Expecting too much from Vega 64?

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
Let me preface this a bit. I'm not too familiar with modern video cards since I was running AMD 7950's in trifire for years. I got a killer deal on a Vega Frontier Edition after the mining boom but just haven't had the time to run any games on the system until recently. Now that I have, the Vega's performance is very strange. I seem to only be getting fps between mid 20's through the 50's with no major spikes or lulls in performance regardless of the game or video settings. I thought maybe it was a cpu bottleneck but its utilization will never jump above 70% on a single core. In GTA5 for example, I tested it with a video encoding and raw image processing in the background with not a single change to the FPS. I lowered the settings to 1920x1080 and the max it could pump out was 90 fps, my old trifire setup would do 130 fps at 2560x1600. In Subnautica, again lowering the graphic settings does nothing and it still only cranks out average fps in the mid 20's, far lower than my trifire setup.

It's running the latest Adrenalin 2020 drivers. I have a mild overclock on the ram and the core is running around 1600Mhz and it's not throttling from what I can tell. Am I doing something wrong? Three 7950's shouldn't be able to hold a candle to a Vega, or am I expecting to much from this card?

The rest of my systems specs are in my sig.

Any help is appreciated.
 

ssnyder28

2[H]4U
Joined
May 9, 2012
Messages
3,705
It may have something to do with your drivers not playing nice? You titled your post that you had a Vega 64 and put in your post that you have a frontier edition. I've heard people say they've had issues getting AMD gaming drivers working properly.
 

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
It may have something to do with your drivers not playing nice? You titled your post that you had a Vega 64 and put in your post that you have a frontier edition. I've heard people say they've had issues getting AMD gaming drivers working properly.
Aren't the driver issues only related to stability/black screens not actual performance? I've messed around with the drivers quite a bit and updated them last night. The performance is only slightly better. The system is running the adrenalin drivers, not pro drivers since they don't allow custom fan curves. The frontier edition is supported by the gaming drivers since its just a Vega 64 with more ram.
 

RazorWind

Supreme [H]ardness
Joined
Feb 11, 2001
Messages
4,248
What does the GPU load/temperature/frequency look like when it's exhibiting this poor performance?
 

chameleoneel

Supreme [H]ardness
Joined
Aug 15, 2005
Messages
5,794
What you are describing is a classic sign of CPU bottlenecking. Your CPU launched in 2012 and is about 1,000mhz slower than most modern CPUs.

At the veryleast, you should use Display Driver Uninstaller aka DDU. And do a clean wipe of the drivers. But in my personal opinion, a major GPU upgrade should be coupled with a re-install of windows. Or in the case of Windows 10, refreshing the installation (keeps your data, wipes Windows clean).

However, I really think you are hardware bottlenecked, here. Doing those things may help a bit. But I do not think you are going to find a magic fix for this. Because I think your Vega is CPU bottlenecked.
 

Dayaks

[H]F Junkie
Joined
Feb 22, 2012
Messages
8,873
Run a timespy and see if you at least get a normal score there.

GTA V is cpu heavy. There’s a good chance of a cpu bottleneck.
 

ssnyder28

2[H]4U
Joined
May 9, 2012
Messages
3,705
Aren't the driver issues only related to stability/black screens not actual performance? I've messed around with the drivers quite a bit and updated them last night. The performance is only slightly better. The system is running the adrenalin drivers, not pro drivers since they don't allow custom fan curves. The frontier edition is supported by the gaming drivers since its just a Vega 64 with more ram.

I am not sure because I don't know enough on the subject. I just know that the FE card is a professional card and i've heard people in the past run into issues but i don't have any specifics. To me it doesn't seem like a cpu bottleneck because your trifire 7950s performed much better and really should be slower overall than a vega FE.
 

RazorWind

Supreme [H]ardness
Joined
Feb 11, 2001
Messages
4,248
What you are describing is a classic sign of CPU bottlenecking. Your CPU launched in 2012 and is about 1,000mhz slower than most modern CPUs.

At the veryleast, you should use Display Driver Uninstaller aka DDU. And do a clean wipe of the drivers. But in my personal opinion, a major GPU upgrade should be coupled with a re-install of windows. Or in the case of Windows 10, refreshing the installation (keeps your data, wipes Windows clean).

However, I really think you are hardware bottlenecked, here. Doing those things may help a bit. But I do not think you are going to find a magic fix for this. Because I think your Vega is CPU bottlenecked.
Not that I fundamentally disagree with you, but he talked about how he went to a single Vega64 from three 7950s and his performance got worse. Shouldn't it just be the same if he's CPU bound, but now with a faster graphics card?
 

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
What does the GPU load/temperature/frequency look like when it's exhibiting this poor performance?
It's sitting at right around 1600Mhz on the core, 1125Mhz on the RAM and about 75C at full load.


What you are describing is a classic sign of CPU bottlenecking. Your CPU launched in 2012 and is about 1,000mhz slower than most modern CPUs.

At the veryleast, you should use Display Driver Uninstaller aka DDU. And do a clean wipe of the drivers. But in my personal opinion, a major GPU upgrade should be coupled with a re-install of windows. Or in the case of Windows 10, refreshing the installation (keeps your data, wipes Windows clean).

However, I really think you are hardware bottlenecked, here. Doing those things may help a bit. But I do not think you are going to find a magic fix for this. Because I think your Vega is CPU bottlenecked.
This was a fresh install of windows, the 7950's were never installed under windows 10. I DDU'ed the drivers the other night when when I installed adrenalin 2020 as well with barely any change.


Not that I fundamentally disagree with you, but he talked about how he went to a single Vega64 from three 7950s and his performance got worse. Shouldn't it just be the same if he's CPU bound, but now with a faster graphics card?
That's what I'm wondering. I know my CPU's are older, but they aren't even close to being fully utilized. I'm not expecting ludicrous performance, just not worse than what I had before. Under gaming loads the GPU will be at 100% use. The few CPU cores that are being used by the game will be around 50%, maybe with a single core getting around 75%. That's pretty far from being the biggest bottleneck in my book.

Why I'm finding this strange is that it's across the board, every game has suffered. It's almost like new vega card has less computing power than the three cards it replaced.
 

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
I am not sure because I don't know enough on the subject. I just know that the FE card is a professional card and i've heard people in the past run into issues but i don't have any specifics. To me it doesn't seem like a cpu bottleneck because your trifire 7950s performed much better and really should be slower overall than a vega FE.
Ah, I've been out of the tech loop for a couple years so I hadn't heard if there's some new driver issues. Most of the problems with the FE cards surround getting the gaming driver to install. There are workarounds, it's just that you don't get the switchable driver feature that AMD advertised.
 

rhansen5_99

2[H]4U
Joined
Nov 12, 2001
Messages
2,153
Hmm, based on you comments I would look at the bios installed on the card. Because you say mining maybe it has a tweaked for bios. You could check with gpuz and cross reference withe tech powerup to make sure. If not that then maybe a thermal limit or power limit is in play?
 

sirmonkey1985

[H]ard|DCer of the Month - July 2010
Joined
Sep 13, 2008
Messages
22,230
i would check other games besides GTAV and cross reference with reviews from the card to see if they match.. GTAV is notorious for performing very poorly on AMD cards and also being very cpu bottlenecked for no reason. the bios theory might be possible as well since most miners that actually took care of their hardware typically undervolted the AMD cards through bios mods. also make sure that you don't have "Chill" enabled in the drivers. ultimately though the vega frontier edition isn't a vega 64, it's a semi professional compute/rendering card and not meant to be a gaming card even though it can still game.
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
Let me preface this a bit. I'm not too familiar with modern video cards since I was running AMD 7950's in trifire for years. I got a killer deal on a Vega Frontier Edition after the mining boom but just haven't had the time to run any games on the system until recently. Now that I have, the Vega's performance is very strange. I seem to only be getting fps between mid 20's through the 50's with no major spikes or lulls in performance regardless of the game or video settings. I thought maybe it was a cpu bottleneck but its utilization will never jump above 70% on a single core. In GTA5 for example, I tested it with a video encoding and raw image processing in the background with not a single change to the FPS. I lowered the settings to 1920x1080 and the max it could pump out was 90 fps, my old trifire setup would do 130 fps at 2560x1600. In Subnautica, again lowering the graphic settings does nothing and it still only cranks out average fps in the mid 20's, far lower than my trifire setup.

It's running the latest Adrenalin 2020 drivers. I have a mild overclock on the ram and the core is running around 1600Mhz and it's not throttling from what I can tell. Am I doing something wrong? Three 7950's shouldn't be able to hold a candle to a Vega, or am I expecting to much from this card?

The rest of my systems specs are in my sig.

Any help is appreciated.
Did you use DDU to clean out your drivers first? or use AMD Driver Factory Reset during installation? The old drivers may have incorrect AMD configurations in the registry.

The default fan curve for Vega FE is very mild maxing out fan speed to like 40% -> Manually up the fan to like 70% and see if it improves the results.

Can you run 3dmark TimeSpy, I can compare it to my Vega FE on my Ryzen 1700X rig using most recent drivers and we can go from there. I do have a very old TimeSpy test here with single card at stock, most of my tests were with 2x Vega FEs
http://www.3dmark.com/spy/3022568
 

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
Hmm, based on you comments I would look at the bios installed on the card. Because you say mining maybe it has a tweaked for bios. You could check with gpuz and cross reference withe tech powerup to make sure. If not that then maybe a thermal limit or power limit is in play?
I just checked the bios, everything looks alright and is running at stock settings with no throttling or power limit set. Radeon chill isn't enabled either.

Did you use DDU to clean out your drivers first? or use AMD Driver Factory Reset during installation? The old drivers may have incorrect AMD configurations in the registry.

The default fan curve for Vega FE is very mild maxing out fan speed to like 40% -> Manually up the fan to like 70% and see if it improves the results.

Can you run 3dmark TimeSpy, I can compare it to my Vega FE on my Ryzen 1700X rig using most recent drivers and we can go from there. I do have a very old TimeSpy test here with single card at stock, most of my tests were with 2x Vega FEs
http://www.3dmark.com/spy/3022568
Yes, DDU was used and cleared out the old drivers before adrenalin 2020 installed. My fan curve is pretty aggressive with the GPU hitting about 75C

Here's the result of the Timespy run. It's actually a bit higher than yours despite my slower CPU's. My overclock was wasn't stable so it's just a bit faster than stock.
https://www.3dmark.com/3dm/44047569
 

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
Here's a screenshot of the settings and CPU usage in timespy.

vega_FE_settings.png
 

primetime

Supreme [H]ardness
Joined
Aug 17, 2005
Messages
7,292
Is the card using the original unmodded bios it was shipped with? If not thats your problem

Also you didnt post your link for timespy score...so it was useless information

Here is what mine does stock https://www.3dmark.com/spy/5458750
7 423
Graphics Score7 233
CPU Score8 725
 

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
Is the card using the original unmodded bios it was shipped with? If not thats your problem

Also you didnt post your link for timespy score...so it was useless information

Here is what mine does stock https://www.3dmark.com/spy/5458750
7 423
Graphics Score7 233
CPU Score8 725
The link is at the bottom of my response to noko. Here it is again.

Few MHz over stock
Scored 6938
https://www.3dmark.com/3dm/44047569

Changed the power settings but otherwise same settings for the core and ram as the last run.
Scored 7359
https://www.3dmark.com/3dm/44048704
 

primetime

Supreme [H]ardness
Joined
Aug 17, 2005
Messages
7,292

primetime

Supreme [H]ardness
Joined
Aug 17, 2005
Messages
7,292
i run as many games as possible at 3200x1800p in some cases...i can only use 75hz for time being anyway so anything above that doesnt really do me any good. I wish my card had 16GB of HBM2 vram ;) May as well try and put it to use
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
I just checked the bios, everything looks alright and is running at stock settings with no throttling or power limit set. Radeon chill isn't enabled either.

Yes, DDU was used and cleared out the old drivers before adrenalin 2020 installed. My fan curve is pretty aggressive with the GPU hitting about 75C

Here's the result of the Timespy run. It's actually a bit higher than yours despite my slower CPU's. My overclock was wasn't stable so it's just a bit faster than stock.
https://www.3dmark.com/3dm/44047569
I will run current drivers, current 3DMark build etc. but it does look like as discussed above you are CPU limited. My Vega FE's undervolt extremely well which can make it perform very good including excellent HMB OCing. I will see if I can capture that to give you come comparisons to go by.
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
Ah, I've been out of the tech loop for a couple years so I hadn't heard if there's some new driver issues. Most of the problems with the FE cards surround getting the gaming driver to install. There are workarounds, it's just that you don't get the switchable driver feature that AMD advertised.
The switchable driver ended I believe on the second Qtr 2019 Pro drivers, 19Q2, the gaming driver that worked with the switchable driver was always behind from the regular gaming drivers so newer games would take over 3 months if not longer. There was a way to get the new gaming drivers to load but it was a pita especially if you had two cards like I did. The registry fix to load now the new drivers works better than all the previous old method. In this case for this card, it was driver hell for good gaming drivers, the Pro drivers were very good for doing professional applications, except they were even worst for newer games than the switchable gaming drivers and had no way to OC your card plus utilities were block in OCing the card, making them pointless if you wanted to game with intensive graphical games.
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
Some 3dMark TimeSpy data points for Vega FE maxing out performance, current optional drivers 20.2.1. For those interested or have a Vega FE. The biggest key for performance increase for an Air Cooled Vega card is HBM memory speed, which I will show.
  • HBM is very sensitive to temperature, keeping it less than 65c will give you a lot of headway for performance increases from OCing the memory, meaning very aggressive fan curves will help dramatically
  • Vege FE was mostly memory restricted but once you crank up the memory speeds it becomes cooling sensitive, in my two Vega FE's OCing the memory gave the most bang or performance increase
  • Once HBM was OC I had no headroom with the GPU core clock mostly due to heat, raise core clock -> HBM temperture goes up making HBM OC lower and overall performance went down. You can increase performance significantly but will eventually hit a heat wall with the blower style cooler. Undervolting the GPU then helps in controlling power and subsequent heat
  • Realistically you are pushing close to 400w out of the card at this point and that blower actually does an amazing job for a blower but to push it further you will need to go to something like water cooling which for me was not worth the hassle for the remaining potential of the card. For others it might be fun, I was tempted to remove the Vega64 Liquid Cooler off and put it on one of the Vega FE's but didn't
All test was done on a 1700x at default CPU speeds using Precision Boost using 16gb of 3200mhz ram using the XMP profile 14-14-14-34
  1. Baseline, Default driver settings Graphics score of 6582. I am using just the graphics score which isolates the GPU performance from the CPU which was relatively consistent between all the test (same settings)
    1. http://www.3dmark.com/spy/10678155
  2. Default driver settings, + fan curve -> 85%@65c. Graphics score of 6846. This is just changing the fan speed curve so that the GPU/HBM is kept cooler. Just keeping the GPU/HBM cooler gave +3.9% Graphics Test 1 and +4.1% on Graphics Test 2
    1. http://www.3dmark.com/spy/10678331
  3. Stock GPU Settings, + fan curve -> 85%@65c.+50%pwr. Graphics score 7291. Upping the power limit now as well. +10.6% for Graphics Test 1 and 10.9% for Test 2 over Baseline. This effectively allows the core clock to go higher.
    1. http://www.3dmark.com/spy/10678438
  4. Stock GPU Settings, + fan curve -> 85%@65c.+50%pwr, HBM 1000mhz. Graphics score 7452. Test 1 + 12.3%, Test 2 12% over baseline. First OC of HBM
    1. http://www.3dmark.com/spy/10678790
  5. Stock GPU Settings, + fan curve -> 90%@65c.+50%pwr, HBM 1050mhz 1020mv, GPU UV 1175mv. Graphics score 7574. Test 1 and Test 2, +15.1% over baseline. Here HBM is being OV slightly plus GPU UV from 1200mv to 1175mv
    1. http://www.3dmark.com/spy/10679213
  6. Stock GPU Settings, + fan curve -> 90%@65c.+50%pwr, HBM 1100mhz 1050mv, GPU UV 1151mv. Test 1 +16.5% and Test 2 + 15.9% Last step for HBM overclocking with additional mem voltage, GPU is UV further down to 1151mv, the performance increase is plateauing
    1. http://www.3dmark.com/spy/10679388
Keeping core clock set to default, fan curve, UV gpu, OC HBM gave over a 16% boost. So why not increase the core clock? It does increase performance but Vega's last 100mhz-200mhz hits power like no other GPU I've ever tested. The issue is even 1625mhz, 25mhz higher than stock setting pushes temperatures up higher than expected, the blower cooler is just no longer sufficient, HBM temperatures goes up and you can no longer maintain the memory OC and performance actually will go the other way. Water cooling you can push that last 5%-10% out of the card plus you will be able to push the HBM even higher, some can do 1200mhz with their HBM.

Here is a graphical link for comparing the Baseline to the last OC:
https://www.3dmark.com/compare/spy/10678155/spy/10679388
 
Last edited:

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
The switchable driver ended I believe on the second Qtr 2019 Pro drivers, 19Q2, the gaming driver that worked with the switchable driver was always behind from the regular gaming drivers so newer games would take over 3 months if not longer. There was a way to get the new gaming drivers to load but it was a pita especially if you had two cards like I did. The registry fix to load now the new drivers works better than all the previous old method. In this case for this card, it was driver hell for good gaming drivers, the Pro drivers were very good for doing professional applications, except they were even worst for newer games than the switchable gaming drivers and had no way to OC your card plus utilities were block in OCing the card, making them pointless if you wanted to game with intensive graphical games.
You're spot on with that statement. The pro drivers are incredibly limiting, and while they work well for professional applications I just couldn't deal with the lack of fan curves. The registry workaround was a fantastic find, since it was so complicated to screw around with the drivers before that.



Some 3dMark TimeSpy data points for Vega FE maxing out performance, current optional drivers 20.2.1. For those that maybe interest or have a Vega FE. The biggest key for performance increase for an Air Cooled Vega card is HBM memory speed, which I will show.
  • HBM is very sensitive to temperature, keeping it less than 65c will give you a lot of headway for performance increases from OCing the memory, meaning very aggressive fan curves will help dramatically
  • Vege FE was mostly memory restricted but once you crank up the memory speeds it becomes cooling sensitive, in my two Vega FE's OCing the memory gave the most bang or performance increase
  • Once HBM was OC I had no headroom with the GPU core clock mostly due to heat, raise core clock -> HBM temperture goes up making HBM OC lower and overall performance went down. You can increase performance significantly but will eventually hit a heat wall with the blower style cooler. Undervolting the GPU then helps in controlling power and subsequent heat
  • Realistically you are pushing close to 400w out of the card at this point and that blower actually does an amazing job for a blower but to push it further you will need to go to something like water cooling which for me was not worth the hassle for the remaining potential of the card. For others it might be fun, I was tempted to remove the Vega64 Liquid Cooler off and put it on one of the Vega FE's but didn't
Wow! Thank's so much for the tips. That's way more than I expected. Like you said, Vega FE's are strange beasts when it comes to overclocking. Unless you mess with the settings you never actually get the full speed of the chip as rated since it starts throttling almost immediately. It seems counterintuitive to lower the core voltage but it works.

I'll have to mess with mine a bit more. Currently it's stable with the following settings. Looks like the ram can be pushed farther if the fan curve is changed.
80%@70C +50%pwr, HBM 1050MHz 1000mV, GPU 1615MHz 1150mV
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
You're spot on with that statement. The pro drivers are incredibly limiting, and while they work well for professional applications I just couldn't deal with the lack of fan curves. The registry workaround was a fantastic find, since it was so complicated to screw around with the drivers before that.




Wow! Thank's so much for the tips. That's way more than I expected. Like you said, Vega FE's are strange beasts when it comes to overclocking. Unless you mess with the settings you never actually get the full speed of the chip as rated since it starts throttling almost immediately. It seems counterintuitive to lower the core voltage but it works.

I'll have to mess with mine a bit more. Currently it's stable with the following settings. Looks like the ram can be pushed farther if the fan curve is changed.
80%@70C +50%pwr, HBM 1050MHz 1000mV, GPU 1615MHz 1150mV
Cool beans! Now my two Vega FE's mined for over a year, 24/7 with memory at 1100mhz (did not need to OV for mining, games are tougher on the memory and I have to OV) with negative power setting and lower clocks to optimize the hashrate/power. They behave exactly same as when I got them. Since they became non-profitable I use them for other stuff and soon they will be joined again hopefully on a Threadripper build. Looks like you got a very good sample there.
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
This is how a VegaFE OC compares to a 5700XT Anniversary Edition at stock speeds (card is factory OC). Just look at the graphics scores or Tests 1 and 2. For 3dMark you can compare the graphics scores even with different systems/cpus. These graphics test are still significantly GPU limited unless one has a very weak CPU. Anyways the 5700XT AE is about 19% faster, the VegaFE still has headroom left but I would need to have better cooling to take advantage of it. I could do a comparison to a 1080 Ti if you want.

VegaOCFeVs1080Ti.png
 

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
Cool beans! Now my two Vega FE's mined for over a year, 24/7 with memory at 1100mhz (did not need to OV for mining, games are tougher on the memory and I have to OV) with negative power setting and lower clocks to optimize the hashrate/power. They behave exactly same as when I got them. Since they became non-profitable I use them for other stuff and soon they will be joined again hopefully on a Threadripper build. Looks like you got a very good sample there.
Maybe it's luck, but more likely it's the fridged climate we have here. :D

Just pushed the hbm to 1100Mhz and scored 7447. Looks like I might be able to go higher even.
https://www.3dmark.com/3dm/44075118

Honestly it looks like my CPU isn't bottlenecking the GPU in any significant way. The fps in your graphic tests are practically the same as what I'm getting. That's rather impressive considering Xeon E5-2630v2's are 7 years old now.
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
Keep going, she can kick some nice ass! Mine is running great with latest drivers!

8184, 8090 GFX - https://www.3dmark.com/spy/10685563

I've giv'n her all she's got captain, an' I canna give her no more.

Like them temps?

View attachment 224760
Yes! how did you do that? If you can cool the HBM and GPU then you can really push the clocks for both. The golden temperature is below 60c which can be hard if you are pushing 400w plus.
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
Here is Sabrewulf Vega 64 compared to my EVGA 1080 Ti (Factory OC at default settings):

SabrewulfVegaOCFeVs1080Ti.png
I see you got the core clock up to 1700mhz with HBM at 1100 that is Vega64 Liquid Cool speeds for the GPU clock and a great HBM OC. On my Vega 64 LC I could get it close to 1800mhz unfortunately the HBM was not so friendly.

Can you do a Time Spy Extreme test so I can compare it to a 1080Ti and 5700XT?
 

NukeULater

Gawd
Joined
Sep 12, 2006
Messages
913
It's getting better every time but there's no way it'll be able to touch Sabrewulf's overclock. It's not stable above 1680Mhz. Who knows, maybe a little more tweeking would help.

90%@65C +50%pwr, HBM 1100MHz 1050mV, GPU 1680MHz 1190mV

Scored 7567
https://www.3dmark.com/3dm/44091899
 

bluestang

Limp Gawd
Joined
Dec 14, 2018
Messages
460
Undervolt and Overclock for Vega 64 :)
https://www.igorslab.de/vega-undervolting-heilsbringer-voodoo-oder-nutzlos-ein-selbstversuch/
Also, since it might be setup by the previous owner with a mining BIOS I would reflash with stock BIOS. But even that doesn't always get rid of the mining tweaks. So even switching/restoring to the original BIOS doesn't always take unless you run a tool because of the previoud modded BIOS that it may have had.

Sometimes you also have to run AMD-ATI-Pixel-Clock-Patcher and ran it as Administrator.
https://www.monitortests.com/forum/Thread-AMD-ATI-Pixel-Clock-Patcher
 

evolucion8

Gawd
Joined
Mar 15, 2006
Messages
917
Looks like CPU bottleneck, the IPC of that CPU along with the latency inherent of the CPU platform due to ECC among other stuff makes them unsuitable for gaming.
 

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
It's getting better every time but there's no way it'll be able to touch Sabrewulf's overclock. It's not stable above 1680Mhz. Who knows, maybe a little more tweeking would help.

90%@65C +50%pwr, HBM 1100MHz 1050mV, GPU 1680MHz 1190mV

Scored 7567
https://www.3dmark.com/3dm/44091899
That is pretty much what I run into, the cooler just can't keep the GPU/HBM cool. Really the HBM is mostly the culprit with temperature which unfortunately is coupled right next to it with a very high heat producing GPU when you push it. Undervolting the GPU allows it to run at a higher clock speed keeping the heat in check, but upping the clock further you just need better cooling. If you can maintain that HBM less than 60c or even cooler you will probably be pushing close to 1800mhz and past 1100mhz HBM.

The Vega 64 LC (I have one of these as well) cooler is fantastic for a reference design, at stock fan curve speeds/temp which keeps it less than 85c, I cannot even get above 975mhz on the HBM, lower it to less than 60c using custom Fan curve -> 1050mhz, 55c 1100mhz. The GPU core will do 1800mhz with stock or slightly increase HBM speeds if I keep it less than 65c. Just the nature and characteristics with these cards.

So if you want more, it will require much more work. Now if you can get a radeon Vega FE liquid Addition cooler -> Just the cooler, I've seen these sold on Ebay a few times (just the cooler) for a good price, that may work out good. I am confident but not sure the Vega 64 LC cooler will fit on the VegeFE since both are AMD reference boards. I've been tempted several times but it just seems a lot of work for 5%-10% more, it would be more for the fun of it and pushing the envelope more than actual benefit in my case. The VegaFE is a fantastically built card with the quietest and least annoying sounding blower of any blower card I've own, I've own many. The VegaFE does extremely well with Pro Render (can be used in Blender and Modo). The other route if you already have a water loop is buying a block and hooking it up which will cost some money and all the hassles of going into custom water loop cooling.

Your last TimeSpy compared to mine, just look at the Graphics Test scores which you are +2.5%. Your card is doing great.

VegaOCFeVsVegaFEOCi.png
 
Last edited:

noko

Supreme [H]ardness
Joined
Apr 14, 2010
Messages
6,720
Looks like CPU bottleneck, the IPC of that CPU along with the latency inherent of the CPU platform due to ECC among other stuff makes them unsuitable for gaming.
His CPU shows limitations but for the graphics score it is irrelevant. He has a +60mhz GPU advantage which gives a +2.5% graphics score. Actually GPU speed may have an actual variance here around that +60hz setting. So the graphical score is valid and showing max potential for his card with his current CPU. Yes for CPU intensive games like GTA V and others, his CPU may limit the max frame rates but if the gameplay is good even that becomes less relevant. Especially if the GPU is predominately at it's limit.

I should actually confirm this by changing my CPU speeds OC/UC to see how much if any the graphic score changes. If I have time, will do.
 
Joined
Jan 18, 2008
Messages
1,014
Here is Sabrewulf Vega 64 compared to my EVGA 1080 Ti (Factory OC at default settings):

I see you got the core clock up to 1700mhz with HBM at 1100 that is Vega64 Liquid Cool speeds for the GPU clock and a great HBM OC. On my Vega 64 LC I could get it close to 1800mhz unfortunately the HBM was not so friendly.

Can you do a Time Spy Extreme test so I can compare it to a 1080Ti and 5700XT?

Don't have Extreme, not buying for $9.99, sorry...
 

viivo

[H]ard|Gawd
Joined
Sep 7, 2005
Messages
1,698
this was/is kinda my thought....but what do we know :)

We know an orgy of synthetic GPU benchmarks has no bearing on the fact of a CPU bottleneck.

What was this thread about? Who cares dude my score went up by 11!!@1
 
Top