Vega 64 hotspot- how hot is too hot?

NattyKathy

[H]ard|Gawd
Joined
Jan 20, 2019
Messages
1,470
is on the edge of 100C too much for V64 hotspot temps? My core stays in the mid-high 50s under load no matter the power limit (aio fans just ramp up) but raising PL from 37%/300W to 50%/330W is enough to push hotspot from high 80s/low 90s to high high 90s.

I'm definitely pushing the H100i's thermal density limits (hotspot drops into the 70s with 280W PL for comparison) but OCd at stock voltage, clocks are higher and much more stable with 330W limit (actual hovers around 310-325) than even 300W so if it's safe, I'd like to push that last lil bit of power thru the core. I know AMD (in?)famously said 110C hotspot is ok for Navi but does that mean that's ok for Vega also- or does Vega have less comprehensive hotspot monitoring that requires more safety margin?
 
I've heard the card throttles at 95C, so ideally you want to stay under that.

throttles at 95C hotspot or 95C general? it seems less clear how the hotspot affects throttling- I saw somebody on Reddit say the max hotspot temp is as high as 105C.
 
I haven't seen anything official from AMD, but I did notice that 105C number on Reddit (I think from HWInfo) so that might be the max safe temp, not that you want to go there on the regular.

Based on my experience with Navi, you are probably good if you can stay under 100C for hotspot, though Vega may be slightly different (there wasn't good information back then, so I don't know for sure).
 
How are you attaching the AIO? Does the bracket you are using also have a fan? What kind of fan are you using?
 
I haven't seen anything official from AMD, but I did notice that 105C number on Reddit (I think from HWInfo) so that might be the max safe temp, not that you want to go there on the regular.

Based on my experience with Navi, you are probably good if you can stay under 100C for hotspot, though Vega may be slightly different (there wasn't good information back then, so I don't know for sure).

Ok thanks for the feedback. Its staying below 100 so I'll leave the +50% PL for now and keep an eye on the temps.

How are you attaching the AIO? Does the bracket you are using also have a fan? What kind of fan are you using?

I've mounted the AIO with a modified Intel X-bracket on the pumphead side and the Vega's original X-bracket on the back. I think the mounting pressure seems OK as I'm only seeing a big hotspot delta with really high power limits- at stock or even more sane settings like undervolted at 180W the hotspot is only a few degrees above the reported core temp. I've retained the reference frontplate, fan, and shroud (only thing removed/replaced was the heatsink itself) so the VRMs etc have ample cooling.
 
I've heard the card throttles at 95C, so ideally you want to stay under that.

Limits on air bios are 85C core, 95C HBM, 105C hotspot. If hotspot throttled at 95C, everyone's reference card would throttle :) Of course you want to stay below that.
 
Limits on air bios are 85C core, 95C HBM, 105C hotspot. If hotspot throttled at 95C, everyone's reference card would throttle :) Of course you want to stay below that.
I've never seen even 90 on my ref card..
Some cards can have might higher hotspot deltas at lower power so sounds like OPs card is fine.
I found undervolting Vega more fun though..
My 'hot and loud' card drawing 120-180W at 68C on the 'shitty blower' at 1.55+GHz.. Apparently Vegas like that are impossible according to haters.
 
Limits on air bios are 85C core, 95C HBM, 105C hotspot. If hotspot throttled at 95C, everyone's reference card would throttle :) Of course you want to stay below that.

Thanks for clarifying. 105C is more reasonable than 95.

I've never seen even 90 on my ref card..
Some cards can have might higher hotspot deltas at lower power so sounds like OPs card is fine.
I found undervolting Vega more fun though..
My 'hot and loud' card drawing 120-180W at 68C on the 'shitty blower' at 1.55+GHz.. Apparently Vegas like that are impossible according to haters.

undervolting is the way to go with Vega! I was running UV during the summer but now that I don't have to worry about overheating my apartment I'm having some high-current fun. You must have really won the silicon lottery to get 1550Mhz+ at <180W :eek: When I had mine at 950mV I saw spikes into the mid 1500s under middling loads but fully-loaded it could only sustain 1450-1470Mhz.

FWIW my hotspot delta is only this large at my 280-330W overclocked. At stock or undervolted the delta is only like 5-10*.
 
AMD cards can run hot, it is known. I wouldn't worry about it. It's rare to see an amd card die from heat.
 
Thanks for clarifying. 105C is more reasonable than 95.

undervolting is the way to go with Vega! I was running UV during the summer but now that I don't have to worry about overheating my apartment I'm having some high-current fun. You must have really won the silicon lottery to get 1550Mhz+ at <180W :eek: When I had mine at 950mV I saw spikes into the mid 1500s under middling loads but fully-loaded it could only sustain 1450-1470Mhz.

FWIW my hotspot delta is only this large at my 280-330W overclocked. At stock or undervolted the delta is only like 5-10*.

5-10° hotspot is considered good. Very interesting how it gets so much more at high OC... poor sillicon.. you meanie!

Hah high current fun is far better than wasting it through a heater ;D I think you have the right idea! That was really the only good side to mining the few times I tried it over the years. Now coming to summer where I am at the moment so something I'd like to try (pushing the Vega) in future with a better cooler and try the higher power approach.. jelly of you! These are the most tweakable cards I've seen since ever, really enjoying the Vega and I hope the next Navi cards or beyond are similar, as I heard they tightened down on the 5700s compared to Vega. I been looking at modding the accelero III on to it but it's a bit of a bitch, enough that I don't have the time.
950mV up to 1500 core is incredible though! Pretty sure mine can't do that, typically 1020-1025mv @1500-1580 highest is 1599 but thats leaving AMD to do the clocking as necessary, I ain't forced anything but voltages. Yours should be better than mine but we could just have two different leakage chips at different ends of the scale that do both well differently. Your one can do less volts but needs more current due to higher power draw? Or you are running a different title/scenario. I ain't turned mine up yet to see what it'll do.. maybe it'll fall on its ass compared to yours in that case.Was yours an early ref or later ref? Mine is from late 2018 in the last lots of stock. Ex-NKD (from here), did the hard yards benching it, great guy to buy from! Also I do only run No Mans Sky currently, it's not always full % utillisation e..g in freighter or in ships but flying around/planetside it will usually push it to the limits. I don't touch the power limit though, do you?
I'm quite curious to test a workload you run now to see how it stacks up. What game/program do you do your power readings with? Win7 though here haha.

Also something interesting, not sure if No Mans Sky update or Driver update but since the prior update I had to back off HBM OC a little as was getting instability, probably using it harder now. Temps slightly higher with season too. I have not touched voltage (set 960-970) but apparently it's static and doesn't change in reality to the HBM regardless, maybe I misunderstood.. was 1100mhz but now 1085 and climbing back up via testing. Have heard of this in the past but curious if you run into it too? I'd only heard of HBM degradation from high voltage/high oc in mining farms prior to updates changing OCs.

Also have you tweaked all the P-state voltages or just P7? I did the lot but definitely need to spend more time on the P5 and under ones. I think when the card is clocking down to save power these become very important for reducing your average power and thus cooling the chip more in lower use scenarios, allowing higher burst clocks to P7...
 
My AIO water cooled daily driver settings: (It'll do more, but those settings run everything I throw at it just fine. Does like the voltage though. Whatever the settings, it runs nice and silent, which was ultimately the purpose)

upload_2019-11-14_11-11-1.png


IMG_20181210_205249806.jpg



See it in action with OSD temps on my post here - Crytek Releases Hardware-Agnostic Raytracing Benchmark "Neon Noir"
 
My AIO water cooled daily driver settings: (It'll do more, but those settings run everything I throw at it just fine. Does like the voltage though. Whatever the settings, it runs nice and silent, which was ultimately the purpose)

View attachment 199692

View attachment 199709


See it in action with OSD temps on my post here - Crytek Releases Hardware-Agnostic Raytracing Benchmark "Neon Noir"

very nice setup there. these cards love being watercooled!

edit to add- I like the VRM heatsinks you added. looks effective!
 
5-10° hotspot is considered good. Very interesting how it gets so much more at high OC... poor sillicon.. you meanie!

Hah high current fun is far better than wasting it through a heater ;D I think you have the right idea! That was really the only good side to mining the few times I tried it over the years. Now coming to summer where I am at the moment so something I'd like to try (pushing the Vega) in future with a better cooler and try the higher power approach.. jelly of you! These are the most tweakable cards I've seen since ever, really enjoying the Vega and I hope the next Navi cards or beyond are similar, as I heard they tightened down on the 5700s compared to Vega. I been looking at modding the accelero III on to it but it's a bit of a bitch, enough that I don't have the time.
950mV up to 1500 core is incredible though! Pretty sure mine can't do that, typically 1020-1025mv @1500-1580 highest is 1599 but thats leaving AMD to do the clocking as necessary, I ain't forced anything but voltages. Yours should be better than mine but we could just have two different leakage chips at different ends of the scale that do both well differently. Your one can do less volts but needs more current due to higher power draw? Or you are running a different title/scenario. I ain't turned mine up yet to see what it'll do.. maybe it'll fall on its ass compared to yours in that case.Was yours an early ref or later ref? Mine is from late 2018 in the last lots of stock. Ex-NKD (from here), did the hard yards benching it, great guy to buy from! Also I do only run No Mans Sky currently, it's not always full % utillisation e..g in freighter or in ships but flying around/planetside it will usually push it to the limits. I don't touch the power limit though, do you?
I'm quite curious to test a workload you run now to see how it stacks up. What game/program do you do your power readings with? Win7 though here haha.

Also something interesting, not sure if No Mans Sky update or Driver update but since the prior update I had to back off HBM OC a little as was getting instability, probably using it harder now. Temps slightly higher with season too. I have not touched voltage (set 960-970) but apparently it's static and doesn't change in reality to the HBM regardless, maybe I misunderstood.. was 1100mhz but now 1085 and climbing back up via testing. Have heard of this in the past but curious if you run into it too? I'd only heard of HBM degradation from high voltage/high oc in mining farms prior to updates changing OCs.

Also have you tweaked all the P-state voltages or just P7? I did the lot but definitely need to spend more time on the P5 and under ones. I think when the card is clocking down to save power these become very important for reducing your average power and thus cooling the chip more in lower use scenarios, allowing higher burst clocks to P7...

Computers are just heaters that happen to be really good at math ;) Summertime is definitely not the time for cranking the GPU heat tho! Does it get very hot where you are? We had a lot of 35C+ days this year where I'm at... I was truly appreciating that undervolt then!

FWIW, I've heard that Navi tweaking is similar to the Vega 56/64 situation- the 5700XT can go to a very high power limit but 5700 is a bit more locked down. Plenty of fun to be had if experiences I've seen here are any indication!

I have been testing with a bunch of things- Heaven, Superposition and Time Spy for synthetics and Far Cry 5, Metro 2033/Last Light/Exodus, Mass Effect Andromeda, Skyrim SE (with hella gfx modz) for games, probably a few others I'm blanking on right now. I played a lot of games this year lol. I've also been using the pathtraced global illumination/ambient occlusion Reshade shader which really pushes the GPU to the limit. Even with P7 at 1690Mhz+ & 1.2v (which usually results in like 1.135v/1650Mhz actual) it's actually somewhat difficult to get the GPU to pull 300W+ but with RTGI it'll run right into the power limit. I think it's worth testing multiple ways though because OC's on these cards behave differently stability-wise under moderate-heavy load vs maxed-out. In particular, Heaven and Superposition make a decent simple & free combination as Heaven with it's inconsistent utilization exposes high-clock-related instability whereas Superposition makes for a heroic effort that will push clocks low and show how things behave under extreme load. My UV-or-OC-no-in-between approach means I can't do anything about Power Limit on the high end- 330W is all I get until/unless I figure out a way to get the Liquid BIOS working properly with new drivers (from my brief, glitchy testing with a LC BIOS it seems like 360W is a good spot that will really keep clocks from floating)- but on the low end I've tried multiple approaches. From my experiences, using a low power limit (180W was what I used most for UV testing) can decrease stability in scenarios where the GPU is running at high utilization; when the load eases up suddenly and then increases again just as fast, huge clock spikes can occur that will cause crashes if the P7 is too high. There's more consistent stability to be had IME with leaving the PL at 220-240W and lowering actual power solely thru dropping the voltages. Of course, this results in power consumption that is both higher and very inconsistent but it can help with frequency spike related crashes. Don't spike the P7! I've found more stability and higher clocks to be had from keeping P6 & P7 moderate and very close together, than from pushing P7 to massive & unrealistic heights.

I don't really know much about OCing the HBM- unfortunately my card reacts very poorly to raised memory clocks so I've mostly kept it at stock. One of these days when I'm fully satisfied with the core clocks and stability I'll mess with the HBM again. As for the P-states, yes I do tweak all of them. I don't know how necessary it is- some folx will say to only touch P6 & P7 (and those are the ones to focus on obviously) but I mess with all of them because that's the kind of person I am.

edit- is there a way to tell my cards manufacture date from the serial #? I bought it from a local Craigslist seller this past May and IIRC he said he'd had it for a year-ish, so it could be early 2018?
 
Last edited:
so, an update- I switched the V64 from the Corsair H100i Pro over to a H115i Pro (240mm vs 280mm rad, same pumphead design), was a bit more courageous with mounting pressure this time, and lo and behold @330W my hotspot dropped all the way to low-mid 80's and core dropped a little to low 50s. I've only tested it with one game and Superposition but with both of those showing consistent temp drops I'm calling it a win. <Steve Ballmer voice> I! LOVE! THIS! COMPANY! GPGPU! I also went all out with a bit of custom paint as I'm mad like that, pics in my other thread here -> https://hardforum.com/threads/vega-64-modded-corsair-aio.1984529/#post-1044393111
 
Computers are just heaters that happen to be really good at math ;) Summertime is definitely not the time for cranking the GPU heat tho! Does it get very hot where you are? We had a lot of 35C+ days this year where I'm at... I was truly appreciating that undervolt then!

FWIW, I've heard that Navi tweaking is similar to the Vega 56/64 situation- the 5700XT can go to a very high power limit but 5700 is a bit more locked down. Plenty of fun to be had if experiences I've seen here are any indication!

I have been testing with a bunch of things- Heaven, Superposition and Time Spy for synthetics and Far Cry 5, Metro 2033/Last Light/Exodus, Mass Effect Andromeda, Skyrim SE (with hella gfx modz) for games, probably a few others I'm blanking on right now. I played a lot of games this year lol. I've also been using the pathtraced global illumination/ambient occlusion Reshade shader which really pushes the GPU to the limit. Even with P7 at 1690Mhz+ & 1.2v (which usually results in like 1.135v/1650Mhz actual) it's actually somewhat difficult to get the GPU to pull 300W+ but with RTGI it'll run right into the power limit. I think it's worth testing multiple ways though because OC's on these cards behave differently stability-wise under moderate-heavy load vs maxed-out. In particular, Heaven and Superposition make a decent simple & free combination as Heaven with it's inconsistent utilization exposes high-clock-related instability whereas Superposition makes for a heroic effort that will push clocks low and show how things behave under extreme load. My UV-or-OC-no-in-between approach means I can't do anything about Power Limit on the high end- 330W is all I get until/unless I figure out a way to get the Liquid BIOS working properly with new drivers (from my brief, glitchy testing with a LC BIOS it seems like 360W is a good spot that will really keep clocks from floating)- but on the low end I've tried multiple approaches. From my experiences, using a low power limit (180W was what I used most for UV testing) can decrease stability in scenarios where the GPU is running at high utilization; when the load eases up suddenly and then increases again just as fast, huge clock spikes can occur that will cause crashes if the P7 is too high. There's more consistent stability to be had IME with leaving the PL at 220-240W and lowering actual power solely thru dropping the voltages. Of course, this results in power consumption that is both higher and very inconsistent but it can help with frequency spike related crashes. Don't spike the P7! I've found more stability and higher clocks to be had from keeping P6 & P7 moderate and very close together, than from pushing P7 to massive & unrealistic heights.

I don't really know much about OCing the HBM- unfortunately my card reacts very poorly to raised memory clocks so I've mostly kept it at stock. One of these days when I'm fully satisfied with the core clocks and stability I'll mess with the HBM again. As for the P-states, yes I do tweak all of them. I don't know how necessary it is- some folx will say to only touch P6 & P7 (and those are the ones to focus on obviously) but I mess with all of them because that's the kind of person I am.

edit- is there a way to tell my cards manufacture date from the serial #? I bought it from a local Craigslist seller this past May and IIRC he said he'd had it for a year-ish, so it could be early 2018?

Damn, good to hear clamping pressure helped. That's a significant drop in temps from that alone... I'd be stoked!
I'm always cautious with that but as long as it's even... worth a try :)

Thank you for explaining the different benchmarks, was always curious about the differences/use of each one as no one ever explains that, just that they are benchmarks lol! I used to just use Kombustor in the past but I don't think I got it going on the V64 this time around (barely had time to play with it other than some NMS - been away for 80% of the time I've owned it, stupid me.. should have bought it with me but the 290X came instead).

Clock spikes are interesting, I've heard that's a major issue with Vega stability as the transients on the PSU are massive... maybe worth looking there if you have a spare to dual power it? I've not had any issues at that stage only sustained load stability recently after some updates to drivers and game. Slowly creeping back to where it used to be though so might have been power related (swapped dual psu feeds to the card around...), will test more shortly to see if it's a plug order preference or just drivers/software changing.

Interesting re: PL left alone and just dropping voltages... 'writes notes down'. And also interesting re P6/P7. Reminds me of the 2600k once you get over 4.4 manual OC is a must due to pstates... Vega is very good at it stock so surprising to hear.

RE: HBM heard it's also more sensitive to heat.

You run your lower P-states much higher than mine. But I also have not run all the titles/benchies you do yet. Will do over this week and report back

upload_2019-11-16_1-15-20.png



PM'd s/n.. Nov 2018.
edit: can't PM you! too new account maybe? or I'm still on some bad goy list for shitposting a few times too much.

181836902184
Lol just realised you are a chick, awesome! Rare thing in the hardware world.
I keep seeing your name as like almost Maori language or similar and not individual words. Silly brain.
 
Last edited:
Back
Top