the game thing is also going to be "it depends" based on which game ...

Actually I meant for overall performance, like gaming, for my new system build. I seem to be finding mixed results on the intraweb thingy.

[...]
I would guess in my case the GTX 295 cards would come out on top for the reason that there is more memory on the card. The ATI I got has only 512 where the 295s have 1700 something ...
[...]
But I suspect that the game thing is also going to be "it depends" based on which game ...

I'd probably say the GTX 295 also, but for different reasons. The HD4870 was roughly equivalent to a GTX 260 in most games. The HD4870X2 is just to HD4870 chips on a single PCB. A GTX 295 is somewhere in between a GTX 260 and a GTX 280 (or 285, depending on what you go with), which puts one of it's GPUs as a little more powerful than a single HD4870. Get my drift? ;)

Also, the HD4870X2 doesn't have 512MB, it comes with 1GB standard. The HD4870 (single version) is the one that comes with 512MB standard.

---

I have a few Nvidia cards over on Seti..A 260 and a Tesla c1060.

When I first bought my 4850 I was completely amazed at it's crunching ability..
<---This rac is 100% ati *grins*

Well, couldn't wait to benchmark it. I figured it would be blazing fast. Well surprise, It doesn't really run benchmarks only ever so slightly faster than a 260.

So here is how it seesm to me. I'm a car guy so I tend to use car analogies...
{
Was wrenching on my Mustang today and hoped in the r32 for a parts run. As I shifted her from 2 to 3 at 12,000rpm a thought occurred to me..

The mustang would blow up over say 7k but she makes 400ft lbs at 2000. Both the Ford and Volkswagon are about 440 horse but they are very different animals. I can't say one is better as I love them both. Each excels in their own way.

Anyway, so this is my analogy, long winded and confusing as it may be:

The Nvidia card has a lot of torque and low revs, The ATI revs high but a short little stroke.

Remember: Horsepower keeps you going..but torque,
.it makes you go..

---