I know. However, the first time the machine did the GPU reset on me was when I _was_ doing something GL-intensive.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Mac Pro (Late 2013) GPU (Driver) Issues
- Thread starter bax2003
- Start date
- Sort by reaction score
I know. However, the first time the machine did the GPU reset on me was when I _was_ doing something GL-intensive.
GPU-B alone was replaced only on my first machine. If you believe me, I don´t remember, i´d have to look through this thread to refresh my memory. There were 2 or 3 GPU replacements on that machine.
Why do you ask
?Adding my issues to the thread. The story so far:
- working as new Mac engineer for advertising agency and we have about 40-ish of MacPro's with D300. Our production environment is Yosemite due to mostly plugins and expensive software holding us back.
- not long after started working here noticed 4 machines which freeze and leave kernel gpuRestart restart in the logs.
- as one person was leaving decided to rebuilt their machine with image I believe it works fine. Before the end of the day I received the call from the person, GPU restart. Give them my own machine which is working fine so connected them with thunderbolt cable and copy straight disk-to-disk. This user has no more trouble. Of note here is that same disk image had issue with one machine and none with another.
- Took Mac to Apple for the 1st time. It came back with GPU A replaced, later on I found out that they could not find an issue and replaced the GPU anyway.
- Put the machine onto (another) users desk. This user also has GPU restart issues with their machine but it's not covered by Apple Service. Copied their disk image straight disk-to-disk. User came back not long after, GPU restart. I would have preferred to wipe the machine and start from the scratch but user asked to keep their settings.
- Asked Apple to take another look this time stressed the issue and asked them to take a look at this thread to which the answer was "we do not work off threads". This time they had a little more thorough look at it. However I received no report although I have tried verbally in the shop, over the email and two separate senior support engineers at Apple. This is no way the treat the customer and it will become important issue in the moment. The guy who phoned in told me to install new os 10.10 and run updates, that's it.
- Remember how I gave my own machine to an user. Well, that image was moved from Mac Pro - 2013 to Mac Pro - Mid 2012. So now I copied that image from Mac Pro 2012 to Mac Pro 2013 that came back from Apple. Boom - crash in Chrome within first hour.
- Booted into external disk, wiped image, reinstalled 10.10.5, run all updates, installed Flash, Chrome & VLC. Went to Chrome, literally the first Flash movie crashed again.
- Called Apple, engineers very rude, do not wish to listen, being defensive, do not answer emails even they gave me their contact details. So naturally but rather unhelpfully I get defensive too. Anyway, arranged 3rd visit to Apple.
- In the shop just now the guy is telling me looking me straight into the face: "It's not hardware issue, it's software and it's VLC that's causing it, Flash & Chrome is fine". Recall back how I chased for report. So I took the machine and came back into the office with it and decided to write this. Of note here is that I was chasing for a report from the second repair. I have never got this. Also, all my users with bad machines reported better experience after I removed Flash & Chrome which does inconvenience them, being an advertising agency. I have not removed VLC from their machines but they hardly ever use it anyway.
Great post - thank you. Never mind my previous question since it was addressed in the first bullet point. Do the affected machines use the same monitor as the unaffected machines? Thanks!
Last edited:
we are using UTC 4 and until now there was no need to upgrade it as it was working fine with Adobe CC 2015. The upgrade is planned for the summer. It's the setup that I found installed here when I joined in November so it's proving a little difficult for the new guy to recommend an expensive upgrade
Thanks for your reply. I edited my previous question after re-reading your excellent post. Sorry for the confusion and thanks for clarifying why your are still using 10.10. I would love to find out if you use the same monitors on all D300 MacPros. Thanks again!
The monitors are all over the place. The machine I sent to Apple had 2x Apple Cinema. Other machines are on various models of Eizo with secondary monitor being some old crap. Tried all variations where I could and replaced cables, including changing adaptor to DVI-to-TB and even VGA-to-TB.
What makes difference for my users is removing Flash, Chrome & upgrading to Sierra. Under this circumstances users (2 of them) report much less OS crashes, from 1-2 times a day on average to 1-2 a week, even less. However graphic intensive software such as Photoshop is crashing much more often. On other machines I only see a couple of Adobe crash reports but on these there is many many more, as well as many many more QuickLook and WindowServer logs (and others) all ending with .resource.diag. They all have this in common:
Event: wakeups
Wakeups: 45222 wakeups over the last 204 seconds (221 wakeups per second average)...
I am not sure if I can read more into this but to my mind "the freeze" is not the only thing that should be focused on. Other applications crash a hell lot more on these two Macs to the point that these users are calling me regularly to recover lost work.
What makes difference for my users is removing Flash, Chrome & upgrading to Sierra. Under this circumstances users (2 of them) report much less OS crashes, from 1-2 times a day on average to 1-2 a week, even less. However graphic intensive software such as Photoshop is crashing much more often. On other machines I only see a couple of Adobe crash reports but on these there is many many more, as well as many many more QuickLook and WindowServer logs (and others) all ending with .resource.diag. They all have this in common:
Event: wakeups
Wakeups: 45222 wakeups over the last 204 seconds (221 wakeups per second average)...
I am not sure if I can read more into this but to my mind "the freeze" is not the only thing that should be focused on. Other applications crash a hell lot more on these two Macs to the point that these users are calling me regularly to recover lost work.
Follow up:
- Booted into internet recovery and used disk utility to format the disk, not just wipe the partition, full disk format.
- Installed Yosemite and all available updates, added it to Casper, then installed Chrome.
- GPU crashed within minutes after launching Chrome for the 1st time. However this time machine recovered. It logged me out but I did not need to reboot. Won't call that a graceful recovery but none the less in previous cases where I had been alerted to the frozen machine by user, I had to reboot.
That Apple guy is
Ok, so Mac Pro can´t deal with Flash and Chrome, but my 5-6 years old MacBook Air can ?
Those type of answers (3rd party is culprit) are just one word: it starts with B and ends with T
BTW, I bought yesterday a Mac Pro 4.1 (sig) and I just can´t believe it...it seems to copes very well with Chrome and Flash
Last edited:
Does anybody know what the highest safe temperature for the MP2013 GPU is?
Does anybody of you monitor the MP2013 temperatures with the TG PRO app?
Does anybody of you monitor the MP2013 temperatures with the TG PRO app?
I never had Chrome nor Flash installed on my D300 MacPro and experienced the same issues in Yosemite (or higher). That being said I also believe that this is a software issue given that I have no crashes in OS X 10.9.5. If Universal Type Client 4 and Adobe CC 2015 are compatible with Mavericks then downgrading to OS X 10.9.5 is a potential solution. If your machines were manufatured after OS X 10.10 was released you won't be able to use internet recovery, however, Apple should be able to install a clean version of OS X 10.9.5 on one of your drives which you can mirror on the other machines.
The issue does not seem to be temperature-dependent. I ran Furmark in Win10 on bootcamp for about an hour pushing the GPU to about 90 degrees and all was fine. The machine fan was silent when these crashes occurred. It's either some hardware going stale or a driver issue, or both - but doesn't seem to be thermal. Or it is something burning out once and then giving problems continuously.
Made last attempt to fix the issue before escalation.
For 2 years I was experiencing the same freezes as everyone (with confirmation in log files). Apple tried to fix it by replacing GPU, with no success.
3 weeks ago I tried to explain that the issue is still presented, therefore they decided to ... replace GPU.
But this time they replaced both GPUs simultaneously and explained that previously they were only replacing a single card.
So far, it works good, with no issues for 3 weeks.
Sierra 10.12.4
For 2 years I was experiencing the same freezes as everyone (with confirmation in log files). Apple tried to fix it by replacing GPU, with no success.
3 weeks ago I tried to explain that the issue is still presented, therefore they decided to ... replace GPU.
But this time they replaced both GPUs simultaneously and explained that previously they were only replacing a single card.
So far, it works good, with no issues for 3 weeks.
Sierra 10.12.4
I have many of these machines but only 4 are not working. All are built from the same disk image. If I take a disk image from the bad machine and put it on good machine, it works fine. Same is the other way around. Here is a course of events:
- a new user had a dud - this machine has been rebuilt using a known working disk image via Casper prior to them joining the company.
- to start off I took a disk image from my own MacPro D300
- took my Mac to their desk and used TB cable to "restore" the dud to my machine
- user is now happy - same software, on one Mac it's not working on another it is (ye fine, there is a caveat but...)
- I took my own disk image and moved it to Mac Pro Mid 2012 - works fine even though the models are different - but I expected that anyway, see caveat in the bullet point above
- took the machine to Apple
- when it came back I used TB cable to "restore" another affected user's disk to the dud - the problem persisted
- then "restored" my own disk from working Mac Pro 2012 to the dud - the problem persisted
- rebuilt machine using Casper once again - problem persisted
- rebuilt machine to fresh Yosemite - problem persisted
The reason to mention Flash and Chrome is that it causes the machine to crash more often. The freeze will happen even when they are both removed. Whenever user reports a crash to me they almost always mention that at the time they were using Chrome. But their Photoshop crashes more than the Photoshop on the machine next to them. It is possible that it's anecdotal but it does not look like it.
@ Idolum I would love to move to Mavericks but to do 40 + machines just because 4 are not working? I do not think that's very reasonable. I would love to move to Sierra and that is happening sooner rather than later.
Weather it's D300 or RAM or VRAM or kekst or something else is besides the point for us, the users but to me the crash report looks like kekst has saved a memory dump from software-on-chip. It starts with ** GPU ASIC Log Start **. I believe ASIC stands for Application-specific integrated circuit. So you tell me if it's hardware or software. It does not help the fact there is 2 GPU's and the logs do not help to show which one is causing the crash.
What's not besides the point is that we all are speculating, we do not know and we stand no chance to figure it out. I have no formed opinion but am trying to be as thorough as possible given the circumstances. My point is Apple stands far better chances of figuring this themselves. They can for example change memory. I wish I have spares but I do not. They can for example take many of those dud D300's and put it in non-dud, to see if it becomes a dud.
Like this, I have to test, spend a lot of time and prove machine my company spent money on is not working as expected. The reason I am upset with Apple is that they are unwilling to develop a test themselves and they have resources and time for it.
Last edited:
That is the key ! 4 years in plenty enough for finding the culprit, even if you change chip by chip - that is if you even want to find it.
If this has happened to popular iMac or MacBook Pro linup, things would be different.
Last edited:
ye well, not the 1st show-stopping thing Apple is not fixing when they should because it's clearly their design. Does this sounds familiar?
https://www.jamf.com/jamf-nation/discussions/9387/problem-renaming-or-moving-folders-on-smb-share
Nothing we can do but speculate, weave webs of misinformation, fear and loathing.
https://www.jamf.com/jamf-nation/discussions/9387/problem-renaming-or-moving-folders-on-smb-share
Nothing we can do but speculate, weave webs of misinformation, fear and loathing.
Right, not sure why you need an update on Sierra. The fact is that there is a difference in hardware, one disk image works on one Mac, not working on another, I am 100% certain of that. Whilst Apple themselves might not have made any revisions (we do not know), the suppliers of parts have. So I will grant you, it could be software that is communicating to a hardware part that has slightly different spec on one machine than on other. GPU crash is only a symptom but the disease could be on any part. We do not know. Neither does Apple. If Apple does now and it's "fixed" with Sierra would it not be fixed with Yosemite by the way of an update? Security update 2017-001 has not fixed Yosemite, I tested that.
In my opinion Sierra is a smoke screen, the problem is mitigated but not resolved. I have upgraded 2 of my machines to Sierra. The full OS freeze is reduced dramatically but it does happen. Other apps crash far more on these machines than on known good machines as testified by comparing Console logs. It's certainly usable but I would not call it a working machine.
If there's a need for Metal API, 10.12.4 is needed... otherwise don't bother. When the API is ready (maybe 10.13?) then most likely Metal becomes mandatory... but before that... there's no rush.
My observation in our office, crash occurs far more frequently at the beginning of the session, as in when the user comes back from lunch. Usually a 2-3 minutes into a session but this is completely anecdotal. My personal experience, left the machine running over the holiday weekend, 4 days. Came back, machine still running, no crash reports. Started using Chrome, crash within minutes.
To me it sounds as if the machine crash more often when it's cooler.
I share your observations. I rarely had problems with machine running warm while video transcoding for example. This could possible be related to GPUs power management and power delivery when there are speed and voltage regulations (Idle / 2D Accel / 3D Accel).