All things Navi 10
13 Jul 2019 9:20:19 UTC
Topic 219203
(moderation:
Inspired by two recent posts in the 'All things Raseon VII / Vega 20' thread...
cecht wrote:
I'm looking forward to someone starting an "All Things Navi 10" forum topic. :)
Here it is!
archae86 wrote:
It will be very interesting to see how well the Navi models perform on Einstein relative to the fabulous performance here of the Radeon VII, which was out of all proportion to the relative gaming performance of that model.
As the Navi models are materially different architecturally and use a very different type of memory, I think only actual Einstein trials will tell us much of how well or how poorly they perform here on the current Gamma Ray Pulsar application--let alone on future applications.
I bought a Sapphire boxed RX 5700 XT 3 days ago to replace an ageing and occasionally flaky MSI R9 280 in [url=https://einsteinathome.org/host/9161845]this host[/url]. Initial observations of cruching performance are good, the card is running out of the box settings with the only tweak I have made being to the fan curve via Wattman.
Crunching performance in this unaltered state is easily on par with most of my overclocked Vega 64's, runtimes are in the order of 5m40s for a single task and 9m25s for two concurrent tasks. Peak card temperature (with my fan curve adjustment) is just 70C at 2700 RPM fan speed and for an AMD blower type fan this is not in the least bit noisy, to my ears all I hear is a gentle whoosh. There is also none of the annoying fan ramping that some of us have experienced with Vega and Radeon VII cards :-)
The truly impressive thing is that given AMD's penchant for setting high default voltages this card is currently averaging a board power of just 179 Watts crunching tasks x2!
So to sum up, what we have is overclocked Vega 64 crunching performance from a significantly cooler, quieter and less power hungry card which with further tweaks could be improved upon still...
BUT
After completing 2 or 3 hundred tasks at both x1 and x2 multiplicity NOT ONE SINGLE TASK has yet validated, there have been no straight errors just consistent invalids.
After a little Googling I discovered some review sites that have attempted folding@home to assess OpenCL performance and they too were unable to produce valid results (sorry no links at the moment). Let's hope future driver improvements can sort this out!!
Currently, if you are considering spending your cash on a Navi card I'd highly recommend you wait ;-)
I notice that your invalid
)
I notice that your invalid results show a mix of simple "validate error" and of "completed, marked as invalid". The second type might possibly reflect slight numeric difference, possibly arising out of details of the implementation, but the first is a more basic sanity check.
The relative rate of the two types seems to have changed, possibly in conjunction with changes in the average completion times. Maybe your 1X tasks typically using 339 seconds elapsed times usually got "completed, marked as invalid", for tasks reported around noon UTC July 11 and also a batch of 2X tasks using about 564 seconds reported around 7 UTC July 11.
Then there is an odd little batch of tasks with elapsed times of about 1150 seconds reported around 13 UTC July 10 with "validate error". Also may more for example around noon UTC July 9. Did you by any chance try 4X, or are these some other sort of anomaly? A small proportion of the 1150 second population were judged "completed, marked as invalid", but the great majority got the worst grade.
Thanks for the report. By the way, I saw a plausible report that while all the initial Navi cards on offer are blower-type single fan cards (probably to a reference design?), that cards with more diverse designs should begin to be available mid-August.
I have a box with abundant cooling and power supply capability current running an RX 570. Your report for the 5700 XT looks like it would upgrade that box nicely if only it actually worked on Einstein Gamma-Ray Pulsar tasks. I'll wait, both for fans more to my liking, and for reports of it working. I'm not eager to repeat my experience with the GTX 2080 (which worked on some batches of Einstein GRP work, and not on others).
Apologies Archae86 you have
)
Apologies Archae86 you have caught me out for rushing my original post! When reviewing my hosts task lists please ignore any results returned before 13:30UTC on the 10th July, any invalids or error while computing results before this time were from the old R9 280 :-)
I would also like to correct one of my original statements (I don't appear to be able to edit my original text)...
Currently, if you are considering spending your cash on a Navi card I'd highly recommend you wait
I should more accurately have said:
Currently, if you are considering spending your cash on a Navi card for compute only I'd highly recommend you wait
I have just downloaded and installed a fresh copy of driver 19.7.1 in the hope things may improve!
If there are any GW gpu tasks out there I could try running those if I still come up bust after the reinstall.
Gavin wrote:cecht wrote: I'm
)
You are a saint.
Ideas are not fixed, nor should they be; we live in model-dependent reality.
Just to update, reinstalling
)
Just to update, reinstalling drivers has made no difference and to be honest I didn't think it would but one has to try these things. I have completed many more tasks now at both x1 and x2 multiplicity, the results are always the same - all tasks complete successfully and satisfy the servers sanity checks only to be declared invalid once compared with the wingmans results.
In view of this I'm beginning to wonder if the project admins would take a look at the results and shed some light on the matter, however I am reluctant to ask directly at this point as the RX 5700 and its drivers are still less than week old and from what I have seen the drivers are very rough and ready in many respects!! That said, tasks continue to crunch successfully and pass the sanity check...
In the meantime I have been trying to get some GW work for the card to have a go at crunching but after several hours I still have none although the server status page suggests there are tasks to send. I will keep trying to snag some of these tasks for the next couple of hours but I think after that I will shut the machine down and await driver updates to see if that resolves the problem. - I am very aware of the number of invalid tasks I have now produced and apologise to all those affected :-)
Gav.
A couple of hours now since
)
A couple of hours now since my last post, still no GW work but have run a small number of GRBP tasks at x3 just for the hell of it. Haven't got any valid results but interestingly still no straight errors or signs of trouble from card or driver.
Shutting the machine down for now.
Gavin wrote:Shutting the
)
I'm very sorry the new card gave this much trouble. Thanks very much for trying so hard and for giving everybody a 'heads up' about the problem. Hopefully AMD will identify and sort out what is causing this fairly quickly.
With any luck, we'll see it returning valid results in the very near future!
Cheers,
Gary.
If it's any consolation, the
)
If it's any consolation, the cryptominers aren't having any better luck with the RX 5700.
https://bitcointalk.org/index.php?topic=5149738.80
https://www.reddit.com/r/gpumining/comments/ca8rsg/rx_5700_hashrate/
As far as I can suss, they seem to think it's an OpenCL kernel issue and are waiting for driver and mining software updates to handle the new card architecture. I wonder whether an updated 'ati' app will be needed for E@H?
Ideas are not fixed, nor should they be; we live in model-dependent reality.
cecht wrote: As far as I can
)
Thanks for the links, I'm heartened to know it's not just me suffering! With luck and if AMD pull their finger out new drivers will come quickly and resolve the issue. Given that the tasks I have run at x1, x2 and now x3 appear to have been computationally correct (yet all invalid), I'm not at this point sure a new einstein app will be needed... He say's :-)
I suppose experiencing trouble for the sake of others is the danger of trying to live on the bleeding edge of technology but I guess someone has to do it! I am though quite surprised other hardened einstein crunchers haven't taken the early plunge like me. Given the anticipatation around the web in the run up to Navi's release I'm surprised I'm the only one here. I have also trawled some of the other project forums and can't find any RX 5700 owners that want to talk...
This begs the questions, what do you all know and why was I kept out of the memo?!!!!!!!!!!!!
All Joking aside I have high hopes that future driver updates will correct the issues and there will be no need for an updated einstein app. Time will tell but despite the current failure to produce a valid task confidence is high here and you all will be the first to know when I complete a validated unit and I thank you for suffering with me :-)
Gav.
Gavin wrote:...I suppose
)
Yeah, we're kinda like Mikey's brothers/friends on the old Life cereal commercial. LOL!
Ideas are not fixed, nor should they be; we live in model-dependent reality.
Hi Folks, Bit of a (belated)
)
Hi Folks,
Bit of a (belated) update, driver version 19.7.2 was released this last mid week. I initially did an 'over the top' upgrade of 19.7.1 and then subsequently a clean install of the driver and ran 30 odd single tasks under both conditions. Sadly (for me :-( ) I have to report there has been no improvement from my first report, all tasks run have completed successfully but were deemed invalid when compared to the wingmans result.
I have shut down the machine again and await the next driver update from AMD... I'll be back as that chap said!