• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Anyone running Vega 64 in a custom loop?

aaronyuri

aaronyuri

aaronyuri

Associate
Joined
23 Mar 2017
Posts
413
Location
Cardiff
I'll be picking Vega up tomorrow, along with an EK water block. I'm going to run it on air initially and, if it's too loud, I'll add it into my loop.

Is there any significant overclocking headroom to be had with watercooling? I'm mostly doing this for the noise it may generate. I game with headphones on, however, so I may not even hear it and end up sending the block back if there's no OC headroom to gain.
 
Vega 64 air cooled reference design is very loud (I have one) I can here it over my headphones but it does all depend on how high you have your headphones.

Fwiw I'll be getting an EK block and backplate in a month or two :)
 
Yes I am, taken from the Vega 64 Owners thread.

https://forums.overclockers.co.uk/posts/31097415/

Yes plenty of headroom. I was running air for 2 weeks whilst I awaited my EK goodies, it was very hot and very noisy, it has the usual AMD reference blower noise. Once you hit around 80c then card throttles, so you have to undervolt and keep 3000rpm on the fan to keep it from doing so. I could only realistically get 1650mhz before it got to hot. Now I'm on water, maximum temps of around 40c, overclocked to 1750/1100 and silence almost :)
 
Great, thanks for the feedback, Tony.

Last reference AMD card I had was a 4870 X2 so I can't quite recall the noise vividly. I do remember it being loud, however.

What size rad and CPU are you running? I have a 360mm thick core and an 1800X @ 4GHz/1.4 (I didn't win the silicon lottery). Just curious how our setups compare and what temps I can expect.
 
Vega 64 reference will be hot and loud if you don't tame it using undervolting. Having said that I am very happy so far with 17.8.2 drivers. To get temperature and noise under control you need to ensure the voltage settings are identical in all 3 voltage fields (State 6, 7 and HBM2).

My manual settings in Wattman (revised)
  • 2nd BIOS enabled (via switch on card moved away from connectors)
  • Clocks State 6 and State 7: 1602 (gives around 1470 core). Any higher and I need to increase the voltage for stability.
  • HBM2 Clock: 1000-1020MHz
  • Fan speed: Min 400, Target 2200
  • Temperature: Max 85, Target 75
  • Power Limit default
  • Voltage Control: (wont actually go below .950mV despite putting in lower numbers)
  • State 6: .950mV
  • State 7: .950mV
  • HBM: .950mV

With these settings my GPU sits at 1470 pretty much constant at 2200RPM, with temperatures of ~77c (depends on ambient). It is actually faster than the stock balanced mode while being quieter and around 8c cooler. I still have leeway to decrease noise by going to 2100 on RPM which increases temperatures to around 79c.

So much better than stock for temperature and noise which gives 85c at2400 RPM and throttling. Though I am still contemplating a watercooling setup just for my Vega just to get that extra performance. :)
 
aaronyuri said:
Great, thanks for the feedback, Tony.

Last reference AMD card I had was a 4870 X2 so I can't quite recall the noise vividly. I do remember it being loud, however.

What size rad and CPU are you running? I have a 360mm thick core and an 1800X @ 4GHz/1.4 (I didn't win the silicon lottery). Just curious how our setups compare and what temps I can expect.
Click to expand...

360mm and 60mm deep :) running a 3770k @ 4.5GHz
 
Guys, does anyone knows if the Vega 64 (air cooled) is power limited compared to Vega 64 (liquid) as per AMD initial presentation?
 
I've done watercooling (even Tec cooling back in the day). Will be fitting an OcUK Techlabs AIO GPU cooler, when my 56 arrives. ICDP thx for the settings.
 
I've received the card today and just had a chance to fit it and have a tweak. Decent increase over my R9 Fury CF setup (63/64 CUs) and holy cow this thing is loud.

The biggest wall with this card appears to be the thermals. I was getting 96 FPS in Doom and then it tanked to 80 FPS within a minute or so. Core was hitting 1500 MHz when it was boosting (1652 set in WattMan) but as soon as the thermals took hold it throttled to 1100 MHz.

GPU freq state 6 & 7: 1652, 950 mV. HMB2 1000 MHz and 950 mV. This works well with the default fan configuration but throttles eventually. If I set the fan speed to 100%, it will cruise along at 1500 core but this isn't feasible long term due to the noise. The waterblock is going on this weekend... :)

More tweaking to come when I get the block on and start clocking the **** out of it!
 
aaronyuri said:
I've received the card today and just had a chance to fit it and have a tweak. Decent increase over my R9 Fury CF setup (63/64 CUs) and holy cow this thing is loud.

The biggest wall with this card appears to be the thermals. I was getting 96 FPS in Doom and then it tanked to 80 FPS within a minute or so. Core was hitting 1500 MHz when it was boosting (1652 set in WattMan) but as soon as the thermals took hold it throttled to 1100 MHz.

GPU freq state 6 & 7: 1652, 950 mV. HMB2 1000 MHz and 950 mV. This works well with the default fan configuration but throttles eventually. If I set the fan speed to 100%, it will cruise along at 1500 core but this isn't feasible long term due to the noise. The waterblock is going on this weekend... :)

More tweaking to come when I get the block on and start clocking the **** out of it!
Click to expand...

Thought that would be the case ;)

Look forward to seeing how you get on once you got that card on water.
 
Noticed that there's a big performance drop when you run HBM2 at anything above 1020 MHz. Tried 1025 MHz/1000 mV but this performed worse than 1020 MHz/950 mV.

Anyone else experienced this?
 
aaronyuri said:
Noticed that there's a big performance drop when you run HBM2 at anything above 1020 MHz. Tried 1025 MHz/1000 mV but this performed worse than 1020 MHz/950 mV.

Anyone else experienced this?
Click to expand...

Pretty sure hbm has ECC so if your hitting errors on the memory you will get performance drop off.
 
I figured something that like was at play. It's better than it crashing, I suppose.

The air cooler is an extremely limiting factor and WattMan isn't the best tool in my eyes. Hopefully the OC support picks up on third party apps and we can get a bit more granularity when tweaking.

I'll get the block on and report back. :) I see tonnes of untapped potential in this card.
 
dont touch the HBM voltage, it breaks the driver!

i'm using an EK block but mostly because the blower is too loud, there isnt really any extra headroom as my chip tops out around 1700mhz; the better bins became liquid cards (and some that weren't good enough, the amount of unstable cards is quite crazy)!

the HBM seems to prefer being cooler, despite it getting pretty toasty still; mine is fine at 1100mhz (quite lucky) but that seems to be on the stock voltage of 1.356V (according to hwinfo), hopefully they fix HBM voltage in the driver soon....

I tried using the liquid bios but that was obviously unstable by default since the stock clocks are 1750!

so unless you won the silicon lottery and your core clock goes to 1700+ you won't hit the power limit on the air bios...

as all the reviewers found, the higher clocks just aren't worth the extra power use; I run at 1550/1000mV core (between 1510-1530 reported depending on load) and 1100 HBM daily atm.
 
I didn't encounter any issues when modifying my HBM voltage. It's cruised along at 950 mV just fine. Could explain why I keep hitting performance issues above 1020 though (having undervolted it). I'm getting 1514-1550 core on 950 mV so may have won the silicon lottery for once! I don't see it stacking up to the XTX bin, however.

Interesting info. I'll get the card under water before making any further changes. What temps are you getting at 1500 core and 1100 HBM?

The performance, when air cooled thermals are taken out of the equation, is substantial. Noticed that throttled performance isn't that much better than my single Fury. Makes me think Vega is Fiji on clockspeed steroids.
 
things still get pretty warm, 40ish celsius on gaming workloads - there is a strange temp sensor called hot spot in the gpu-z beta that gets as high as 60

if running something memory intensive (like crypto coins) the HBM can get closer to 60 too.

I think the core voltage doesn't actually go down below the hbm memory setting currently so didnt really try to get the ultimate undervolt yet, I tried 970mV but nothing seemed to change so left it at 1000mV!

you definitely need to set the blower RPM to jet engine levels to avoid throttling when the core is at 1200mV, it wasn't super noisy undervolted but the blower really couldn't handle memory loads at all; I benchmarked a crypto load and the HBM temps went through the roof!

the performance is great at the undervolt, around 24k graphics in firestrike which actually improved slightly with the HBM under water (there is some sort of thermal throttle built in).
 
40 isn't that bad for gaming workloads considering 85 on air is a guaranteed temp! What loop setup are you running? Have you got the EK backplate? (I've got one ready to go on).

I just threw it on 100% fan, 950 mV core/HBM and started pushing the HBM (topping at 1020 before perf loss). I then pushed the core to 1652 in WattMan but it was getting late at this point.

It was, on occasions, trickling to 100 FPS in Doom (4k, Vulkan, and maxed out settings). Throttling was crippling it to 76 FPS with peaks to 80-82 once the fan was put back to stock settings. That's a varying 15-20% loss due to throttling (albeit in one game). I'm not a big fan of synthetics. I picked Doom as my test bed as it's probably Vega's best game so I'll see those 1-2 FPS increases with fine tuning whereas I'd see nothing in others, I suspect.

All this talk is itching me to tweak! I'll raise the voltages to 1000/1050 mV later and see if I can get the HBM higher.
 
You must log in or register to reply here.
Back
Top Bottom