Question regarding core clocks on R9 cards

[ImpulsE69]

I just got a Sapphire Tri-X 290x (New Edition) and I'm attempting to do some testing on it. I haven't used a ATI card since a 4870 so I am not up to speed on the way these newer cards work.

When I am doing benchmarking, I am seeing what looks like hitching/stuttering and I can see the core clock go down rather than stay pegged at the top speed I gave it. This in turn makes the FPS go down, but I would think it would work the other way - ie: The core clock stay pegged, and FPS drop if it can't handle it. But instead the core will go up and down constantly from 1050hz down to 700hz or so for a moment, then back up at random. Almost like it is throttling down (maybe due to power?). Temps are stable in the 50-70 range.

Am I missing something here? I am just running a default test on a aged version of MSI Afterburner at the moment. I can see that the GPU usage goes up and down rather than staying at 100%. Running Win7 - SP1, Latest Catalyst Drivers and 2500k @ 4.2. I'm using a Corsair 650W PSU (coming off a GTX 480 - so powerwise they are similar, the Tri-X uses a bit more).

 

[Elixer]

What exactly are you testing?
Synthetic tests, actual games (which?).

 

[ImpulsE69]

What exactly are you testing?
Synthetic tests, actual games (which?).

Synthetic FPS stress testing (the one built into afterburner actually).

 

[Stuka87]

What are its temps, and what voltage is it set too? The card will throttle if it hits too high of a temperature.

And both AMD and nVidia do limit their cards during some stress tests due to high heat generation, such as furmark.

 

[SimianR]

I just got a Sapphire Tri-X 290x and I'm attempting to do some testing on it. I haven't used a ATI card since a 4870 so I am not up to speed on the way these newer cards work.

When I am doing benchmarking, I am seeing what looks like hitching/stuttering and I can see the core clock go down rather than stay pegged at the top speed I gave it. This in turn makes the FPS go down, but I would think it would work the other way - ie: The core clock stay pegged, and FPS drop if it can't handle it. But instead the core will go up and down constantly from 1050hz down to 700hz or so for a moment, then back up at random. Almost like it is throttling down (maybe due to power?). Temps are stable in the 50-70 range.

Am I missing something here? I am just running a default test on a aged version of MSI Afterburner at the moment. I can see that the GPU usage goes up and down rather than staying at 100%. Running Win7 - SP1, Latest Catalyst Drivers and 2500k @ 4.2. I'm using a Corsair 650W PSU (coming off a GTX 480 - so powerwise they are similar, the Tri-X uses a bit more).

The card will throttle the clockspeed down in a number of scenarios to save power. If you have vsync enabled for example and the refresh rate is at 60hz the card will throttle itself if it the framerate is at 60fps. It will also throttle if it gets too hot, uses too much power, and in certain scenarios I've seen it throttle in CPU limited scenarios as well. You could always go into the Catalyst Control Center and change the power limit to 20% to see if it's a power issue (at least through software). It could also be that in this case AMD has enabled some sort of power virus check (eg - furmark) to throttle the card in this particular synthetic test.

 

[ImpulsE69]

Wow, what an adventure. Per the advice of many forums, I installed Valley to do some benchmarking and fps comparisons on overclocking. I didn't take it up too much, and everything was working well. Seems whatever issue I was having was limited to the other program.

At some point I wasn't even doing anything, everything was idle and the screen goes black. I was like..oh man . I reboot to find Valley had installed some sort of browser plugin and a bunch of malware that took over my computer causing all sorts of glitches and crashes. Try as I might I couldn't get rid of it, and at some point as I played around MS decided my Windows was no longer valid.

I just quick restored an image from last night, and reinstalled the drivers and everything is good now.

To answer some of the questions above. I had Vsync off (was getting around 290FPS in the test I was doing). The temp never got above 70 during this time. I think it was just related to the the old program. It's an older, less bloated version. The valley test showed me what I needed to know before it went bonkers.

 

[KingFatty]

Where did you download your Valley install from, man that's crazy what happened!

Good to see the issue was related to your particular test, but I'm guessing it's not helpful for AMD to do those kinds of things to gimp the card temporarily without some info helping the user understand why the card seems to perform less than its best. I mean, maybe they can put a little notice in Catalyst control manager or something that mentions they are causing the funny behavior when the card drops performance or something, that it's on purpose.

 

[ImpulsE69]

I honestly thought I got it from their own site.

As an update, I found I still had the issue. I did some more googling and found this is a common issue. The issue ended up being that I had enabled overclocking in CCC, but was doing the clock changes in afterburner. I believe it is because there is some sort of power throttling in CCC. The solution was simply to disable overclocking in CCC and make the changes in afterburner.

For now that solves the issue. Once I get around to truly OC'ing it, I'll try to figure out the odd CCC settings for tweaking it.

 

[ImpulsE69]

I've about had it with this card already.

Some serious design flaws in terms of voltage control. Spent most the day working through the random idle black screens that plague these r200 cards that AMD seems not to want to fix. I would say almost 3 years is plenty of time to acknowledge and fix these issues.

In game it works great, but out of game when it isn't doing anything is annoying because apparently the memory voltage is tied to the core voltage and with all the throttling it can't keep enough voltage for the memory at idle.

I think I've finally worked around it by installing the latest AB and creating mulitple profiles (one for 2D one for 3D) then using Rivatuner to toggle between them automatically, but the fact I have to set this up for every fricking executable is downright stupid.

I may end up returning this and going with a 980 or ti instead.

 

[Hitman928]

You shouldn't have to do all that and you shouldn't be getting random idle black screens. Which card do you have in particular? If you uninstall afterburner, do you still have the issues?

3rd party oc utilities can cause issues if not designed for your specific card. Afterburner has been known to cause weird issues like this. You can look at this thread in the Nvidia section to see that it effects Nvidia too:

http://forums.anandtech.com/showthread.php?t=2432430

Try everything stock without any 3rd part apps installed and see if you still have the same issues.

 

[ImpulsE69]

You shouldn't have to do all that and you shouldn't be getting random idle black screens. Which card do you have in particular? If you uninstall afterburner, do you still have the issues?

3rd party oc utilities can cause issues if not designed for your specific card. Afterburner has been known to cause weird issues like this. You can look at this thread in the Nvidia section to see that it effects Nvidia too:

http://forums.anandtech.com/showthread.php?t=2432430

Try everything stock without any 3rd part apps installed and see if you still have the same issues.

It apparently isn't vendor specific, and it is well documented. I found 100's of threads/posts of people with the same issue dating back from 2013 to now on all ranges of 290/290x cards. No one fix seems to work for everyone. Many people RMA, but have the same issue with new cards. In some cases, PSU was thought to be the blame, but I think in many cases that was ruled out. (In games PSU is the culprit in many cases, but not in desktop idle).

Had I known to look for this prior to purchasing I probably wouldn't have. I was shocked to see how many people have this issue and it hasn't been addressed - or at least narrowed down to what exactly causes it 100%. I believe what I have done has 'fixed' it for now. I will run a few days like this and see how it goes. So far since I made the changes it's been good.

I read the thread you posted, and yea there is some similarity there. On the AMD side, it is due to voltage being set the same for core and memory, causing the memory to crash when it is clocked down and suddenly asked for full clock (supposedly). This happens regardless of overclocking or not. (Keep in mind, no it doesn't happen on all cards. It is thought to be related to the ASIC quality and voltage needs). I've read quite a bit about this, but still finding new things. It happens with stock settings, but I have not tried uninstalling AB and just running CCC (and honestly I have no interest in that since it doesn't give me any info externally).

I 'could' potentially just turn on constant voltage and force everything full throttle all the time, but I would rather not do that either. I just find it annoying I ran w/o issue on my 480 for 4 years.

 

[Elixer]

Hmm, most of the 'black screen' issues I have seen were because of a faulty card or PSU issues, and I think that some OEMs also had a new BIOS to fix voltage issues, but, unless you got really old stock, I don't think that would be it this time.
Haven't heard much about these kinds of issues lately though.

You know if you got the new version of the Tri-X (custom PCB), or the older version, based on the reference design?

 

[ImpulsE69]

This one: SAPPHIRE 100361-4L Radeon R9 290X 4GB 512-Bit GDDR5 PCI Express 3.0 Tri-X OC(UEFI) Video Card, New Edition

I don't know it could be, we shall see. Like I said, I and many others, can run in games (TW3, Mordor, etc etc) for hours and no issues. It isn't until it is sitting idle that it black screens.

One thing I didn't mention was I downclocked the memory to "reference" speeds on 2D. Since I've done that I haven't had any issues so far.

 

[Stuka87]

The only black screen issue I know of is the wake from display sleep bug. Disabling display sleep prevents the issue entirely. Just leave a screen saver going.

 

[ImpulsE69]

The only black screen issue I know of is the wake from display sleep bug. Disabling display sleep prevents the issue entirely. Just leave a screen saver going.

It's not that.

 

[brandonmatic]

It's not that.

Sorry to hear about the problems with your card. I also just got the Sapphire Tri-X 290X 4L (running on a CX600 PSU) and haven't run into that problem myself. Some online posts suggest that you try adding +25mV to the core VDDC as a fix. But maybe your best bet is to RMA and see if you can get a card that doesn't have this issue. Or maybe you can contact Sapphire to see if they can send you a new bios to fix the problem. It might be improperly spec-ed voltage.

Good luck and let us know how it turns out.

Edit: GPU-Z reports my VDDC as 0.984V at idle, ASIC quality as 78.6%, and I have a zero mV offset in Trixx. What are you running?

https://www.sapphireforum.com/showthread.php?33235-R9-290-Black-Screen

 

[ImpulsE69]

VDDC is .984 as well (with 2D profile to under clock memory to 1250 instead of 1350) I didn't add voltage but I read alot of threads about doing that. I opted for the less heat method.

ASIC is 71.1%

So far after making the change above I haven't had any further issues.

My 3D profile is a light OC of 1121/1451 and it is stable after running for hours. (I haven't really tried to push it).

Still not sure if I should RMA or not, but I have another week before i need to make that decision.

 

[SimianR]

The only time I've experienced the black screen "bug" is when I've tried undervolting too far, like -50mv or more. The second I set the voltage to default value it seems to fix it. It definitely does seem to be a voltage issue related to the memory (or controller).

 

[spat55]

I've about had it with this card already.

Some serious design flaws in terms of voltage control. Spent most the day working through the random idle black screens that plague these r200 cards that AMD seems not to want to fix. I would say almost 3 years is plenty of time to acknowledge and fix these issues.

In game it works great, but out of game when it isn't doing anything is annoying because apparently the memory voltage is tied to the core voltage and with all the throttling it can't keep enough voltage for the memory at idle.

I think I've finally worked around it by installing the latest AB and creating mulitple profiles (one for 2D one for 3D) then using Rivatuner to toggle between them automatically, but the fact I have to set this up for every fricking executable is downright stupid.

I may end up returning this and going with a 980 or ti instead.

The card has been out since November 2013 so not nearly even 2 years, not had an issue with my R9 290 really so must just be something bad in your rig not liking it.

 

[Shmee]

I got both of mine off ebay, and they have been rock solid. I am not OCing them though, I just leave them at stock.

 

[MTDEW]

I know Sapphire released an updated Bios for the R9 290 Tri-x cards awhile back that has a +25mv by default that fixes the black screen issue due to too low of memory voltage in 2D modes on some GPUs.
Any Sapphire Tri-x 290 OC should have the +25mv by default with bios version 015.044.000.011.000000 or later. LINK

Now as for the Sapphire R9 290X Tri-X OC , the newest Sapphire bios I have seen floating around is 015.046.000.000.000000 HERE.
So I can only assume that would have the same +25mv "fix" as the 290's, but I don't know for sure.

Easy way to check if your bios has the +25mv is to use MSI Afterburner and "reset to defaults" and see if your GPU defaults to +25mv.

Of course as you know,this still only helps if memory is run at or near stock speeds.
If you're overclocking the memory too much in 3D, it still doesn't completely solve the issue.
Because once you overclock the memory in 3D modes, it also overclocks the memory in some 2D modes where as you already know the extra 3D voltages dont kick in resulting in a blackscreen in 2D modes, yet fully stable in 3D modes due to the extra voltage needed kicking in in 3D mode.
(actually that's just an oversimplified explanation, it has more to do with the Internal memory controller not always being able to up the voltage fast enough when switching between different memory clocks in 2D mode which results in the blackscreen)

To keep the memory fully stable when overclocking it you either have to...
1:find your max stable memory overclock with using just 2D voltage and use that memory speed for both 2D and 3D.
2: Use Different profiles for 2D and 3D modes so you're not OCing the memory when stuck in 2D voltage.
3: Force a constant voltage where the memory and core always gets your selected 3D voltage whether in 2D or 3D mode. (which is a horrible solution IMO)

I found its best just to use a memory overclock that works just fine with 2D voltages and use that speed for both 2D and 3D and be done with it.
You don't really gain much OCing the memory vs overclocking the core anyway.
You're much better off concentrating on your core overclock if you want more performance.

Oh and you can max out your Power limit % to prevent unwanted throttling while gaming if its happening when it shouldn't be.
Your core will still throttle sometimes in less demanding games if you use vsync, it will throttle the core to just as much speed as it needs to maintain your "selected v-synced FPS" in that case.
Example: v-sync set a 60 - if the GPU can maintain 60FPS in that game with a speed of 700mhz, then that is what it will use with v-sync on.

 

[ImpulsE69]

I know Sapphire released an updated Bios for the R9 290 Tri-x cards awhile back that has a +25mv by default that fixes the black screen issue due to too low of memory voltage in 2D modes on some GPUs.
Any Sapphire Tri-x 290 OC should have the +25mv by default with bios version 015.044.000.011.000000 or later. LINK

Now as for the Sapphire R9 290X Tri-X OC , the newest Sapphire bios I have seen floating around is 015.046.000.000.000000 HERE.
So I can only assume that would have the same +25mv "fix" as the 290's, but I don't know for sure.

Easy way to check if your bios has the +25mv is to use MSI Afterburner and "reset to defaults" and see if your GPU defaults to +25mv.

Thanks for the info. Mine always was defaulted to +6mv. This particular card seems too new for anyone to have much on, but I did find mention somewhere of a bios revision newer than what I have directly from Sapphire (mine is newer than what you have listed). It also has Samsung memory as opposed most all having Hynix or Elpida.

Going to reach out to Sapphire. Everything has been stable as I mentioned - but I'd prefer not to have to do profiling.

 

[brandonmatic]

I have BIOS Version 015.048.000.038.000000 (UEFI bios). I have zero mV VDDC offset according to Trixx. Mine is also Samsung memory - I think all of the 4L Tri-X cards use Samsung memory.

 

[MTDEW]

Also when searching around for overclock advice on your 290x, it's best to keep in mind that Sapphires Tri-x cards don't use the same newer AMD Powertune that the reference and other 290/290x GPUs use to throttle that are based on both thermal and power limitations.

The Sapphire Tri-x core throttling is based on power limits only.(and of course TjMax) (and it's a GOOD THING!)
Which is why I recommended upping the power limit % to max when overclocking to prevent throttling.

You can read about the difference in the way the Tri-x version throttles right here in the Anandtech Review of the 290 Tri-X
Here are some quotes from the review.

For the 290 Tri-X OC Sapphire has reverted to traditional power and temperature based throttling, opting not to use the functionality of next generation PowerTune. This means that the 290 Tri-X OC does not offer the ability to throttle based on fan speeds, nor does it offer the ability to adjust the temperature it throttles at, instead throttling at Hawaii’s TjMax. This implementation caught us off guard at first since we had expected everyone to use next generation PowerTune, however as it turns out this is something that board partners get to decide for themselves on their customized cards.

Sapphire for their part has told us that based on the ample cooling performance of the Tri-X cooler that they've opted to use a traditional thermal management implementation in order to better sustain performance. Though we can’t readily test Sapphire’s statements about sustainability, we certainly can’t argue against Sapphire’s statement on the performance of their cooler. We’ll see the full breakdown in our benchmark section, but they are having absolutely no problem balancing noise and temperatures right now without next generation PowerTune.

Meanwhile for FurMark, the reference 290 would throttle here based on both thermal and power limitations, whereas the Sapphire 290 is only limited by power. As a result it’s able to maintain higher clockspeeds and hence higher power consumption levels than the reference 290. Or in other words, the reference 290 was held back by its cooler here, while the 290 Tri-X OC is held back only by its board power limits.

 

[MTDEW]

Also when searching around for overclock advice on your 290x, it's best to keep in mind that Sapphires Tri-x cards don't use the same newer AMD Powertune that the reference and other 290/290x GPUs use to throttle that are based on both thermal and power limitations.

The Sapphire Tri-x core throttling is based on its set power limits only.(and of course TjMax) (and it's a GOOD THING!)
Which is why I recommended upping the power limit % to max when overclocking to prevent throttling.

You can read about the difference in the way the Tri-x version throttles right here in the Anandtech Review of the 290 Tri-X
Here are some quotes from the review.

For the 290 Tri-X OC Sapphire has reverted to traditional power and temperature based throttling, opting not to use the functionality of next generation PowerTune. This means that the 290 Tri-X OC does not offer the ability to throttle based on fan speeds, nor does it offer the ability to adjust the temperature it throttles at, instead throttling at Hawaii’s TjMax. This implementation caught us off guard at first since we had expected everyone to use next generation PowerTune, however as it turns out this is something that board partners get to decide for themselves on their customized cards.

Sapphire for their part has told us that based on the ample cooling performance of the Tri-X cooler that they've opted to use a traditional thermal management implementation in order to better sustain performance. Though we can’t readily test Sapphire’s statements about sustainability, we certainly can’t argue against Sapphire’s statement on the performance of their cooler. We’ll see the full breakdown in our benchmark section, but they are having absolutely no problem balancing noise and temperatures right now without next generation PowerTune.

Meanwhile for FurMark, the reference 290 would throttle here based on both thermal and power limitations, whereas the Sapphire 290 is only limited by power. As a result it’s able to maintain higher clockspeeds and hence higher power consumption levels than the reference 290. Or in other words, the reference 290 was held back by its cooler here, while the 290 Tri-X OC is held back only by its board power limits.