Welcome to MilkyWay@home

Upcoming Cayman architecture.


Advanced search

Message boards : Number crunching : Upcoming Cayman architecture.
Message board moderation

To post messages, you must log in.

Previous · 1 · 2 · 3 · 4 · 5 · Next

AuthorMessage
Profilekashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
100 million credit badge14 year member badge
Message 45225 - Posted: 23 Dec 2010, 1:18:40 UTC

Interesting, I didn't think of the power capping of Cayman affecting the FurMark power figures. I know that my 5870 was drawing slightly more power on the core VRMs than figures I had seen of FurMark being run. That made me believe that FurMark was a fair indication of power draw when running MilkyWay or DNETC on AMD cards. Less power used on the memory though. Collatz draws considerably less than the other 2 projects on the core VRMs, but more on the memory. The single precision projects do not benefit from Cayman's 24% increase in double precision performance, so there's a possibility that Cypress would be more efficient on those. The extra memory is definitely a disadvantage for crunching use.

Thanks for the tip about experimenting with reducing the GPU voltage, I'll give that a go after Christmas. My power bill is approaching the max I can afford, so every time the tariff rises I will have to turn something down or off. Removing the 5870 and the floor fan should keep me going for a while, I'll see how that works when the next bill arrives. In winter I can partly justify it because it warms my room beautifully, but in summer it just spins the meter and results in a sauna room.

Yes I also believe a 1 GB 6950 would be a good Cayman choice for crunching, but thought the next smaller series would be the best for me if I'm still going then. I wouldn't mind switching to a single core GPU actually, the dual core ones can pump out a lot of work for the power used but cause more problems with cooling and sometimes drivers too.
ID: 45225 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
55degrees

Send message
Joined: 8 Sep 09
Posts: 62
Credit: 61,330,584
RAC: 0
50 million credit badge12 year member badge
Message 45226 - Posted: 23 Dec 2010, 2:04:30 UTC - in response to Message 45221.  

ID: 45226 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
55degrees

Send message
Joined: 8 Sep 09
Posts: 62
Credit: 61,330,584
RAC: 0
50 million credit badge12 year member badge
Message 45227 - Posted: 23 Dec 2010, 2:06:46 UTC - in response to Message 45212.  

Yes an additional delaying factor is possibly that they would wish to make 6990 OpenCL capable on both cores. If only newer drivers are available for 6990 and the OpenCL dual core bug is not fixed then 6990 may not be able to be used on MilkyWay.

I had a little try of PrimeGrid's OpenCL application. I mistakenly thought that the OpenCL bug on 5970 had been fixed. Didn't realise for some time that all the tasks completed and successfully reported on the second core were invalid after being checked by a wingman. When I came back to MilkyWay the same thing happened and I wasn't aware of it until I noticed a lower than usual daily total and a mass of pending tasks. Didn't know the OpenCL driver would muck up CAL applications as well. Will have to revert to an earlier driver when I get a chance. At least DNETC is working, still freezes occasionally of course on a 5970, but I doubt that will ever be fixed.



I was not aware of noted opencl driver issues with cal apps. wtf? f = flipper, for the forum police.

by freezes, do you mean the infuriating post gpu crunched WU then cpu hanging, that happens to just a lucky few? oh yeah, with shivers, I spent so much, WAY too much, time trying to figure out that mess; gave up. the wrapper folk have no idea and not getting help from distributed.net, so I agree it very likely never gets fixed.
ID: 45227 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilekashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
100 million credit badge14 year member badge
Message 45228 - Posted: 23 Dec 2010, 3:09:25 UTC - in response to Message 45227.  


I was not aware of noted opencl driver issues with cal apps. wtf? f = flipper, for the forum police.

Yes I couldn't understand how or why the OpenCL bug affects CAL applications either. It certainly does though, have a look at my computer and you will see a few invalids dribbling through from tasks I completed a day or so ago. Validated and invalidated tasks clear from the database quickly here but there should still be one or more invalids there most times for the next few days. Notice all invalids are on GPU 1, which is the second 5970 core. The bug only affects the 5970, recent single core GPUs are supported by OpenCL.


by freezes, do you mean the infuriating post gpu crunched WU then cpu hanging, that happens to just a lucky few? oh yeah, with shivers, I spent so much, WAY too much, time trying to figure out that mess; gave up. the wrapper folk have no idea and not getting help from distributed.net, so I agree it very likely never gets fixed.

That's the one, the DNETC task gets stuck, usually at or near the end of processing and sits there not progressing until the watchdog timer kicks in. When this happens the DNETC CPU usage rises to a full core's worth and stays there while the task is stuck. At least DNETC works most of the time, if the 6990 suffers from the OpenCL bug like the 5970, DNETC may be the only BOINC project that is able to make use of both 6990 cores because there will be no earlier working CAL drivers available for it.
ID: 45228 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Zydor
Avatar

Send message
Joined: 24 Feb 09
Posts: 620
Credit: 100,587,625
RAC: 0
100 million credit badge13 year member badgeextraordinary contributions badge
Message 45229 - Posted: 23 Dec 2010, 3:15:11 UTC

Whilst I believe the fuss will die down once the 6990 is out - and NVIDIA will have achieved their objective of bringing forward the 580 as an attempted 6990 spoiler, its only a short term event, the AMD machine will roll on once the 6990 is out, NVIDIA have nothing to respond with.

The real battle is the upcoming new family aka AMD APUs etc. In all this, Intel is the dark horse that could well come up the inside rail whilst traditional paper battles are refought AMD v NVIDIA. Intel tried before to pitch into the graphics market and failed.

I cant see Intel making the same mistake again. I personally am awaiting informed comment - as opposed to ill informed or spoiler rumour - on what their offering will be. The latter will not be known in a definitive sense for a good while yet. I hope they make it, the graphics market needs strong competition, and for 2/3 years NVIDIA has not supplied it. Without that effective competition - and I could care less if its NVIDIA or Intel or both - we will be in for years of relatively high pricing for consumer related graphics technologies howsoever they manifest themselves. There will be no AMD incentive to keep pricing down

The upcoming world of the APU etc is the real battle/question - the 69xx/580 saga is just talking points in the long run, as it will all be over on release of the 6990 Jan / Feb, which all expect will be way ahead of the 580.

Regards
Zy
ID: 45229 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile[SETI.USA]Tank_Master

Send message
Joined: 2 Sep 07
Posts: 24
Credit: 26,244,596
RAC: 0
20 million credit badge14 year member badge
Message 45231 - Posted: 23 Dec 2010, 4:04:01 UTC
Last modified: 23 Dec 2010, 4:21:04 UTC

just to comment, i am starting to return results with my ATI 6970 (i do have 10.12 drivers with the openCL SDK installed from AMD)

they are taking 2min per WU on my system.
ID: 45231 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilekashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
100 million credit badge14 year member badge
Message 45232 - Posted: 23 Dec 2010, 5:34:52 UTC

Do you find running at 800 MHz helps much with the temperature? I would consider lowering the memory speed first before lowering the default core speed but everyone has their own way of doing things and that's fine. I understand some prefer to keep the memory speed high if they have Collatz as a backup project.
ID: 45232 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile[SETI.USA]Tank_Master

Send message
Joined: 2 Sep 07
Posts: 24
Credit: 26,244,596
RAC: 0
20 million credit badge14 year member badge
Message 45233 - Posted: 23 Dec 2010, 7:00:14 UTC

im running stock speeds, and stock app, but with fan forced to 50%. my temps on WM are 74C for the card after running for several hours now.

and correction to the post above, the seperation_17 takes 3 min and the seperation_16 takes 2 min to complete.
ID: 45233 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilekashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
100 million credit badge14 year member badge
Message 45234 - Posted: 23 Dec 2010, 7:19:59 UTC
Last modified: 23 Dec 2010, 7:24:48 UTC

That's strange, I thought default speed was 880 MHz. Perhaps it's the power limiting circuitry lowering it to 800 Mhz.
Ah I know, the driver is lowering the speed to 800 MHz after the completion of the task and that is what is being reported. I had something like that once a while a go when I used to run one task at a time.
ID: 45234 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile[SETI.USA]Tank_Master

Send message
Joined: 2 Sep 07
Posts: 24
Credit: 26,244,596
RAC: 0
20 million credit badge14 year member badge
Message 45242 - Posted: 23 Dec 2010, 17:47:23 UTC

humm, i seem to have come acorss an issue now, the server is only sending me ATI work, even though i have it set to both ATI and nvidia, so my geforce 295 is sitting idle... i upgraded boinc to 6.12.8 and its still doing the same.
ID: 45242 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileThe Gas Giant
Avatar

Send message
Joined: 24 Dec 07
Posts: 1947
Credit: 240,884,648
RAC: 0
200 million credit badge14 year member badge
Message 45244 - Posted: 23 Dec 2010, 19:18:55 UTC - in response to Message 45242.  

humm, i seem to have come acorss an issue now, the server is only sending me ATI work, even though i have it set to both ATI and nvidia, so my geforce 295 is sitting idle... i upgraded boinc to 6.12.8 and its still doing the same.

The system is smart enough to only send work to the superior card... ;P
ID: 45244 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile[SETI.USA]Tank_Master

Send message
Joined: 2 Sep 07
Posts: 24
Credit: 26,244,596
RAC: 0
20 million credit badge14 year member badge
Message 45258 - Posted: 24 Dec 2010, 3:22:01 UTC

i turned off RRI and now it seems to be working fine (getting work for both cards as its asking for more than one WU at a time, but asking less often)
ID: 45258 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
55degrees

Send message
Joined: 8 Sep 09
Posts: 62
Credit: 61,330,584
RAC: 0
50 million credit badge12 year member badge
Message 45271 - Posted: 24 Dec 2010, 15:57:14 UTC - in response to Message 45229.  

Whilst I believe the fuss will die down once the 6990 is out - and NVIDIA will have achieved their objective of bringing forward the 580 as an attempted 6990 spoiler, its only a short term event, the AMD machine will roll on once the 6990 is out, NVIDIA have nothing to respond with.


I agree, and with et apes, and kashi, that essentially amd is doing what they want and thinking long term. the 6970 might not decimate, but hold on. that the 6970 should not be compared to the 580 and wait for the 6990, etc. but I'm not sold the 6990 in-wait gets them anymore than boasting. the price for it if anything like the 5970 is out there, smaller audience.

I am not challenging what amd bring to the table, but how they serve it.

has anyone oc'd the 6970? what voltages possible, etc.?
ID: 45271 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Mogdy

Send message
Joined: 26 Dec 09
Posts: 9
Credit: 20,140,488
RAC: 0
20 million credit badge12 year member badge
Message 45296 - Posted: 25 Dec 2010, 18:57:10 UTC - in response to Message 45233.  

im running stock speeds, and stock app, but with fan forced to 50%. my temps on WM are 74C for the card after running for several hours now.

and correction to the post above, the seperation_17 takes 3 min and the seperation_16 takes 2 min to complete.


i think you have a problem, look at my results.
and for voltage, now you can't go over 1.175v but with trixx you can go below
i'll try.
in crossfire, the max i can go is 900-1400 after there is lock up.
ID: 45296 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileArif Mert Kapicioglu

Send message
Joined: 14 Dec 09
Posts: 161
Credit: 589,318,064
RAC: 0
500 million credit badge12 year member badge
Message 45297 - Posted: 25 Dec 2010, 20:07:13 UTC
Last modified: 25 Dec 2010, 20:07:49 UTC

Ok people, i got my first 6970 finally:). At +%20 power and 880 core clock, here is my host, you can check the results. I know the times are already posted by other people. You can check also mine. Hope will help.
Take care

http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=248968&offset=0&show_names=0&state=2
ID: 45297 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileArif Mert Kapicioglu

Send message
Joined: 14 Dec 09
Posts: 161
Credit: 589,318,064
RAC: 0
500 million credit badge12 year member badge
Message 45300 - Posted: 25 Dec 2010, 22:21:06 UTC - in response to Message 45297.  

Ok people, i got my first 6970 finally:). At +%20 power and 880 core clock, here is my host, you can check the results. I know the times are already posted by other people. You can check also mine. Hope will help.
Take care

http://milkyway.cs.rpi.edu/milkyway/results.php?hostid=248968&offset=0&show_names=0&state=2


For 880 core clock, i learned that power tune should be at 0 (am i right?), however i increased the power tune to +%20 and core clock to 950. If anybody understands the power tune option, please advice me for what to do at stock core clock and 950 core to save heat.

The max temp is 80C at 880 core, 85C at 950 core. (The value that writes on GPU Temp. #2 on GPU-Z)
ID: 45300 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profile[SETI.USA]Tank_Master

Send message
Joined: 2 Sep 07
Posts: 24
Credit: 26,244,596
RAC: 0
20 million credit badge14 year member badge
Message 45303 - Posted: 26 Dec 2010, 7:55:54 UTC

i uped the power to +20% and the times reduced by about 1/3rd... from 2 and 3 min to 1:12 and 1:50 for each type of WU. I will overclock it later and see what it can do. you can see info on the powerTune here: http://www.anandtech.com/show/4061/amds-radeon-hd-6970-radeon-hd-6950/7
ID: 45303 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Profilekashi

Send message
Joined: 30 Dec 07
Posts: 311
Credit: 149,490,184
RAC: 0
100 million credit badge14 year member badge
Message 45304 - Posted: 26 Dec 2010, 9:51:25 UTC
Last modified: 26 Dec 2010, 9:53:48 UTC

Glad you have it running faster. Looks like it was throttling after all. I knew something was happening because your times were slower than they should have been.

Thanks for the link to the explanation of PowerTune. I had only read a small amount of one review. There's so many game benchmarks in reviews now that my eyes glaze over after a few pages so I rarely read them anymore. An interesting concept and very useful for reducing power consumption during review benchmarks. It certainly fooled me when I was speculating about the relative power efficiency by using figures from a review.

It is surprising that AnandTech tests show Distributed.net client does not trigger it. On both my 5870 and 5970, DNETC pulls more power and runs slightly hotter than MilkyWay. Must be the load on the double precision Cayman parts causing PowerTune throttling to be activated, while the single precision parts use less power on Cayman plus there are 4% less shaders on Cayman XT compared to Cypress XT and Hemlock XT cores.
ID: 45304 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
ProfileArif Mert Kapicioglu

Send message
Joined: 14 Dec 09
Posts: 161
Credit: 589,318,064
RAC: 0
500 million credit badge12 year member badge
Message 45321 - Posted: 26 Dec 2010, 19:39:58 UTC - in response to Message 45303.  

i uped the power to +20% and the times reduced by about 1/3rd... from 2 and 3 min to 1:12 and 1:50 for each type of WU. I will overclock it later and see what it can do. you can see info on the powerTune here: http://www.anandtech.com/show/4061/amds-radeon-hd-6970-radeon-hd-6950/7


Are you running crossfire? I couldn't see the host in your data. I just lowered the core to 850 and memory to 685 via MSI Afterburner Beta 5 and still the temps are really high on the second one, easily exceeding 95 degree. I will check the temps tomorrow at a constant cooled temperature of 25. Hope they wont exceed 80:(
ID: 45321 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Zydor
Avatar

Send message
Joined: 24 Feb 09
Posts: 620
Credit: 100,587,625
RAC: 0
100 million credit badge13 year member badgeextraordinary contributions badge
Message 45322 - Posted: 26 Dec 2010, 19:51:27 UTC

Keep an eye on VRM temperatures, they often get missed/forgotten about in the concern over Core temperature. Its the VRMs that cause 5970s incorrectly handled to burn out, as there is a very wide difference between core temps and VRM temperatures on a 5970. The 5970s have an inherent design issue that causes this, but does bare watching for in a 6970 (and upcoming 6990) until it becomes clear what the situation with the new design is.

I have no idea what the VRM temp way points are for a 6970, but for a 5970 they are:

Up to 100 degrees - ideal, usually core temps are around 75-80
Up to 110 degrees - tolerable, will be fine, watch for increases, core temps usually around 85-90
Up to 115 degrees - pushing your luck, need to back off, core temps usually 90+
115-120 degree - card life will drastically reduce, must cool further
120+ - card devolts, will burn out fairly quickly if run 7x24 at these temps

GPU-Z is a good one to use to monitor VRMs as well as core temps

Regards
Zy
ID: 45322 · Rating: 0 · rate: Rate + / Rate - Report as offensive     Reply Quote
Previous · 1 · 2 · 3 · 4 · 5 · Next

Message boards : Number crunching : Upcoming Cayman architecture.

©2022 Astroinformatics Group