How good is an RX590 at DC ?

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
23,109
11,861
136
And in what projects ? I have come into one.
 

VirtualLarry

No Lifer
Aug 25, 2001
53,782
8,019
126
Shrug. It's a Polaris GPU. Not horrible at compute, can do 30MH/sec in ETH, but under-performs at F@H. (300K PPD)
 

Skillz

Senior member
Feb 14, 2014
605
541
136
Probably not a lot as stated above.

I'd toss is on Einstein@home though, see what it can do there.
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
23,109
11,861
136
Shrug. It's a Polaris GPU. Not horrible at compute, can do 30MH/sec in ETH, but under-performs at F@H. (300K PPD)
I thought AMD was good at FP type projects. I was told even at this date that a 7970 was pretty good as some, but I don't have that anymore.
 

VirtualLarry

No Lifer
Aug 25, 2001
53,782
8,019
126
7950/7970 were FP64 beasts. Polaris doesn't share that advantage, specific to FP64, though it does alright with FP32. In Windows, need to set the driver to "Compute Mode".
 

Icecold

Golden Member
Nov 15, 2004
1,070
983
146
One thing that Polaris does decent on is there's an OpenCL specific version of MQueens for Yoyo@home that only runs on a small handful of AMD cards such as Polaris. I couldn't get it to run on a Firepro S9150 or Radeon 7950 but it should run on Polaris and IIRC runs on a Vega 56. Pretty niche though and I think there's only binaries for Windows, and you have to manually put them in the correct folder.


Yoyo@home webiste said:
16 December 2020 GPU version for M Queens
We have now a first GPU version for the M Queens project for Windows. It requires an OpenCL 2.0 GPU. The version can be downloaded from github.com. The zip contains an installation instruction, the app and the app_info.xml.
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
23,109
11,861
136
It will be on linux.....
 

Icecold

Golden Member
Nov 15, 2004
1,070
983
146
It will be on linux.....
I'm not aware of any projects on Linux where an RX 590 will perform better than even some of your 'lower end' Nvidia cards you don't always run(2060, 1070ti, etc.). Even Einstein which was historically pretty good on AMD cards runs pretty well on Nvidia cards these days as far as I know.
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
23,109
11,861
136
I'm not aware of any projects on Linux where an RX 590 will perform better than even some of your 'lower end' Nvidia cards you don't always run(2060, 1070ti, etc.). Even Einstein which was historically pretty good on AMD cards runs pretty well on Nvidia cards these days as far as I know.
Oh well, I guess another paper weight.
 

Icecold

Golden Member
Nov 15, 2004
1,070
983
146
Even with video cards being down in price somewhat recently, it's probably still worth at least $250-$300. I'd throw it on the FS/T section of the forum.
 

Skillz

Senior member
Feb 14, 2014
605
541
136
Good advice. I'd sell it while it's still worth something.

As other have said, only some AMD cards have crazy good FP64 performance such as the 7970. If that's the kind of projects your wanting to do good in then look for Radeon VII cards.
 

StefanR5R

Elite Member
Dec 10, 2016
4,683
5,767
136
At this time, there are only two DC projects which benefit from FP64 support in GPUs:
  1. MilkyWay@home of course,

  2. one or more of PrimeGrid's Genefer based subprojects:
    GFN-21 (for as long as the test candidates don't become too big to be viable for Genefer's FP64 based 'ocl' transform);
    not sure of the smaller GFNs.
(Of course that's somewhat off-topic to the thread's subject which does not have decent hardware support for FP64, only minimal compatibility support.)

Edit: In FP32 based projects, AMD GPUs hold up well (more or less) against NVidia GPUs in Einstein@home, and that's it. In other projects, NVidia GPUs tend to be way ahead. But that's only my superficial, unreliable observation, as I haven't used AMD GPUs in DC myself for some years now.
 
Last edited:

mmonnin03

Member
Nov 7, 2006
80
84
91
Polaris cards perform pretty well in E@H due to HBM as E@H is GPU memory speed sensitive. Per watt it at least used to be pretty good, not sure how it stacks up with Ampere now though. in E@H on polaris cards, there wasn't much if a difference between 470, 480, 570, 580 so a 590 may not be worth extra over those cards. OCing the mem gave the most boost while reducing core volts dropped power. Similar to mining.
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
23,109
11,861
136
Polaris cards perform pretty well in E@H due to HBM as E@H is GPU memory speed sensitive. Per watt it at least used to be pretty good, not sure how it stacks up with Ampere now though. in E@H on polaris cards, there wasn't much if a difference between 470, 480, 570, 580 so a 590 may not be worth extra over those cards. OCing the mem gave the most boost while reducing core volts dropped power. Similar to mining.
I got it free. Just wondering where I should use it.
 

Shmee

Memory and Storage, Graphics Cards
Super Moderator
Sep 13, 2008
6,032
1,401
136
Polaris cards perform pretty well in E@H due to HBM as E@H is GPU memory speed sensitive. Per watt it at least used to be pretty good, not sure how it stacks up with Ampere now though. in E@H on polaris cards, there wasn't much if a difference between 470, 480, 570, 580 so a 590 may not be worth extra over those cards. OCing the mem gave the most boost while reducing core volts dropped power. Similar to mining.
Just a correction, possibly, Polaris used GDDR5, not HBM. Unless you were talking about the bandwidth of said GDDR5 in general, I wasn't sure. Fiji and Vega used HBM.

Anyway, congrats on the free RX 590 Mark! Just curious, what model/make is it?

As some have mentioned, at this point it might be most valuable as an older midrange gaming card. I figure it might do decent in Milkyway@home and ETH mining, but I am not certain, might not be worth it. I know back in the day, before I got into mining at all, but I did some DC, like F@H, and then BOINC, Geforce always did better at folding but Radeons did better at projects like Milkyway etc. Of course this was back when I had a GTX 260 and then an HD 5870, all on a first gen i7 Bloomslfield/Gulftown on X58.
 

Markfw

CPU Moderator, VC&G Moderator, Elite Member
Super Moderator
May 16, 2002
23,109
11,861
136
Just a correction, possibly, Polaris used GDDR5, not HBM. Unless you were talking about the bandwidth of said GDDR5 in general, I wasn't sure. Fiji and Vega used HBM.

Anyway, congrats on the free RX 590 Mark! Just curious, what model/make is it?

As some have mentioned, at this point it might be most valuable as an older midrange gaming card. I figure it might do decent in Milkyway@home and ETH mining, but I am not certain, might not be worth it. I know back in the day, before I got into mining at all, but I did some DC, like F@H, and then BOINC, Geforce always did better at folding but Radeons did better at projects like Milkyway etc. Of course this was back when I had a GTX 260 and then an HD 5870, all on a first gen i7 Bloomslfield/Gulftown on X58.
Its a XFX Radeon RX 590 Fatboy from 2018. Like I said, no idea what to do with it. Worse case ? My new 12700F has 9800 GTX+ from like 14 years ago, and this would replace that.
 

Assimilator1

Elite Member
Nov 4, 1999
23,975
450
126
Whilst their not as good at Milkyway as the HD79x0s, they're still pretty decent at it :).
Gaming wise the RX 580 I have is roughly the same as GTX 1060, but the RX 580 has considerably better FP64 than that card (hence why I ended up buying one).
Looking at my old Milkyway benchmark thread, my RX 580 was faster than an RTX 2080 Ti! (although an HD 7970 took less than 1/2 the time my RX 580 did!).

I think the RX 590 was about 5% faster than the 580? (IIRC from an AnandTech review).

I recently ran Einstein on my RX 580 (and will do a little tomorrow too), but for now you can still see a few results from it here, if you want to see how it compares to other GPUs (I've no idea there). Actually I'll copy them here, let's see if the formatting holds....

Task ID Workunit ID Computer Sent Time reported or deadline Status Run time CPU time Granted credit Application
LATeah3012L03_756.0_0_0.0_24482007_0 612488579 11395023 6 Mar 2022 12:36:08 UTC 6 Mar 2022 13:45:57 UTC Completed and validated 533.02 68.20 3,465.00 Gamma-ray pulsar binary search #1 on GPUs v1.22 (FGRPopencl1K-ati)
windows_x86_64
LATeah3012L03_756.0_0_0.0_24473700_0 612488565 11395023 6 Mar 2022 12:36:09 UTC 6 Mar 2022 13:45:57 UTC Completed and validated 645.72 59.81 3,465.00 Gamma-ray pulsar binary search #1 on GPUs v1.22 (FGRPopencl1K-ati)
windows_x86_64
LATeah3012L03_756.0_0_0.0_23572071_1 612487100 11395023 6 Mar 2022 12:36:09 UTC 6 Mar 2022 16:48:25 UTC Completed and validated 514.86 61.48 3,465.00 Gamma-ray pulsar binary search #1 on GPUs v1.22 (FGRPopencl1K-ati)
windows_x86_64
LATeah3012L03_756.0_0_0.0_19488222_0 612462280 11395023 6 Mar 2022 12:36:09 UTC 6 Mar 2022 13:45:57 UTC Completed and validated 543.23 66.61 3,465.00 Gamma-ray pulsar binary search #1 on GPUs v1.22 (FGRPopencl1K-ati)
windows_x86_64
 
Last edited:
  • Like
Reactions: Markfw

mmonnin03

Member
Nov 7, 2006
80
84
91
Just a correction, possibly, Polaris used GDDR5, not HBM. Unless you were talking about the bandwidth of said GDDR5 in general, I wasn't sure. Fiji and Vega used HBM.

Anyway, congrats on the free RX 590 Mark! Just curious, what model/make is it?

As some have mentioned, at this point it might be most valuable as an older midrange gaming card. I figure it might do decent in Milkyway@home and ETH mining, but I am not certain, might not be worth it. I know back in the day, before I got into mining at all, but I did some DC, like F@H, and then BOINC, Geforce always did better at folding but Radeons did better at projects like Milkyway etc. Of course this was back when I had a GTX 260 and then an HD 5870, all on a first gen i7 Bloomslfield/Gulftown on X58.
Ah true. Dunno why it performs well in E@H then. Not at ampere level or VII levels but they were the cards of choice for awhile. Can't say I used my 580 for much else than E@H.
 

Assimilator1

Elite Member
Nov 4, 1999
23,975
450
126
It does well in E@H because it has better FP64 than the same generation Nvidia cards.
e.g (according to wiki figures) :-

RX 580 386 Gflops
RX 590 423-445 Gflops
GTX 1060 120-137 Gflops
GTX 1070 Ti 244-256 Gflops
GTX 1080 Ti 332-354 Gflops
Titan XP 337-380 Gflops
RTX 2070 203-233 Gflops
RTX 2080 Ti 367-421Gflops
And for fun, the HD 7970 GHz edition 1024-1075 Gflops!
HD 7870 XT 710 Gflops

Of course that doesn't speak as to how well each GPU uses that processing power, but as per my earlier post talking about MW benchmarks, my RX 580 was quicker than a RTX 2080 Ti. Some stats from MW benchmark thread :-

GPU statistics - Average Run Time to Complete 1 MW v1.46 227.62 credit WU :-

HD 7970, GPU 1200 MHz(!) (CPU, Xeon E5 ES 10 core @2.7 GHz, HT off) ..... 38.2s .... tictoc
R9 290, GPU 1000 MHz, (CPU, Xeon E5 ES 10 core @2.7 GHz, HT off) ........... 70.9s .... tictoc
HD 7870 XT 3GB(DS), GPU 925 MHz (CPU, C2 Q9550 @3.58 GHz) ................ 73.2s .... Assimilator1
RX 580 8GB, GPU 1350 MHz (CPU, i7 4930k @4.1 GHz) .................................. 97.3s .... Assimilator1
RTX 2080 Ti, GPU ???? MHz (CPU, i7-8700K @4.7 GHz no AVX) ................... 110.6s .... IEC
 

StefanR5R

Elite Member
Dec 10, 2016
4,683
5,767
136
FP64 support is playing only a secondary role at Einstein@home though:
In 2017 Christian Beer (E@h moderator) said:
Double precision is only needed at the end of a task for a short time. The test is done at the beginning to determine if this can be done on the GPU or CPU. [In the latter case] the app should fall back to CPU processing of this part of the computation.
(source)
In 2018 Gary Roberts (E@h moderator) said:
If you are talking about the FGRPB1G search using GPUs, only the final followup stage (<10% of the calculation) uses double precision. If the GPU doesn't have DP capability, the followup stage would use a CPU core - and take a lot longer. So DP capability is quite desirable. My impression is it doesn't need to be particularly 'strong'.

At the moment on an AMD HD7850 GPU running 2 concurrent tasks, the followup stage takes around 20-30 secs out of an overall crunch time of perhaps 28 mins (two tasks completed). This is about a fifth of what it used to be on previous data files so the nature of the data has changed somewhat recently.
(source)
 
  • Like
Reactions: Assimilator1

ASK THE COMMUNITY