• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

Something to consider (ATI)

K

Kendrak

[H]ard|DCer of the Year 2009
Joined
Aug 29, 2001
Messages
21,141
The 4850 X2:
When I first tested (and published) this article, I wasn't certain that my results were accurate because of a very low power draw reading. After a full week of testing and re-testing, I can now be certain that my results were accurate all along. To my surprise, the Sapphire Radeon HD 4850 X2 exhibited such a low consumption at idle that it actually measured slightly less than a single Radeon HD 4850. But that's not the real news, because once I confirmed the results of my testing under full load this product stepped out and demanded my attention. At a mere 160W of VGA power consumption under load, the Sapphire Radeon HD 4850 X2 actually consumes 10W less power than the GeForce 9800 GX2. This might not mean anything to you yet, but consider this: it also consumes 83W less power than a single Radeon 4870, and a full 160W less than the 4870 X2 under load. That's very impressive energy efficiency in my book, and something that Sapphire should be quite proud of.
Click to expand...

http://benchmarkreviews.com/index.php?option=com_content&task=view&id=244&Itemid=72&limit=1&limitstart=10

Now if one was to make a rig with 2 of these...... and with the impending lambda WU. AND a nice price drop ($330 AR at newegg)

Things are getting interesting. (Ohh and as a bonus, It probably could run Crysis:p)

 
What kind of ppd are those 4850's doing now?


 
Correct me if im wrong but isnt a 4850 a slowr clocked 4870 with ddr3 isntead of ddr5?
 
That is correct, sl1982. 4850 is the same GPU with the same shader count as a 4870. They are just clocked slower and use the GDDR3 memory. We all know that folding is a non-memory intensive load for a GPU so the 4850 is going to be the best for ppd/$ and also ppd/Watt. GDDR5 seems to be some power-hungry stuff. Whenever I get around to building my new HTPC/folding rig plural 4850X2s will be considered if the price is right.


 
The problem is that a 4850 will get around 2800 - 3200 ppd right now and a 9600gso gets around 2600 - 2800 when overclocked running a large 1392 atom protein.
In other words it's almost a wash ppd wise.
Of course some new aspects of the Dec. 10th ATI driver have not been implemented in folding@home yet, and who knows what that will bring. But unless the new driver and CAL cause the ppd to double or more nvidia still looks good for folding.
 
yes the 4850 is slower..

its factory downclocked but like flogge said.. its the exact same gpu and shader just gddr3 and downclocked.. this was something ati specificly did to save them money when it came to fabricating the waffers.. where as nvidia still relies on the old make everything 1 gpu and if it cant pass the test turn it into a slower version.. and thats where the gtx 280 and 260 stand.. but its actually starting to cost nvidia money by doing that..

but i believe what sapphire has done with the energy efficiency thing in idle is that the second core basicly goes in standby when idle.. that and the fact that amd gave all the card manufactures the freedom to modify what ever they wanted.. they are not forced to stick with the reference design like the older 3k series cards and nvidia cards.. so as time goes you will start seeing some much more efficient cards and some crazier designs like the 4850x2 which is not an ATI reference design.. its basicly just replacing the 4870x2 memory chips with gddr3.. since the gpu's are identical.. much more cost effective and a brilliant design because it will be a quad gpu/crossfire design that is much more suited for people using 1920x1200 resolutions to run that setup..

so my hats off to sapphire and ATI for doing this..

i posted an article about ATI and the RV770 gpu that Anand from anandtech.com did.. its its a good article and makes to relize just how nvidia and ati really work and how they trying to stay one step ahead of eachother.. also makes you relize how screwed nvidia really is with their business design compared to ATI's..

http://www.hardforum.com/showthread.php?t=1372953
 
I'm looking forward to new, much much larger WU. The drop in NV cards might not be done. And the ATI cards, at the ppd that you stated, are only using about 1/2 of their shaders.
 
Isn't it also going to take a relatively fast cpu to feed all those shaders?
 
Kendrak said:
I'm looking forward to new, much much larger WU. The drop in NV cards might not be done. And the ATI cards, at the ppd that you stated, are only using about 1/2 of their shaders.
Click to expand...

Larger WU and better use of the ATI shaders is great news. I do have an ATI 4850 and I would love to see it better put to use.
 
tkam said:
Isn't it also going to take a relatively fast cpu to feed all those shaders?
Click to expand...


no because the gpu client isnt cpu dependent.. so it the program wont be cpu bottlenecked.. the client only requires the gpu and shader.. thus you can pretty much get the same ppd with a crappy 1ghz processor with a nvidia 8/9/200 or ati hd series card as you would with a 4x4 16 core system running the same gfx card..
 
sirmonkey1985 said:
no because the gpu client isnt cpu dependent.. so it the program wont be cpu bottlenecked.. the client only requires the gpu and shader.. thus you can pretty much get the same ppd with a crappy 1ghz processor with a nvidia 8/9/200 or ati hd series card as you would with a 4x4 16 core system running the same gfx card..
Click to expand...

That's also not true. CPU speed has a distinct effect on GPU client performance. I had to isolate one of the cores of my CPU from the SMP client and allocate it for the GPU client to get it to run at full speed. A slower chip or dual-core would not have enough speed to run an instance of the SMP client concurrently with a GPU client at full speed. Even with future improvements in client efficiency and CPU usage, a certain amount of CPU speed is still necessary in order to not be bottlenecking anything.
 
I hit the wall on CPU feeding Vid card in my main folding box.

I found out that a E2140 is NOT enough to feed 3X 9800GX2 cards (6 cores)
It was plenty to feed 2X 9800GX2 cards, now a E4500 does the job just fine.

Now I know that ATI cards are a bit more CPU hungry (and from what I understand they are working on that, (Evil might be able to fill us in more on that) But I would think that a E4500 or above, should be plenty to feed 2 of these cards.
 
Kendrak said:
Now I know that ATI cards are a bit more CPU hungry (and from what I understand they are working on that, (Evil might be able to fill us in more on that) But I would think that a E4500 or above, should be plenty to feed 2 of these cards.
Click to expand...

I doubt it. My 4870 GPU client takes around 40-60% of one of my cores, which is running at 3.8GHz. A stock E4500 at 2.2GHz would be able to run maybe three clients at most across both cores, but not four, at least not with the current client's CPU usage level. The new CAL DLLs that Evil keeps talking about are supposed to reduce CPU usage a lot, but I'll believe it when I see it, and that also depends on when the final ones are actually released.

Folding performance in general is alright with current ATI cards (currently hitting ~3000PPD with large proteins, was around 3700-3800 before with smaller ones), but CPU usage is really the big issue now, especially for budget multi-GPU boxen to be a viable option.
 
Zero82z said:
IThe new CAL DLLs that Evil keeps talking about are supposed to reduce CPU usage a lot, but I'll believe it when I see it, and that also depends on when the final ones are actually released.
Click to expand...
They have been released for public testing on November 11, 2008. CAL 1.2.1
http://foldingforum.org/viewtopic.php?f=51&t=6847

(This beta users Client 6.20 RC3, but the new client is 6.23.)
(I suggest you download the beta with the new dll's and the update the client part to 6.23)

Cal 1.3 was released with the Catalyst 8.12 drives, but F@H will not use them until some more tests can be completed to insure there is no issues.

Also, new core will be released in a few days that will help ppd slightly.

I have 4 GPU's running on a AMD 9350e (Quad - 2.0Ghz) so CPU usage is not an issue for me. I average 17-20% CPU usage per client.
 
EvilAlchemist said:
Also, new core will be released in a few days that will help ppd slightly..
Click to expand...
That's good to know because my sole Ati client stopped working a few days ago. Every time I start it I get a message stating that the core crashed. I am now forced to run the console client on that machine until I resolve the issue. :(

 
EvilAlchemist said:
They have been released for public testing on November 11, 2008. CAL 1.2.1
http://foldingforum.org/viewtopic.php?f=51&t=6847

(This beta users Client 6.20 RC3, but the new client is 6.23.)
(I suggest you download the beta with the new dll's and the update the client part to 6.23)

Cal 1.3 was released with the Catalyst 8.12 drives, but F@H will not use them until some more tests can be completed to insure there is no issues.
Click to expand...

I will definitely try that.
 
When their avg PPD catches up to the GTX280, I'll consider building a box around them.
 
EvilAlchemist said:
They have been released for public testing on November 11, 2008. CAL 1.2.1
http://foldingforum.org/viewtopic.php?f=51&t=6847

(This beta users Client 6.20 RC3, but the new client is 6.23.)
(I suggest you download the beta with the new dll's and the update the client part to 6.23)

Cal 1.3 was released with the Catalyst 8.12 drives, but F@H will not use them until some more tests can be completed to insure there is no issues.
Click to expand...

Zero82z said:
I will definitely try that.
Click to expand...

Turns out I was already using the 1.2.1 CAL DLLs. I guess I'll just wait for the updated core then.
 
I think it will be mid-next week (Dec 15-17) before they get this ATI core out to the public.

I feel that it is ready now, but the team wants to make 100% sure there are no problems this weekend.
(There has been 0 problems yet which is a good thing)
 
EvilAlchemist said:
I feel that it is ready now, but the team wants to make 100% sure there are no problems this weekend.
(There has been 0 problems yet which is a good thing)
Click to expand...

Definitely a good thing. Problem-free cores have been somewhat of a rarity lately ;).
 
More importantly what are the PPD results.

 
My gains so far on a 4870 are around 200PPD on the 384 pointers and around 100 on the 511s


 
I estimate an increase of about 500PPD across the board, or in other words, my output is up to about 200PPD less than I was getting with the old small proteins. On the flip side, CPU usage for the client is back up to 100% of one of my cores, as opposed to about 40-50% with 1.18b.
 
Zero82z said:
I estimate an increase of about 500PPD across the board, or in other words, my output is up to about 200PPD less than I was getting with the old small proteins. On the flip side, CPU usage for the client is back up to 100% of one of my cores, as opposed to about 40-50% with 1.18b.
Click to expand...
Unless it's a small amount, I haven't noticed an increase in performance from what I can see but I'm running a 3850.
 
Zero82z said:
That could be why. It's definitely performing a lot faster for me.
Click to expand...
I've only seen the 384 pointers so far and if they are the lower PPD WUs, then I'm seeing a modest increase of about 100 PPD. I won't know for sure until I get more variety of WUs.
 
I think most of the boost that we are hoping for on the 4850/4870 is the utilization of all of the extra shaders those cards have.
 
There should be small performance increases for all boards, but you may only notice it with the larger WUs. This core update does utilize the shaders on the 48XX boards better, but the small proteins are still fighting me a little and we are continuing to work on that.
 
Kendrak said:
I think most of the boost that we are hoping for on the 4850/4870 is the utilization of all of the extra shaders those cards have.
APOLLO said:
I've only seen the 384 pointers so far and if they are the lower PPD WUs, then I'm seeing a modest increase of about 100 PPD. I won't know for sure until I get more variety of WUs.
Click to expand...
Click to expand...
After a couple more days running the new core I noticed a 300PPD increase on other WUs I received. Although these WUs seem to be in the minority, this puts my HD3850 to around 2500PPD max so far. Considering that my 8800GT cards can drop to as low as 3500PPD, this was unexpected. :cool:
 
You must log in or register to reply here.
Back
Top