Opened a new thread - as was pointed out - the first one is now long and messy, and since 7970s been out a while now, the initial rush and problems are over (as such), so this thread should be more practical now.

Regards
Zy

...... 12.1 Drivers, what are release day RC drivers ??? 

Thats your problem. 12.1 does not support 7970s. You need the AMD drivers release for 7.XXX. The problem is its now the brave world of OpenCL, and to get the 7970 running OpenCL apps, you need the drivers they released when 7970s were released. Those drivers do not support CAL, as CAL is officially dead. So the new drivers are native OpenCL written for the new GCN architecture of 7970s. Without them, and running 12.1, the card just sits and cycles doing nothing because its looking for OpenCL support and can only see CAL which it doesnt understand - or doesnt seem to. DNETC have a CAL app nearly running on a 7970, but I'm not convinced they will finally do it ... see what happens ..

Guru3D are carrying them, these are the next version after the release day set - its an RC - Release Candidate - set, but they work fine

http://downloads.guru3d.com/download.php?det=2844

Load those up, and you should be fine. I copied my app_info below just in case:

<app_info>
<app>
<name>milkyway</name>
</app>
<file_info>
<name>milkyway_separation_0.82_windows_x86_64__cuda_opencl.exe</name>
<executable />
</file_info>
<app_version>
<app_name>milkyway</app_name>
<version_num>82</version_num>
<flops>1.0e11</flops>
<avg_ncpus>0.05</avg_ncpus>
<max_ncpus>0.5</max_ncpus>
<plan_class>ati14ati</plan_class>
<coproc>
<type>ATI</type>
<count>0.5</count>
</coproc>
<cmdline>--gpu-disable-checkpointing</cmdline>
<file_ref>
<file_name>milkyway_separation_0.82_windows_x86_64__cuda_opencl.exe</file_name>
<main_program/>
</file_ref>
</app_version>
</app_info>

Regards
Zy

The 12.1 Drivers did run the Collatz Wu's as I ran a few of them but then realized they wouldn't get Validated. I tried Slickers app release and the Wu's took 3 times as long to run as they did without the app file.

I'll try the Drivers & app file you suggested although the app file is the same as I was using. I'll get back in awhile & let you know how it worked out ... Thanks

---

STE\/E

Collatz will not run on 12.1 / 7970. What happens is the card takes the app, processes it, but it does not get validated. That card is looking for OpenCL apps, Collatz is a CAL app, and so will not run successfully.

Slicker released an OpenCL version for the 7970, and included in his post a download link which brings down a zip file. Look at the read me file inside the zip file, and it explains it all. He also included an off line test so we could get the parameters sorted out without borking WUs. It works fine, I've not yet got round to testing extensively the parameters he gives, but the OpenCL app does run ok.

There will be more of this as the change from CAL to OpenCL takes hold. Its a fundamental shift, and a new mindset, until it all settles more in circa 18 months. OpenCL is way slower than CAL, but the shift in hardware from 40nm to 28nm will to a degree, mask perceptions and help the transition. However it is looked at and viewed, the shift to OpenCL (by both NVidia and AMD) has started, and the days of propriatory GPU languages has been buried, never to return - at least not in the consumer world.

Now OpenCL is mainstream, its bound to get faster as time goes on, as developments of the Core language march on. It will never be as fast as propriatory languages ( aka Stream/CUDA), but thats not the yardstick used to bring it in. They are measuring all this by cross-platform compatibility. A world away from us for sure - to a degreee - but even in BOINCLand we will see benefits .... we are now running an OpenCL app put togther by an NVidia compiler on AMD hardware. Will take time to mature et al ... gonna be fun :)

Regards
Zy

OpenCL is way slower than CAL,

This doesn't really mean anything. OpenCL is a specification for an API. It's too abstract to call it slow. Speed will come from usage as well as the implementation. The problem is AMD's OpenCL compiler is kind of bad at times. CAL was kind of awful and you had to write kernels in the IL pseudo-assembly; there were things like Brook to compile C-like languages to it, but they disappeared at some point as AMD stopped shipping them and there was some kind of void as far as I can tell for what to actually use. For something heavily ALU bound like the separation stuff, the failings of the compiler become more apparent.

Actually I think most of the problem with the OpenCL stuff for separation is I couldn't effectively replace the exp() approximation used in the IL kernel in it. The standard exp it uses requires dynamic indexing into some kind of math constant table they use which necessitates a switch to a texture clause and is quite slow (the texture clause thing isn't true anymore in GCN and the OpenCL kernel is as fast as the IL kernel although I still think it shouldn't be but I haven't really looked into why yet). I tried once to get as close as I could to replacing it but it didn't work so well and I decided I needed direct access to 2 IL instructions to come close to it. I'll probably try again at some point.

1/29/2012 3:25:04 PM | Milkyway@Home | File referenced in app_info.xml does not exist: milkyway_separation_0.82_windows_x86_64__cuda_opencl.exe

---

STE\/E

Go here...

http://www.arkayn.us/forum/index.php?action=downloads;cat=11

You are looking for the Windows 64 bit cuda OpenCL app (last one on the list of three under windows 64 bit)

Regards
Zy

lol ...

1/29/2012 4:25:08 PM | | ATI GPU 0: ATI unknown (CAL version 1.4.1658, 3072MB, 3033MB available, 9472 GFLOPS peak)
1/29/2012 4:25:08 PM | | OpenCL: ATI GPU 0: Tahiti (driver version CAL 1.4.1658 (VM), device version OpenCL 1.1 AMD-APP (844.5), 6144MB)
1/29/2012 4:25:08 PM | | ATI GPU is OpenCL-capable
1/29/2012 4:25:08 PM | Milkyway@Home | Found app_info.xml; using anonymous platform
1/29/2012 4:28:36 PM | Milkyway@Home | Sending scheduler request: Requested by user.
1/29/2012 4:28:36 PM | Milkyway@Home | Requesting new tasks for ATI
1/29/2012 4:28:37 PM | Milkyway@Home | Scheduler request completed: got 0 new tasks
1/29/2012 4:28:37 PM | Milkyway@Home | No tasks sent
1/29/2012 4:28:37 PM | Milkyway@Home | Message from server: Your app_info.xml file doesn't have a usable version of MilkyWay@Home N-Body Simulation.

---

STE\/E

It doesnt, but you dont need one - ignore it.

It should run ok if you loaded the app_info and the files from the downloaded zip file.

Regards
Zy

Okay but not getting any work right now ... Thanks for all the help Z

EDIT: Okay I got a few W's & they ran 2 @ 102 Seconds (Single Card will add Second Card later) ... Just using the Max ATI ccc 1125 Core setting & no power added & Memory @ 700

---

STE\/E

Spot on - exactly what was expected - your install is good :)

Just a case now of playing with the voltages ... if you are into that .... if you are I'll post a dead safe start set of settings to get you going, and you move on from there to whatever level re voltage you want to go to. Make sure the power slider is set to 20% full on in CCC if you are o/c re voltages. The power slider raises the card TDP, without that you will be restricted on voltage changes.

If voltage o/c use afterburner - get the latest Beta RC from Guru3D, (RC 11 I think), its been tweeked to give full support to 7970s. Works well - could do worse ...

Regards
Zy

Okay PM me or Post the safe starting point. I have MSI Afterburner v2.2.0 BETA 11 & TRIXX installed, Afterburner doesn't show the Voltage, doesn't really read the card right at all, still say's it's at 925 Core ...

EDIT: Another Question, which side should the Switch be on 1 or 2, from reading 2 is the Default side & 1 the Overclock side unless I read it wrong ... ???

---

STE\/E

Quirks of Afterburner et el ... I'll put together a PM on it, dont want to put details of unofficial mode on the forum, else some will go into play mode and burn cards.

I'm reloading my Driver at present on the 7970 machine, I'll get the PM to you in about 30 mins

Regards
Zy

Okay, no problem ... Thanks

---

STE\/E

Extract from Guru3D review on 7970:

"If you peek to the right of the PCB on the photo in-between the 1 and 2 position you'll notice a Switch, it's actually a Dual BIOS toggle switch. Setting two returns the card to factory default, setting 1 is an unprotected mode which allows you to overclock and tweak."

Relevant page of review (look near the bottom)

http://www.guru3d.com/article/amd-radeon-hd-7970-review/6

That review and the two others 7970 related at Guru3D well worth reading, thoroughly reccommended

Regards
Zy

Okay Z, I got your PM but won't mess with it until tomorrow as it's getting late here & I'm trying to figure out some Network & Firewall Problems, have it mostly sorted out but needs a little Tweaking yet.

---

STE\/E

There is a new Version of the ASUS GPU Tweak Utility out - ASUS GPU Tweak V2.0.3, it is very good, and well worth a download. The link below is for the utility on 7970 cards. It may work on other cards despite the title on the main screen, I have no idea.

They have also folded into it GPU-Z. The result is a neat collection of tools, combining some of the Afterburner and GPU-Z functionality. The tool set is not as extensive as Afterburner, but its close .... real nice. Not had a chance to test thoroughly yet, but so far so good.

Anyone using tools on cards, this is certainly worth a peak thats for sure. Select your OS at the drop down box on this link:

http://uk.asus.com/Graphics_Cards/AMD_Series/HD79703GD5/#download

Regards
Zy

This all sounds like far too much faffing about. In the end I decided to refuse my 7970 delivery and send it back. It's not worth spending over £400 on when it is barely faster (for MW purposes) than my 5870.

Its over twice as fast.

Your twin 5870 is producing two WUs every 74 seconds.

A twin 7970 produces four WUs in 88 seconds, and thats running an OpenCL app which puts a (roughly 30%+) slow down on the 7970

On first opening, without any tweaking or non standard actions, and running at stock speed, the 7970 will do its first WU in 52 seconds on its own, no doubling up with app_info

Regards
Zy

A twin 7970 produces four WUs in 88 seconds, and thats running an OpenCL app which puts a (roughly 30%+) slow down on the 7970

Not true

Mine is producing 4 every 88 seconds right now - look at the results on my 7970 machine. Its running app_info with two WUs per card, they both produce at 88 second point = 4 per 88 seconds.

Every 7970 on start up does produce its first WU in 52 seconds - every time without fail at stock speeds.

Last to get them running was Steve yesterday, his first card on an app_infor stock speed produced two in 102 seconds which equates to one rolling off every 51 seconds.

Regards
Zy