• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Having a 3-way... with RX 580s

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.
mackerel

mackerel

Member
​
Joined
Mar 7, 2008
3xcfno.jpg

As mentioned in another thread, I just received 3x RX 580s. Primarily to get one steam achievement... am I crazy? Probably... anyway, simple enough plan. I previously checked my Asrock X99 Extreme3 board supports 3 way Crossfire, running 16/16/4 with a 40 lane CPU like the Xeon I have. Based on TDP, the three GPUs would be around 555W so the existing 550W in there isn't going to do. I'll just grab the old Corsair TX750 from another system... it's in the 1700/Vega system. Doh! So that's now shut down and offline until I back fill it.

The case I have didn't have a gap between the lowest PCIe slot and the PSU shroud. I thought I'd use a PCIe extension I bought for another project, but the system was unstable AF. I've noticed in the past, nvidia cards seem to work fine with it, but AMD cards really didn't like it. This was no exception. Ok, let's take the whole lot out of the case then...

3x-cf.jpeg

And away we go! Without any tuning, I got 15386 in Fire Strike regular. I think the CPU might be holding it back... it's an E5 2683 v3 Xeon (ES), and although it has a ton of cores, the clock only goes up to 2.3 GHz. That run was enough to get me the achievement, so I'm done... or am I? I'll try to OC it later. I've submitted it to the bot, easy gold medals as no one else is running 3 way Crossfire it seems. 4x and 2x yes, but not 3x. Snag is PC is only Win7 so no Time Spy. Don't think I have a spare disk at moment for a temporary install...
 
Looks great, sir, but you mention you got gold medals in the bot, what does it do? Is there a reward of some sort?

Woe, barely enough room to squeeze a sheet of paper between cards
 
Taco - Hwbot.org

Its a databse/repository for overclocking and holds competitions. No rewards outside of pride and supporting the team. See benchmarking section too for more details.
 
Even if older, a Fury is a higher performing card right? I've yet to run with 2x CF or even 1x to see how it scales. It might just be that 3x sucks, especially as the 3rd card is on a 4x slot. Also I'm not that familiar with what impacts the scoring of the subtests. My arrangement was only ahead of Johan's in 3D test 2, and lagged in the other areas. I'm still occasionally looking for a cheap 5820k or similar as that should help on the CPU side for this type of benching.

The middle card is rather suffocated so it might hold back the lot.

I'm also not sure if it is a Crossfire thing, but GPU-z behaves oddly when Crossfire is enabled. It is somewhat unresponsive and the reported values for the non-main cards seem to be nonsense. I think it might just be the driver isolating the cards from detection, as everything is fine when Crossfire is off. This is unlike nvidia as regardless of SLI things work as expected.

I've put them on a distributed challenge elsewhere now, which ends Wednesday afternoon so I'll pick up testing again after that. Oh, on that challenge, the 580's are slower than my stock 280X... I mentioned it elsewhere, and others say cards after the 200 series seem to have gone backwards in that compute application. Just as well I didn't buy it for that! Comparing stats, the 580 has more compute potential, but less ram bandwidth, so I wonder if that might be a contributing factor.

Long term, I'll probably run two in one system, and the 3rd will go elsewhere. I'll have to have a think how best to deploy my GPUs as I'm not going to be short on them in the near future... looking to accumulate ethereum in short term. On that note, I haven't tested those cards with mining yet.
 
The number of threads and CPU speed directly impact Physics and combined tests in FS. Could be the driver too. From experience AMD performance varies widely depending on driver version. If I have time either tonight or tomorrow night I can put the RX580 I have into my Ryzen just to give you a single card comparison
 
Some stock performance results could be useful as a sanity check. Actually, I have one of the 580's "spare" at the moment (to allow the other two to breathe) so could slap that into another system temporarily and see where that gets me. At least in my case it'll be the same card, even if not the same CPU.
 
Try running something like gpupi with all 3 to see how they do. That will eliminate the cpu as a variable.
 
Lochekey said:
Try running something like gpupi with all 3 to see how they do. That will eliminate the cpu as a variable.
Click to expand...

Good point, I have no idea what software like that can use multiple GPUs. I was wondering if the hwbot listings for multiple GPUs even makes sense in some cases, in case the software can't use more than one anyway. I haven't looked into if that is or isn't the case yet.
 
Some benches do some don't some of the older 3dMark01 and Aquamark won't benefit from a second GPU. Pretty much any bench that loves high CPU speed doesn't like CFX or SLI.
Newer 3d mark does though like 03 and up and catzilla
 
IIRC AMD DEtuned their mainstream GPUs compute after the 200~300 series. The cards where running 1/4 and nvidia I think was running 1/16, so AMD dropped theirs back also. This way you pay for the FirePro\Workstation GPU that is more $$$. If you look at the number of subs for video cards, The 580 has like 1k while the 480 has 8k+.
 
This isn't FP64 workload, probably integer actually. I'm not aware of any nerfing going on in that area, but the workload I'm running is more of an nvidia one anyway.
 
Thanks for results - how does the tessellation setting affect results? I was also shown a 3x 480 result elsewhere that scored almost double mine, but I haven't looked at it in detail yet. As said, currently I have two cards running on a compute challenge, and I got the dates mixed up slightly. It ends tomorrow, not tonight, so I wont do further testing until then. It does mean I can quick run 2x to compare scores. Actually, if crossfire is off, it'll still run 1x right? So I could see the scaling effects.

On the CPU side, I just put in bids on ebay for a 5820k and 6800k. Based on other recent finishing prices, I think I have maybe 50% chance of "winning" the 5820k and no real chance on the 6800k as I put in a low bid, but you never know... will find out tonight if I get either.
 
Disabling tesselation will give a boost and it's legal on most 3Dmark benches because it's a driver modification, you'll find it in wattman. Time Spy is an exception which I'm not sure why but it is. Yes you should be able to just disable CFX in software and get a single result. Check GPUz it should tell you if it's using a single or still in CFX.
I realized today that my card has a factory OC so the stock run at 1425 really isn't the "reference" speed, it also has a +50% power limit and up to 200mV of voltage leeway
 
Ok, I'll have to look further into hwbot legal optimisations for future submissions. For now I'll keep it stock so I have like for like comparisons.

I know right now crossfire is off as I had to turn it off. While it is on, I'm seeing weird behaviour in GPU-z and other applications that want to address each card individually. I haven't given any time to figuring that out.
 
Most likely driver IMO which causes weird issues. I used to run SLI quite a bit and drivers are even more important for multicard setups than single
 
You must log in or register to reply here.

Similar threads

Overclockers.com
FRONTPAGE Gigabyte Radeon RX 7900 GRE Gaming OC 16G Graphics Card Review
Replies
0
Views
163
Overclockers.com
Overclockers.com
spiritedandy
RX 7900 XTX Black screen Mouse freezing in bios . General freezing instability
Replies
9
Views
456
spiritedandy
spiritedandy
mtdewcowboy
A Little Upgrade
Replies
2
Views
324
Tír na nÓg
Tír na nÓg
M
Overclocking Fx 6300
Replies
5
Views
408
Blaylock
Blaylock
rainless
HEY! Wait a minute... (RTX 4060 PCIe)
2
Replies
34
Views
869
rainless
rainless
Back