Discussion in 'Videocards - AMD Radeon' started by OnnA, Jun 11, 2019.
As of 19.12.3 MSI Afterburner now works, no more flicker with a custom fan curve.
it worked in 19.12.1 and 19.12.2 but only on the desktop and games would still flicker but yes all games now seem to work properly @Unwinder
PCIe 3.0 Vs PCIe 4.0
are the powerplay tables still working on the rx 5700? what are the main disadvantages of using them? im worried that if i put the xt bios on it, it will brake after a few weeks/months to the point of not even being able to restore the og bios for warranty. the xt is only like 12% more expensive where i live, so its actually a better value if this mod didnt exist
SPPT's work fine yes
disadavantage? zero as long as its cooled properly
5700 cards do not like their mem chips to be oc'd much 80Mhz even with an SPPT allowing +50% power limit
Max I can run my card at is 1930/1830 at 1.081v +20% power limit, temps - core under 80c, mem under 90c
anything over that causes black screens or restarts
Hmm so a black screen on the new system as well.
Thus it's unrelated to the older aged PC hardware particularly stuff like RAM's.
It's unlikely though not written off entirely to be GPU hardware related though there are at least two separate black screen issues, idle and full load possibly four depending on if it's recoverable or a full crash/shutdown of the system when this happens.
Driver wise well 19.7.5 was always the most stable here though it has issues fixed in 19.8.1 though these all the way up until 19.12.1 have various trade-offs and aren't fully stable fixing some things but risking the black screen problem.
19.12.2 and 19.12.3 increases the frequency of these happening but how and what's causing them anyway.
GPU first party or third party designs changes to SOC voltages and clocks seem unlikely as well since the issue is reported for Sapphire, Gigabyte, MSI, Asus and so on including AMD's 5700, 5700XT and the 5700XTX or 50th AE variants all.
My own problems are wholly down to system idle black screens and not when the GPU is at any load such as gaming which might be potentially caused by Wattman and Overdrive being a bit so-so with what's inputted and what's actually used though even stock configurations encounter problems and again a mix of GPU being under load and idle and also varying from D3D9 to D3D11 to D3D12 and Vulkan and a mix of well running titles and AMD optimized ones to known problematic titles so not much there.
Replicating the issue is somewhat possible but very random, it's always when the GPU is boosting up which with the wonky behavior of Adrenaline 2020 and Wattman especially a less ideal steeper fan curve usually shows this by just moving the mouse cursor up or down whether HW acceleration of the browser is on or not or even when on the desktop of the OS itself as the GPU clocks should shoot up slightly and as a result the fan stepping immediately goes from the first stage to the second one due to these not being actual curves but more like stairs and shifting from one to the other directly with no gradual changes in fan RPM up or down but just switching completely from one state to the next as temps or GPU load changes even if it's slight.
Seems more common after a higher demanding GPU load such as when gaming but might be coincidence not entirely sure and kinda hard to narrow down a very random infrequent issue at all and some users don't encounter the problem what so ever though the ones that do and the reported hardware and GPU model is all over the place from Intel to AMD to older and the newest hardware and just about any AMD Navi GPU and driver combination including 19.7.x too for that matter so it's not like it's guaranteed to be stable far as I can tell at least though it's certainly worked best for me personally but again this entire issue is just incredibly random and hard to track down.
Next up well that's checking on a different display but that's going to take a while though the first thing will be a actual certified DP cable instead of this half decade old one, doubt it changes anything but worth a test.
I expect it's something in the drivers though but then what's the deal with that and it not affecting every system, minor hardware faults in the GPU seem unlikely too but there's so much I don't know or haven't learned on how Navi works though there's also regressions in stability for Vega and Polaris GPU models possibly others too particularly for Adrenaline 2020 so can't really say anything though it being a driver problem would be nice as that makes it a fully software related issue and then a fix in some newer driver to come.
What more is there to say, well not much I guess and I haven't learned much about the issue other than just reading reports on how it's still a thing and back and forth about why and how it happens and all kinds of suggestions which might or might not work, randomness and sporadic behavior says to me it's not like these do much but then there could be multiple conditions or factors for which black screen problem occurs from unstable overclocks and recovering or not from this to being idle or on a higher GPU workload and all sorts of possibilities that are a bit beyond my knowledge and understanding for this all works.
Would *really* like a way to fully lock the GPU to P1 instead of P0 to call them this so instead of zero mode it'll be forced to use the low clock voltage and clocks particularly voltage since it still fluctuates but even with a lower GPU load this thing *is* a curve so values from dialing in manual values can be -+ 100mv or so depening on actual clock speeds long as these are above the fully idle state which seem to be around 0, 6 or 30 Mhz for how Wattman reports it at least and 0 - 4% GPU load or so with well everything in this screen just differs a bit or is offset a bit so kinda hard to tell for sure ha ha.
Disabling this sleep or zero state would make for testing voltage and clock speed stability same as the black screen shutdown issue that Vega 64 had earlier though this was related to the VRAM P states and idle so locking it to P1 instead resolved it or rather worked around it, GDDR6 here with Navi is a bit more of a unknown and needs third party programs to do anything but do a few changes up from the stock specs of 875 Mhz to 950 Mhz and this fluctuates a bit too when idle though at much lower values than what can be set for the GPU state for when it's under load though a similar lock to the active state would be handy. (Alternatively forcing it by just using a 144 Hz 2560x1440 or higher display, wonder if that could actually work around the issue by adding another behavior though intentional but it sets a higher idle state for at least some things though one of these might be VRAM operating at full speeds instead of some middle ground which isn't entirely ideal either, something else to try maybe ha ha.)
Well just theories and ideas really, getting somewhere though nothing's a solution or fix for what's going on but this confirms at least that the earlier hardware had nothing to do with the issue so here's hoping it's just a pending driver software fix.
(And even if it can be kinda replicated it can still take forever for it to happen or it might not happen at all because well no idea what causes it or why or where and all sorts of questions but it's something though, not much. )
Bit of a lengthy post kinda suck at explanations too but yeah that's the current state for this fun little quirk on my end.
Definitively helped the GPU a bit by putting a better *everything* in the new system and removing some bottlenecks though honestly 2050 - 2100 Mhz or 1900 with a 3 - 4% performance loss but drastically lower power consumption and much lower temps yeah.
Dial it in to 2000 instead of 2100 default or 2050 it kinda varies from one driver to another and then the -50 Mhz or so and either 1900 (1950) or 1950 (2000) clock speed and voltages around 1.000 to 1.050 leaving some headroom as needed.
(Plus testing at stock voltage first to ensure stability and then testing with modified lower voltages and confirming everything is still stable.)
Voltages curving also means the three states kinda work together to get some range of how it curves though idle can still drop lower although a higher P1 state can make it so it keeps around .800 to .900 although it offshoots the set value by as much as 100mv from what I've been testing or at least how Wattman reports it depending on how it curves here.
Full load is a bit better but can still go higher than set particularly GPU moments where the clocks spike a bit higher or where it maintains boost for a longer period though well it has that thing by always being some 50Mhz within the max boost even at 1500 Mhz it'll be around 1450 and only limited by some software thing here for what it's doing instead of being able to hit the completely achievable target value same as what's going on with Vega and the boost state also being some 50 Mhz within the max targeted value even when lowered down to what's completely doable for these cards.
Bit extra there, lower makes the GPU even more efficient but under 1700 Mhz it's starting to near a 10% performance trade-off though compared to the still significant power consumption possible reductions and power draw (Each GPU being a bit different after all.) and the lowered junction temperate for the GPU core that's still pretty low though a noticeable reduction in framerate when testing.
(Though the 30 - 40% fan speed instead of upwards of double and lessened power draw and further reduction in temps though not as high as the initial gains is nice possibly also as something like a low demand profile such as for older titles.)
Jonas I always give up reading your posts. Too long. So if I read the first part correctly, do you have the black screen problem with 19.12.1 drivers? And by black screen, do you mean the monitor disconnects and the system freezes and you have to hard reset? And is it also happening while idling? That's bad. 19.12.1 are working fine for me. This is such a mess.
Mine is currently set at a 2070MHz target clock (reaches 2040MHz max) at 1.18v and memory clocked at 925MHz with 50%+ power limit.
Yeah it's a bit lengthy as it's been what five and a half months now I think.
So 19.7.5 works here without the black screen sporadic problems but it breaks other stuff fixed in later drivers though these all the way up until 19.12.1 have occasional black screen issues.
19.12.2/19.12.3 increases the frequency of these to almost as bad as the 19.8.x drivers which where the previous worst performing ones for this particular issue.
Always when idle and never when actually gaming which also differs from some reports where it's always when gaming never when idle.
Plus the split between black screen and then the display comes back and black screen and having to do a system restart or shutdown. (Sometimes Windows Key, Ctrl, Shift and B to restart the driver can be needed but if that won't work a shutdown and reboot is required.)
Various thoughts and ideas since I've been trying to figure out what's going on, suspecting it's just the drivers and I have some ways to at least narrow down when it occurs but nothing like a actual solution to the fix or even getting closer to understanding how or why or where this is happening.
Though despite the difficulties in actually doing a replication of the issue so far it's all just pointing to a software problem which is good, though not 100% for everyone which makes testing difficult without a easy to replicate condition.
(Even the one I have can be a bit hit & miss for when it happens.)
So all in all, I expect it'll take a while to get this fixed fully.
(Same as the fluctuating clock speeds problem which has also persisted though there is some confirmation that enhanced sync is getting fixed soon at least so that's something since these cards came out much as it's taking a bit of time.)
Interesting that it only happens when idling and that it is still happening with the new drivers. I thought it was only happening with games now. It's even worse than I thought. The worst drivers for me were the 19.9.x drivers where it would black screen while idling within a few minutes of booting Windows. There is no one driver that works correctly for all. As I said 19.12.1 is working fine for me but as we can see not for you. And the worst part, it was declared 'fixed' within 2 drivers releases by AMD after acknowledging the issue after 4 months since NAVI release...
Jonas what monitor are you using?
rflair, have you also experienced the black screen freeze issue since you got the card?
Earlier 2016 display of one of Dell's the UP2716D fairly OK for a low to mid range 60 Hz 2560x1440 8-bit + FRC (~10-bit but not true.) and lack of features like G/Free-Sync or HDR though the plan is to upgrade to a 120 Hz later in 2020 ideally with whatever is next for AMD or NVIDIA for GPU's though first also seeing how this little driver situation plays out particularly now that the Adrenaline yearly feature driver is out and bug fixes and updates can be a focus in full for the Windows driver development team over on AMD plus whatever NVIDIA will be doing next for the 440 to 445 series on their end.
Still liking the GPU hardware but I am not entirely sure as to how things are going with the software for AMD's GPU division but that's what I'll be checking for the January and February drivers and how much of these bugs they can verify, replicate and fix.
EDIT: I do also understand that it simply takes time and these issues are difficult to track down and verify since it's so random and some users get hit by them bad enough they exchange the cards and others well either it's limited or nothing at all and there's no common factor at all from what I can tell to really tie it together into something that's easy enough to confirm and make a test case for or how to say.
I do not get Black Screen freezes. But I get full freeze when I use Radeon Overlay in media player and take screenshots via it.
As that happens, image stays frozen. But I can unpause, pause media (only audio goes on as image is frozen) for a while before full system freeze happens. But even then image is still frozen.
Impressive to see the results of what Navi can do when pushed a bit higher, hitting anywhere close to 2300 boost is a really good achievement and even a smaller 5Mhz for memory above the usual cap. Very nice.
As a proud Navi owner -> please update Your profile