strongest performing milkyway@home AMD card

hello fellow crunchers

i am currently using a 6950 (with 6970 shaders) for milkyway at home. Running 24/7 it would total 269,000 credits/day.

what i would like to know is which card of AMD is the fastest in terms of milkyway crunching, because i cannot really get more information out of the statistics than the top computers running with AMD AMD Radeon HD 7870/7950/7970/R9 280X series (Tahiti) (3072MB) driver: 1.4.1848 OpenCL: 1.2

is it worth changing from a (practically) 6970 to a 7970 or is it more reasonable to get a r9 290?
could anybody give me concrete numbers, i.e. how long your card (and specify the card ofc) needs for MilkyWay@Home v1.02 (opencl_amd_ati) operations?

i am somewhat confused because i heard that the older cards may perform better than the newer ones?

my card is running with 1000MHz (standard clock 820) and it takes 34 (42 standard clocked) seconds to finish one wu.

appreciate your help.

khryl

You might wish to read through this thread:
http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=3551&postid=61794#61794

Since you only consider AMD/ATI cards you might take a look at this list I made:
ATI 4650 AGP-no 320
ATI 4670-no 320
ATI 5550-no 320
AMD 5770-no 800
AMD 5870-yes 1600
AMD 6770-no 800
AMD 6850-no 960
AMD 6870-no 1120
AMD 6950-yes 1408
AMD 6970-yes 1536
AMD 7750-yes 512
AMD 7770-yes 640
AMD 7790-yes 896
AMD 7850-yes 1024
AMD 7870-yes 1280
AMD 7950-yes 1792
AMD 7970-yes 2048
AMD 7990-yes 4096
AMD R7 260x 896
AMD R9 270x 1280
AMD R9 280x 2048
AMD R9 290x 2816


The first column is the card version, the yes/no means whether it will work on MilkyWay or not while the last column is the Shaders, ie mini cpu cores, on each card. The more Shaders on the card the faster it crunches, all other things being equal. Card speed, overclocking etc will all affect that, but you cannot overcome pure horsepower, and that is what the Shaders are. In your example you have a 6950 which has 1408 Shaders, while a 7970 has 2048 Shaders, meaning the 7970 IS faster.

I have a 7970, NOT overclocked at all and am seeing this:
27.06 3.00 pending MilkyWay@Home v1.02 (opencl_amd_ati)

So mine is 5 seconds faster per work unit than yours is, using your non oc'd numbers.

The second problem you talked about was:
"i am somewhat confused because i heard that the older cards may perform better than the newer ones?"

This refers to the Nvidia brand gpu cards as the newer ones dual-precision capabilities is diminished in the newer cards. MilkyWay requires dual-precision cards, which all, or nearly all, Nvidia cards have it built-in. That was the yes/no column in my list, not all AMD/ATI cards are dual precision!

thank you for your response!
really helped me a lot.

but:

I have a 7970, NOT overclocked at all and am seeing this:
27.06 3.00 pending MilkyWay@Home v1.02 (opencl_amd_ati)

So mine is 5 seconds faster per work unit than yours is, using your non oc'd numbers.

wait a second. unless my brain is lagging right now this is not quite true.

first off, i flashed my 6950, so it now has 1536 shaders, technically making it a 6970 for mw@home as i understand it.

your work unit would,according to your statement, take 30 seconds, while my card needs 42 seconds @standard clock (820MHz). The best I can reach is 34 seconds with 1010MHz clock. so in fact your card is 12 seconds faster than mine, is that correct? or did I not catch sth here

anyway, thank you for the list, helping me a lot. does this mean that there is no difference between taking a 7970 or a r9 280x, except for the price? is one of them more suitable for overclocking than the other?

thank you for your response!
really helped me a lot.

but:

I have a 7970, NOT overclocked at all and am seeing this:
27.06 3.00 pending MilkyWay@Home v1.02 (opencl_amd_ati)

So mine is 5 seconds faster per work unit than yours is, using your non oc'd numbers.

wait a second. unless my brain is lagging right now this is not quite true.

first off, i flashed my 6950, so it now has 1536 shaders, technically making it a 6970 for mw@home as i understand it.

your work unit would,according to your statement, take 30 seconds, while my card needs 42 seconds @standard clock (820MHz). The best I can reach is 34 seconds with 1010MHz clock. so in fact your card is 12 seconds faster than mine, is that correct? or did I not catch sth here

anyway, thank you for the list, helping me a lot. does this mean that there is no difference between taking a 7970 or a r9 280x, except for the price? is one of them more suitable for overclocking than the other?

Yes you are right, my math was 'fuzzy' as was my head I guess.

Yes the cards are identical EXCEPT starting with the R series cards AMD has permanently removed something, damn I forgot what right now, that we crunchers needed. Thru software it is back in again, on the pre R series cards it is done thru hardware though. That means it SHOULD work at all Boinc Projects, but doesn't due to some projects using older version of the Server side software. From the top gpu list it looks like MW IS one of those that DOES support the R series cards as it says this "AMD Radeon HD 7870/7950/7970/R9 280X series (Tahiti)".

You can go here http://products.amd.com/en-us/GraphicCardResult.aspx and compare the specs on up to 5 cards at a time.

thank you for your response!
really helped me a lot.

but:

I have a 7970, NOT overclocked at all and am seeing this:
27.06 3.00 pending MilkyWay@Home v1.02 (opencl_amd_ati)

So mine is 5 seconds faster per work unit than yours is, using your non oc'd numbers.

wait a second. unless my brain is lagging right now this is not quite true.

first off, i flashed my 6950, so it now has 1536 shaders, technically making it a 6970 for mw@home as i understand it.

your work unit would,according to your statement, take 30 seconds, while my card needs 42 seconds @standard clock (820MHz). The best I can reach is 34 seconds with 1010MHz clock. so in fact your card is 12 seconds faster than mine, is that correct? or did I not catch sth here

anyway, thank you for the list, helping me a lot. does this mean that there is no difference between taking a 7970 or a r9 280x, except for the price? is one of them more suitable for overclocking than the other?

Yes you are right, my math was 'fuzzy' as was my head I guess.

Yes the cards are identical EXCEPT starting with the R series cards AMD has permanently removed something, damn I forgot what right now, that we crunchers needed. Thru software it is back in again, on the pre R series cards it is done thru hardware though. That means it SHOULD work at all Boinc Projects, but doesn't due to some projects using older version of the Server side software. From the top gpu list it looks like MW IS one of those that DOES support the R series cards as it says this "AMD Radeon HD 7870/7950/7970/R9 280X series (Tahiti)".

You can go here http://products.amd.com/en-us/GraphicCardResult.aspx and compare the specs on up to 5 cards at a time.

oh i did not know that either. i really hope they do support milkyway though, cause as recently as yesterday i ordered a xfx r9 280x black which clocks @1080MHz core. in another thread i believe to have already seen someone posting his/her time for a work unit running with a 280x, so i should be fine.

thank you especially and anyone else for the responses, though! :)

accidentally posted twice, sorry

Excuse an old sailor.
But!
Nvidia cards have quite poor DP perfomance, up and until the GTX- Titan series of GPU:s.

A GTX 970- for example "only" have a (DP)PPP of 109 GFLOPS, while the much, much older Tahiti, AMD r9 280 (which you still can get for less than €160) (7950) has a (DP) PPP of 836 GFLOPS. That's roughly eight times the DP processing power of the nvidia card (and the Nvidia card costs about €290).

The Tahiti card has a SP/DP divisor of 1/4, while the Nvidia card have a divisor of 1/32.

While you are absolutely right in that not all AMD cards are DP capable. The same goes for Nvidia GPU:s. There are actually fewer DP capable Nvidia GPU:S and what there is, is less capable at DP calculations than an AMD GPU (cent for cent!)...

@khrylxtko

There is essentially no difference between a 7970 and a 280X. Same Tahiti chip, same shader count, same 1/4 SP/DP divisor, same amount of memmory, same 384- bit bus, maybe a 10% higher clock.
So if you can get your hands on a 7970 for less money than the r9 280X- go for it.

And yes. The r9 290 (Hawaii) is faring worse than the 280X when it comes to DP calculation (although the Hawaii chip has more stream processors). Then again, the 290 is better when gaming.
So...

I don't think it makes sense to switch from 6970 to 7970.
Currently the HD 7970 (280X) is the best performer here due to high DP performance. The 290 series (Hawaii) have heavily reduced DP performance, so it's not a good choice for this project.
Theoretically the FirePro S9100 might be even better if it doesn't have crippled DP performance like the other consumer Hawaii SKUs. But I don't think anybody tested it here yet.

thanks for all the replies.

i have today received a xfx r9 280x black 1080MHz standard clock.

the card is running smoothly and i do see a difference. it takes the card 25 seconds for 1 work unit to finish @1080MHz, which is already 9 seconds faster than the 34 it took the flashed 6950@ 1000MHz (standard 820MHz)

I am currently trying around how to get the most out of its power, crunching 1 work unit seems to be way more ineffective than it was on my 6950.

same for 2 WUs

running 3 WUs at the same time it slowly starts to approach 99/100% usage, which is nice.

is there a specific amount of simultaneously run work units that s proven to be the most efficient? or is 3,4,6,8 or whatever just personal preference?

thanks for all the replies.

i have today received a xfx r9 280x black 1080MHz standard clock.

the card is running smoothly and i do see a difference. it takes the card 25 seconds for 1 work unit to finish @1080MHz, which is already 9 seconds faster than the 34 it took the flashed 6950@ 1000MHz (standard 820MHz)

I am currently trying around how to get the most out of its power, crunching 1 work unit seems to be way more ineffective than it was on my 6950.

same for 2 WUs

running 3 WUs at the same time it slowly starts to approach 99/100% usage, which is nice.

is there a specific amount of simultaneously run work units that s proven to be the most efficient? or is 3,4,6,8 or whatever just personal preference?

It's more of a trial and error process as to what works best on your machine. Keep in mind though that if you leave a cpu core free just to feed the gpu it will run faster then if you do not.

Now that you aren't using your old card maybe you could offer it to one of your teammates that they could then upgrade to from their current older one.

Hey guys, I know this thread is a couple of weeks old now, but I was wanting to ask a question. My main system has an R290X, but that is because it is also my gaming system. My other systems have a 7970 and 6970. At this point I have been unable to get my 290X system to obtain new work units.

I have removed and re-added the project, and even have gone so far as to completely remove and reinstall BOINC. No matter what I do I cannot get this 290X to get MW@H work units.

Anyone else having this problem, or know of any suggestions on how to fix it? I'm out of ideas, and while it may not be my fastest cruncher, I really hate the fact that I can't get MW@H units to keep it occupied when I'm not using it for gaming.

Hey guys, I know this thread is a couple of weeks old now, but I was wanting to ask a question. My main system has an R290X, but that is because it is also my gaming system. My other systems have a 7970 and 6970. At this point I have been unable to get my 290X system to obtain new work units.

I have removed and re-added the project, and even have gone so far as to completely remove and reinstall BOINC. No matter what I do I cannot get this 290X to get MW@H work units.

Anyone else having this problem, or know of any suggestions on how to fix it? I'm out of ideas, and while it may not be my fastest cruncher, I really hate the fact that I can't get MW@H units to keep it occupied when I'm not using it for gaming.

See this thread.
http://milkyway.cs.rpi.edu/milkyway/forum_thread.php?id=3721

---

khrylxtko (and anyone else running a R290)

Could you check out my benchmarking thread (linked by swiftmallard, 2nd post) & post a time for your R290?
We don't have any R290 times atm.

TIA :)

---

Team AnandTech - SETI@H, DPAD, F@H, MW@H, A@H, LHC, POGS, R@H, Einstein@H, DHEP, WCG

Main rig - Ryzen 5 3600, MSI B450 G.Pro C. AC, RTX 3060Ti 8GB, 32GB DDR4 3200, Win 10 64bit
2nd rig - i7 4930k @4.1 GHz, HD 7870 XT 3GB(DS), 16GB DDR3 1866, Win7

hey pal,

i am running a 280x, though