Associate
- Joined
- 28 Nov 2018
- Posts
- 337
- Location
- Monstrocity
That's nowhere near as slapdash as I was imagining! 
-
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.
The RX Vega 56 Owners Thread
- Thread starter Kaapstad
- Start date
I've been doing some more Tuning adjustments, and I thought I would share my findings.
The following may not seem like much at first glance, but gaining an additional 1% clocks at 1500+ MHz average on Vega can consume an additional 5 to 50 Watts as you go higher. Further, each additional 5 Watts increases the average fan speed on the reference fan by 50 to 100 RPM, depending on ambient temperature.
The Tuning profile settings can be observed in the below images. Note that both Tuning profiles have been tested to ensure stability.
Image - Tuning profile with original settings (as before)
Image - Tuning profile with adjusted settings
A standard 20 minute run of 3DMark Time Spy Stress Test was used to test the difference, with data carefully collected from only the start and to the end of the test. Two displays were used to run the test and simultaneously collect data using HWiNFO64 v6.32 from the secondary display. The below images capture both displays as part of a single desktop image.
Image - Time Spy Stress Test data for original settings
Image - Time Spy Stress Test data for adjusted settings
The important parts to look at are the average clock speed and average power usage. Also note that the clock difference will be slightly higher under standard gaming loads, and there should be further power saving under heavier workloads that push the average core clock towards the 1496 MHz mark.
The following may not seem like much at first glance, but gaining an additional 1% clocks at 1500+ MHz average on Vega can consume an additional 5 to 50 Watts as you go higher. Further, each additional 5 Watts increases the average fan speed on the reference fan by 50 to 100 RPM, depending on ambient temperature.
The Tuning profile settings can be observed in the below images. Note that both Tuning profiles have been tested to ensure stability.
Image - Tuning profile with original settings (as before)
Image - Tuning profile with adjusted settings
A standard 20 minute run of 3DMark Time Spy Stress Test was used to test the difference, with data carefully collected from only the start and to the end of the test. Two displays were used to run the test and simultaneously collect data using HWiNFO64 v6.32 from the secondary display. The below images capture both displays as part of a single desktop image.
Image - Time Spy Stress Test data for original settings
Image - Time Spy Stress Test data for adjusted settings
The important parts to look at are the average clock speed and average power usage. Also note that the clock difference will be slightly higher under standard gaming loads, and there should be further power saving under heavier workloads that push the average core clock towards the 1496 MHz mark.
Last edited:
A few notes and a final update.
If running significantly lowered voltages, changing GPU clock states below state 7 appears to have little or no effect. Any observed improvements were likely because of ambient conditions and slightly improved efficiency. This is possibly because between 960 and 965 mV appears to be the most efficient point, and going a little higher to 970 mV (975 mV actual) results in almost no performance gain without significantly improved cooling.
Vega is clearly designed to run with active throttling being the norm at stock settings, so changing GPU clock states could be useful in that configuration. However, even at lower voltages, GPU state 5 voltage should still be kept at least 10 mV (12.5 mV actual) below state 7 to allow momentary throttling to work as intended if needed in rare cases.
I did notice a strange anomaly when significantly increasing GPU state 6 clocks. I find this results in slightly increased maximum boost clocks, and this can be consistently shown in GPU-Z.
Image - GPU-Z Fullscreen Render Test with GPU clock state 6 at stock
Image - GPU-Z Fullscreen Render Test with GPU clock state 6 at 1590 MHz
I finally settled on the Tuning profile below.
Image - Tuning profile with final settings
Image - Time Spy Stress Test data for final settings
If running significantly lowered voltages, changing GPU clock states below state 7 appears to have little or no effect. Any observed improvements were likely because of ambient conditions and slightly improved efficiency. This is possibly because between 960 and 965 mV appears to be the most efficient point, and going a little higher to 970 mV (975 mV actual) results in almost no performance gain without significantly improved cooling.
Vega is clearly designed to run with active throttling being the norm at stock settings, so changing GPU clock states could be useful in that configuration. However, even at lower voltages, GPU state 5 voltage should still be kept at least 10 mV (12.5 mV actual) below state 7 to allow momentary throttling to work as intended if needed in rare cases.
I did notice a strange anomaly when significantly increasing GPU state 6 clocks. I find this results in slightly increased maximum boost clocks, and this can be consistently shown in GPU-Z.
Image - GPU-Z Fullscreen Render Test with GPU clock state 6 at stock
Image - GPU-Z Fullscreen Render Test with GPU clock state 6 at 1590 MHz
I finally settled on the Tuning profile below.
Image - Tuning profile with final settings
Image - Time Spy Stress Test data for final settings
Last edited:
Bumping this thread as I've finally got around to upgrading my system to a Ryzen 5600x and should now be capable of getting the best out of this card.
I'd like to flash the bios to Vega 64 and also understand the best thing to do to get the most out of this card is to undervolt but I'm an absolute beginner and was hoping someone could point me at a guide (for beginners like myself!) to show me really clearly how to do both.
Thanks in advance for your help
I'd like to flash the bios to Vega 64 and also understand the best thing to do to get the most out of this card is to undervolt but I'm an absolute beginner and was hoping someone could point me at a guide (for beginners like myself!) to show me really clearly how to do both.
Thanks in advance for your help
Did you flash the bios first or just leave as is and follow the guide you just posted?
I just left the original BIOS on it.
Ok cool thank you, just watched that guide on youtube and really clear, I'll give that a go as a starter and wait to see if anyone can advise if it it worth flashing the bios to squeeze some extra performance out of it
Soldato
- Joined
- 23 Jun 2004
- Posts
- 4,881
- Location
- Blackburn
Definately worth flashing the bios in my opinion. I found overclocking the hbm gave me a larger performance increase than clocking the core. With a vega 64 bios I can clock the hbm upto 1100mhz where as 945mhz was the max I could get beforehand
When running highly efficient voltage levels on reference Vega, the power limiter needs to be put to +50% for stability.
I suspect this ensures voltage and power limitations remain at the bios level only, as the values are matched at this point, and therefore stops the driver from running its own algorithm. Interestingly, it's very similar to LLC in that sense, where it needs additional voltage to absorb the imbalance it creates due to artificially inserting itself into a natural electric process.
Anyway, in my experience, voltage levels at anything below approximately 1025 mV will invariably cause instability with the power limiter left at default, while a +50% setting will allow voltages as low as 950 mV at stock clocks.
I suspect this ensures voltage and power limitations remain at the bios level only, as the values are matched at this point, and therefore stops the driver from running its own algorithm. Interestingly, it's very similar to LLC in that sense, where it needs additional voltage to absorb the imbalance it creates due to artificially inserting itself into a natural electric process.
Anyway, in my experience, voltage levels at anything below approximately 1025 mV will invariably cause instability with the power limiter left at default, while a +50% setting will allow voltages as low as 950 mV at stock clocks.
Last edited:
After 3 years my msi blow out blower vega56 fan started to act up, so at the worst possible time as you cant buy cards atm.
So, time to do the party trick with third-party cooling solution then.
A Raijintek Morpheus II it then was to be.
First seated attempt didn't work out, 105c hot spot.
crash.
second seating after fan exchange as it was an old one and a bit thicker layer on the core.
Now it worked.
No temps over 83c after hours of gaming.
and silence in the library.
Was a bit of tinkering to get it done.
so all good.
So, time to do the party trick with third-party cooling solution then.
A Raijintek Morpheus II it then was to be.
First seated attempt didn't work out, 105c hot spot.
crash.
second seating after fan exchange as it was an old one and a bit thicker layer on the core.
Now it worked.
No temps over 83c after hours of gaming.
and silence in the library.
Was a bit of tinkering to get it done.
so all good.
Associate
- Joined
- 20 Feb 2021
- Posts
- 8
I hope I never need to replace the cooler on any graphics card. From what I've seen in videos I'd probably pull my hair out. Good that it worked for you eventually.
What I was a bit surprised with reading this thread, is a couple of people reporting hot spot temps over 100°C and pretty large differences between core and hot spot. Does it take a lot of time for the hot spot to settle? Is is very game dependent? I've run a match of War Thunder in 1440p everything maxed out and it went a bit higher (17°C above core temp) than with Furmark (14° above core temp).
By the way, you can breathe some extra life into your Vega 56 in War Thunder by switching to DX12. It gave me +15% fps over DX11 on Maximum preset in 1080p. It beats a Sapphire 5700 XT running in default DX11 mode.
What I was a bit surprised with reading this thread, is a couple of people reporting hot spot temps over 100°C and pretty large differences between core and hot spot. Does it take a lot of time for the hot spot to settle? Is is very game dependent? I've run a match of War Thunder in 1440p everything maxed out and it went a bit higher (17°C above core temp) than with Furmark (14° above core temp).
By the way, you can breathe some extra life into your Vega 56 in War Thunder by switching to DX12. It gave me +15% fps over DX11 on Maximum preset in 1080p. It beats a Sapphire 5700 XT running in default DX11 mode.
Furmark have driver limits.
if the program would run at full power it fry cards.
I had no choice, fan started to make noises and reached end of life.
due to HBM ram and how the temps are measured stuff can heat up rampantly.
still really happy with the change as noise was reduced massively due to the old blower wasn't silent.
Associate
- Joined
- 20 Feb 2021
- Posts
- 8
Oh and big thanks to whoever explained how the HBM clock works. I naively set the core states 6 and 7 to 932 mV and 1030 mV respectively, and wondered why the HBM clock wasn't going beyond 500 MHz in many instances. Today I learned that the HBM voltage needs to be reduced to that of the lowest core power state for which I want maximum memory performance, so I set the HBM voltage to 932 mV as well and it now stays at 900 MHz, which is a big boost in benchmarks and as an added bonus of course, the total power draw drops further and I'm not hitting the power limit of 180W any more and have more stable frame times. Mind blown, lol. Why doesn't AMD explain this stuff somewhere? =)
Final Fantasy XV go brrrr: ~10.000 pts on Standard/1080p (1640 Mhz @ 1030 mV)
EDIT: Fooled myself. The HBM voltage slider does nothing when you change it. It's the custom voltage toggle that, when enabled, lowers memory voltage and makes it run at 900 MHz.
Final Fantasy XV go brrrr: ~10.000 pts on Standard/1080p (1640 Mhz @ 1030 mV)
EDIT: Fooled myself. The HBM voltage slider does nothing when you change it. It's the custom voltage toggle that, when enabled, lowers memory voltage and makes it run at 900 MHz.
Last edited:
Soldato
- Joined
- 23 Jun 2004
- Posts
- 4,881
- Location
- Blackburn
Mine is still doing pretty good tbf, most games I get 80-100fps at 1440p. Running at 1630/1100mhz
Didn't realise I had notifications still on for this thread! Still stuck with my pulse. Upgraded to a 48" 4k oled as my monitor so its struggling a bit 
. OC has got more unstable too. Was hopeful to have upgraded by now of course. Not touched any new game releases in a year 
. OC has got more unstable too. Was hopeful to have upgraded by now of course. Not touched any new game releases in a year
Associate
- Joined
- 20 Feb 2021
- Posts
- 8
Another Pulse owner here:
Power Limit: 180W (stock)
Core: P7 = 1640 MHz @ 1010 mV / P4 = 1336 MHz @ 860 mV
Mem: 880 MHz @ some lower voltage
For the memory clock and voltage I used OverrideNTool and its soft power play table editor, because when I use Radeon Settings it was always dropping mem clock to 500 MHz until the core was in P7, making the card run higher core clocks (and power draw) to compensate in VSync scenarios.
EDIT: I have 3 fun ideas why only Pulse owners respond:
Power Limit: 180W (stock)
Core: P7 = 1640 MHz @ 1010 mV / P4 = 1336 MHz @ 860 mV
Mem: 880 MHz @ some lower voltage
For the memory clock and voltage I used OverrideNTool and its soft power play table editor, because when I use Radeon Settings it was always dropping mem clock to 500 MHz until the core was in P7, making the card run higher core clocks (and power draw) to compensate in VSync scenarios.
EDIT: I have 3 fun ideas why only Pulse owners respond:
- They are the only Vega 56's that still work.
- They are the only Vega 56's that sold in numbers.
- Sapphire owners are more invested, because Sapphire is the goto AMD board partner.
Last edited: