Radeon VII having problems?..

nEo717 said:
I found URL on EKWB website a few days back for Radeon VII from google search for Radeon VII coolers... The card is also already in their configurator - I'll see if I kept the URL to the block when get home, if so, I'll post URL.
Click to expand...

I see that it is in their cooling configurator, but when you get to the "what to do you want to WC" of course GPU cannot be selected..
 
Maddness said:
I think pressure should be put on them to change this. For such an expensive video card, that's not good enough.
Click to expand...

wouldn't be surprised if it had something to do with the AIB's so that there's a reason for customers to buy their cards over direct from AMD. just my guess, but you may be right some pressure could get them to change it.
 
I would suggest warranty going forward is being reviewed. Let's face it, it has been a while since AMD had been in direct card sales in any volume. This is going to change and I would expect updated warranties to follow.
 
IdiotInCharge said:
You can spend significantly more and still get a one year warranty ;)
Click to expand...

I realise that. But don't the AIBs give a 3 great warranty on these cards. Please correct me if I'm wrong. Also Nvidia offers 3 years on there 20 series. They do need it though, with all there problems.
 
One can bet if you have an issue after one year AMD would actually help. XFX does the same too. Then again, we don't have to worry about space invaders either :p
 
Sapphire has a 2 year. I'm not sure about the others. And I don't think a 1 year warranty on this product is impressive either. It's not.
 
I don't think the 1 year warranty is really all that big of a deal unless there is something actually wrong with the cards; I would wager nearly everyone has, or has used, a video card that is more than a decade old and still has no issues. I have a system at work that has an X58 mobo with GTS250 and i7 975 that has been running 24/7 for nearly it's entire life and still kicking; it's very likely a non-issue.
 
Captindecisive said:
I don't think the 1 year warranty is really all that big of a deal unless there is something actually wrong with the cards; I would wager nearly everyone has, or has used, a video card that is more than a decade old and still has no issues. I have a system at work that has an X58 mobo with GTS250 and i7 975 that has been running 24/7 for nearly it's entire life and still kicking; it's very likely a non-issue.
Click to expand...

It's totally not a big deal until you actually need it and then you're SOL.
 
Maddness said:
It's totally not a big deal until you actually need it and then you're SOL.
Click to expand...

It's true, but you could literally say that about anything; if there did happen to be a real issue with the cards, I think AMD would either honor a return past the warranty period, or at least be forced to do something about it if they knowingly sold hardware that was going to fail in a short period. Not necessarily ideal, but I'm still not really concerned about it. As Kyle said, it's fairly likely they'll go ahead and extend it anyways after some discussion.
 
So, just got my VII on lunch and started working on the temps/tweaking stuff.

System:
Intel Core i5 9600k @4.8GHz (-3 AVX offset)
EVGA Z370 Micro
16GB Corsair Vengeance LPX DDR4 2400
AData SX8200 Pro 512GB NVME
Sound Blaster X-AE5
Seasonic Prime Titanium 850w
AMD Radeon VII
Noctua NH-U12S
NZXT H400 micro - 2 NZXT Aer F front intake, 1x Noctua Chromax NF-S12 rear exhaust, 1x Noctua Chromax NF-A15 top exhaust (all fans set for 1050rpm)
Windows 10 v1809
Adrenalin 19.2.2

I'll keep this post updated/edited as I move along with my testing.

Currently:
I have been running Furmark at 1280x720 with 4xAA for about an hour | Min FPS: 161 - Max FPS: 189 - Avg FPS: 176. Ambient temps ~70-73F
Max Junction temp = 111c | Hovering in the 107-111 range, when it hits 111 it drops the GPU clock to 1500-1520MHz for a second or two until Junction drops back down
Max GPU temp = 84c | It has been sitting at this temp since it reached the peak using default fan profile
Max GPU speed = 1764MHz | This was early on and after heat soak it has been bouncing between 1700-1720MHz
Memory speed = 1000MHz | This has been steady with no variations
GPU PWR = Seeing the GPU PWR jump between 218-228w and drops down to 150-170w when Junction temp gets to 111 and it drops the speed

I spent the first half hour running Furmark at the stock 1042mv @ 1801MHz on the Wattman scale
This current session has been running at 940mv @ 1801MHz on the Wattman scale

Furmark seems to not care about the change in voltage to the GPU, this was expected as it did the same when I was testing with my Vega 56 with 64 bios with undervolting. I will mess more with undervolting in gaming/benchmark scenarios and report back those findings.

Once I've finished with all the testing at stock and finding the undervolt limit, I'll be re-doing the testing with the washers under the retention screws to report temp difference and any undervolting changes I see.

Here is a comparison of a quick test using Time Spy stress test in 3D Mark - 20 Passes (note 1043mv is stock)



Of note, at 940mv the temp of the GPU itself was higher, but the fans were running 700rpm less without a change in clock speeds but with a very noticeable change in power draw throughtout the 20 loops. Bare in mind that this is just a core mv drop, I have not yet started messing with the Power Target. I'll be doing a lot of testing in the next few days before changing anything and post all the details and logs that I can for those interested. I'll actually likely just create a new post with the entirety of my testing and logs in the Video Cards forum and link to it from this post.

Likely just within margin of error, but it passed the 20 loop stress test with 99% at 940mv and 98.6% at 1043mv

3D Mark Time Spy @ default settings @ 940mv = 8403
3D Mark Time Spy @ default settings @ 1043mv = 8413

This is ~30% higher than my previous result with my V56 @ V64bios with core at 1550 @ 1000mv and memory at 1050MHz @ 1020mv


Looks like 940mv is the lowest I can go with this card on stock cooling without any washers. Going to 935mv causes crashes and instability, regardless of power target, so settled at 941mv. I'll use the stock 1043mv and undervolted 941mv for testing and logs at stock.


Here's a 20 loop Time Spy at 941mv with Power Target set to +10%. I don't see any drops below 1700MHz on this one and still no power usage over 198w when the stock 1043mv at 0% Power Target hit 233w. Stability passed at 99.6% on this run. I'll be keeping these settings in a sheet so I can replicate them with in-game testing as well.






I've done a lot of testing with multiple games in long sessions (8-12 hours) since I've gotten the card. I haven't had any noticeable differences in the numbers from the Timespy stress test sheets posted here. As such, I don't see any real reason to post anything other than the Timespy stress test numbers so I'll be redoing those without the washers in multiple configurations and then redoing those same tests in the same way with the washers installed. I'll post up those sheets in a separate thread when I'm done in the next few days.
 

Attachments

  • IMG_20190214_130922.jpg
    IMG_20190214_130922.jpg
    571.4 KB · Views: 0
Last edited:
This thing's dead in the water now that Nvidia bent over and started supporting FreeSync.

AMD actually has to compete on performance now.
 
Been tinkering with mine for about 3 hours now. Seems stable at 2000/1000, fans locked at 75%. Actual clock is bouncing around 1930-1960. Temps are low 60s during play.

Will say while tinkering with UV/OC, when it crashes, it crashes hard now. Much uglier than I've seen in a long time. There is no "recovery" in windows. Almost always a hard or soft lock. Reset key is needed.
 
jbc029 said:
Been tinkering with mine for about 3 hours now. Seems stable at 2000/1000, fans locked at 75%. Actual clock is bouncing around 1930-1960. Temps are low 60s during play.

Will say while tinkering with UV/OC, when it crashes, it crashes hard now. Much uglier than I've seen in a long time. There is no "recovery" in windows. Almost always a hard or soft lock. Reset key is needed.
Click to expand...

I will have WC'ing testing in the next week finished...Parts are still rolling in from Amazon as we speak. I hope to be able to actively WC the VRMs but if that does not work I have some heavy duty copper heatsinks with quality thermal tape ready. The main limiting factor is that my 75+lb case is sitting about 6" off the ground and with my spine/pelvic injury I cannot easy pick it up and move it. I have VEGAs in that loop now, and while I do have a spare PCI-E 16X slot open, it is only a 2.0 lane from the chipset which will really impact performance. Also unsure how well a VII will play with RX VEGAs together.
 
  • Like
Reactions: GHRTW
like this
I just got my VII this afternoon from Fedex. Too bad I'm still at the office and won't get to dick around with it until 7 or 8pm when I get home. I'm curious to see how it does vs my current Titan Xp.

As for hard crashes though - the latest NV drivers have been doing that to my Titan Xps (both in SLI and single card setups), so crashy drivers is kind of a thing everywhere. I'm not going to be pointing fingers to any one particular side of the fence on that one.
 
jbc029 said:
Been tinkering with mine for about 3 hours now. Seems stable at 2000/1000, fans locked at 75%. Actual clock is bouncing around 1930-1960. Temps are low 60s during play.

Will say while tinkering with UV/OC, when it crashes, it crashes hard now. Much uglier than I've seen in a long time. There is no "recovery" in windows. Almost always a hard or soft lock. Reset key is needed.
Click to expand...

I noticed during my testing that it only hard crashes if you go too low. I hard crash at 925mv or below at stock speeds, but only soft crash from 926mv to 939mv.
 
jbc029 said:
Been tinkering with mine for about 3 hours now. Seems stable at 2000/1000, fans locked at 75%. Actual clock is bouncing around 1930-1960. Temps are low 60s during play.

Will say while tinkering with UV/OC, when it crashes, it crashes hard now. Much uglier than I've seen in a long time. There is no "recovery" in windows. Almost always a hard or soft lock. Reset key is needed.
Click to expand...
What are your voltage and power target settings?
 
1.15 V, power target +10%. 1950 was stable with no power target increase and 1.05V.

Edit: HBM was stable at 1100 also, but I didnt see any improvment at all from it compared to the stock 1000, from what I could see. So I put it back to stock.

Edit 2: should also note that my ambient temp is lower than normal also. Rig is in the basement and temp is normally around 50 F during the winter down there.
 
Last edited:
9820 Timespy https://www.3dmark.com/spy/6250557

13520 Firestrike Extreme https://www.3dmark.com/fs/18348938

23015 Firestrike https://www.3dmark.com/fs/18317741

7416 Firestrike Ultra https://www.3dmark.com/fs/18347438

4906 Timespy Extreme https://www.3dmark.com/spy/6251662

6102 Superposition. 1080p extreme.

Running at 2000/1200 at 1088mv. It's taken everything I can throw at it at that voltage. I've seen it boost to 2045 with the profile thou. +20% power target.

Been trying for 2035/1200. I get 9980 Timespy with that. Almost stable but needs more juice. I've noticed on my card if I lower HBM to 1150 my jct temps drop by 7-8c. I was flirting with 2100/1150 but ran out of voltage room. With HBM at 1200 Temp is my limiter though.


https://pastebin.com/2qWE8mC2
 
Last edited:
Looks like some OCing Fun cards the Radeon 7 is. Old Vega temperature was king on OCing. Lowering the voltage allowed cooler temps and thus more OCing. Raise voltage and power drawl and heat just killed how much you could do.
 
Maye88 said:
9820 Timespy https://www.3dmark.com/spy/6250557

13520 Firestrike Extreme https://www.3dmark.com/fs/18348938

23015 Firestrike https://www.3dmark.com/fs/18317741

7416 Firestrike Ultra https://www.3dmark.com/fs/18347438

4906 Timespy Extreme https://www.3dmark.com/spy/6251662

6102 Superposition. 1080p extreme.

Running at 2000/1200 at 1088mv. It's taken everything I can throw at it at that voltage. I've seen it boost to 2045 with the profile thou. +20% power target.

Been trying for 2035/1200. I get 9980 Timespy with that. Almost stable but needs more juice. I've noticed on my card if I lower HBM to 1150 my jct temps drop by 7-8c. I was flirting with 2100/1150 but ran out of voltage room. With HBM at 1200 Temp is my limiter though.


https://pastebin.com/2qWE8mC2
Click to expand...
You should try to run Firestrike Ultra stress test (min 20 runs).
 
You must log in or register to reply here.
Back
Top