Off topic posts from "Any solid reasons for running v6?"

[Rattledagger]

If you haven't even read the install guide, I'm not sure how much more anyone could help you find the info you were looking for.

Frankly I've got better things to do than to read every single word, and trying to follow every single link of the 29-page "windows install guide".

In addition to this being untrue, I still cannot see that you have provided a link to any web page that states that this is true.

V7 DOES work with XP, Vista, 7, and 8.

In case it wasn't clear, it was the FAQ stating "What platforms does the client run on?" and including a list with XP, 2003, 2008 and 7. Not yet including windows 8 was expected, but including the two server-OS and not Vista seemed strange, since most won't include server-OS in their listings.

[JimF]

I run v7 on two PCs with Nvidia cards only, but am being forced against my will to consider v6 again for another PC due to my old nemesis, running mixed AMD/Nvidia cards. In particular, I want to fold on a GTX 560 while using an HD 7770 for other projects.

I can actually get the GTX 560 to fold easily enough with the right combination of GPU core/Cuda indices, but then it reverts back to the wrong card (the HD 7770) upon restarting v7. I have tried deleting the work folder, etc. to get rid of the HD 7770 and Core_16, but they keep coming back. In the past, I have sometimes had better luck (not perfect by any means) getting v6 to work on a particular card, and am about to try it again.

Either that, or else BOINC detects the cards properly.

[art_l_j_PlanetAMD64]

I run v7 on two PCs with Nvidia cards only, but am being forced against my will to consider v6 again for another PC due to my old nemesis, running mixed AMD/Nvidia cards. In particular, I want to fold on a GTX 560 while using an HD 7770 for other projects.

I can actually get the GTX 560 to fold easily enough with the right combination of GPU core/Cuda indices, but then it reverts back to the wrong card (the HD 7770) upon restarting v7. I have tried deleting the work folder, etc. to get rid of the HD 7770 and Core_16, but they keep coming back. In the past, I have sometimes had better luck (not perfect by any means) getting v6 to work on a particular card, and am about to try it again.

Either that, or else BOINC detects the cards properly.

I haven't tried it, so I don't know if it would work, but couldn't you set 'gpu' to 'false' (disabling auto-detect), and then manually add the GPU slot(s) you want? You should use FAHControl to make the changes, of course, but here is a sample 'config.xml' that might fix this problem:

Code: Select all

<config>
  <!-- User Information -->
  <user value="art_l_j_PlanetAMD64"/> <!-- Enter your user name here -->
  <team value="45862"/>         <!-- Your team number -->
  <passkey value="********************************"/>       <!-- 32 hexadecimal characters if provided -->

  <smp value="true"/>       <!-- If true, attempt to autoconfigure SMP -->
  <gpu value="false"/>      <!-- If true, attempt to autoconfigure GPUs -->

  <!-- Work Unit Control -->
  <next-unit-percentage value="100"/>

  <!-- Folding Slots -->
  <slot id="0" type="SMP"/>
  <slot id="1" type="GPU"/>

</config>

You would have to manually set the 'gpu-index', 'opencl-index', and 'cuda-index' to the required values, but you say that you already know what these values should be. I believe that this would turn off the 'autoconfigure GPUs' function which is causing your problem, but would still allow you to manually create the GPU slot(s) that you want. I have a WinXP system with a GPU WU that will be finished in about 20 minutes, so I will try it there and let you know if it works.

[JimF]

art,

Thanks a lot; it is just what I need.
But v6 picked up the Fermi card right away, and ignored the AMD card, so I think I will run with it until the next install.

[art_l_j_PlanetAMD64]

art,

Thanks a lot; it is just what I need.
But v6 picked up the Fermi card right away, and ignored the AMD card, so I think I will run with it until the next install.

You're welcome! I have just completed what I said I was going to do, which was to:

• set 'gpu' to 'false' (disabling auto-detect)
• deleted the GPU slot (id 0)
• saved the Configuration
• shutdown the computer, including removing the power plug for 20 seconds
• started up the computer
• waited until FAHControl was up and running
• manually added the GPU slot
• saw that the GPU slot is working OK

When I manually added the GPU slot, I left the 'gpu-index', 'opencl-index', and 'cuda-index' at their default value of -1, but you say that you already know what these values should be, to fold on the GTX 560 while using the HD 7770 for other projects.

[JimF]

You would think that v7 would allow you to start up from the last known good configuration, or something like that. But they want to figure it out for themselves. It is another case where trying to make things easier for you actually makes them much harder.

[art_l_j_PlanetAMD64]

You would think that v7 would allow you to start up from the last known good configuration, or something like that. But they want to figure it out for themselves. It is another case where trying to make things easier for you actually makes them much harder.

Well, they're also trying to make it super-simple for the first-time FAH user as well, a job which I think they've done admirably.

The Custom Installation - Optional (Advanced and above) seems to do most of what you are suggesting, with this exception:

Note: The installer (GPU & SMP) will automatically install one or more FAHClient slots to match the computer hardware detected. If multiple CPU cores are detected, an SMP slot is installed, otherwise a Uniprocessor slot is installed as the fall back option. If one or more GPUs are detected, one or more GPU slots are also installed automatically. If neither multicores or GPU are detected, a Uniprocessor slot is installed by default.

This is the point where the following options could be made available to the Advanced or Expert user:

1. Continue with the auto-detection and installation of slots ('smp' and 'gpu' set to 'true'), or
2. Import the existing configuration ('smp' and 'gpu' set to 'false'), or
3. The user will manually add slots after FAHControl starts ('smp' and 'gpu' set to 'false')

The only option that might be more work for the developers is the second one (Import the existing configuration), the first and third options already exist. And even the second option shouldn't be too much work, as the code to read the configuration file (config.xml) already exists, assuming a previous v7 installation (there is no config.xml file in older versions of FAH). Option 2 could even give you the choice of whether or not you wanted auto-detection and installation of slots to be enabled ('smp' and 'gpu' set to 'true') or disabled ('smp' and 'gpu' set to 'false') when the FAH v7 installation has been completed.

[TonyStewart14]

I agree- I sometimes have to dump uniprocessor WUs because they are downloaded automatically if I don't have the slots pre-configured and I want SMP instead.

[mmonnin]

Multiple GPU setups.
viewtopic.php?p=199379#p199379

[codysluder]

It is another case where trying to make things easier for you actually makes them much harder.

As you know, the V7 client is still being developed and it's not perfect yet, as you point out. Consider the following facts.

The vast majority of donor has one or zero GPUs in their system. Automatic configuration is certainly easier for them since it works flawlessly. Of the small fraction of Donors who have two or more supported GPUs, a lot will have two from the same company, whether it's a single board with two GPUs or simply two GPU cards from their favorite vendor. Again, automatic configuration is flawless.

Installing GPUs from two different companies was impossible in V6. Sometimes it works in V7 and sometimes it doesn't. We can hope that the team that's working on V7 fixes the bugs in auto-configuration that you've encountered soon so it will be easier for those rather obscure cases where it currently doesn't work.

[Rattledagger]

The vast majority of donor has one or zero GPUs in their system. Automatic configuration is certainly easier for them since it works flawlessly.

I wouldn't really call the automatic configuration "flawlessly" for Amd-GPU-users, since SMP and GPU will compete for the same cpu-resources and can give severe performance-hits for SMP.

[bruce]

There are two separate auto-config settings: gpu=true/false and cpu=true/false. I'm talking specifically about GPU configuration, not SMP configuration. you're correct in stating that using both autoconfig settings concurrently has problems with AMD GPUs, but crashing a WU due to enumeration issues is more serious than a performance issue where both both WUs will be completed successfully but earn fewer points because of performance issues.

WUs which are A) fast and correct are better than WUs which are B) slow and correct but slow and correct is still better than C) never (machine hung) and/or (D) slow and incorrect. If I were to assign priorities, I'd eliminate C & D before worrying about eliminating B.

As I've said elsewhere, the PG considers WUs one-at-a-time, not combinations. They do benchmarking by running a single WU, not a combination of WUs. So the AMD+SMP issue never comes up.

I expect that future changes to the V7 client will take more combinations into consideration, but that's still a work-in-progress, not one that's included in the current version.

[Nathan_P]

Installing GPUs from two different companies was impossible in V6.

Err no it wasn't, I had a mix of Asus fermi, EVGA GT200 and Leadtek/Asus G92 Gpu's in my gpu folding rig. They ran just fine together under v6. However I will accept that not everyone could get it to work - it did take me about a week to get them to play nice together

[art_l_j_PlanetAMD64]

Installing GPUs from two different companies was impossible in V6.

Err no it wasn't, I had a mix of Asus fermi, EVGA GT200 and Leadtek/Asus G92 Gpu's in my gpu folding rig. They ran just fine together under v6. However I will accept that not everyone could get it to work - it did take me about a week to get them to play nice together.

I'm not sure, but I think that codysluder meant NVidia versus AMD/ATI, not different vendors of all-NVidia cards.

[JimF]

The vast majority of donor has one or zero GPUs in their system. Automatic configuration is certainly easier for them since it works flawlessly.

Yes, I am not indicting V7 for not doing its job. It does it well in many cases. But you don't have many options when things go wrong, at least not obvious ones. The problem could be addressed fairly easily it would seem by allowing you to configure it yourself and simply save your settings. V7 automates you out of the system a little too readily.