GPU2 -> GPU3 Transition Guide
Officially Supported Operating Systems:
Windows XP (32 bit & 64 bit)
Windows Vista (32 bit & 64 bit)
Windows 7 (32 bit & 64 bit)
GPU3 BETA Client Download Locations:
Systray 6.32: (
http://www.stanford.edu/~friedrim/.Folding@home-systray-632.msi)
Console 6.31 For Windows XP: (
http://www.stanford.edu/~friedrim/.Folding@home-Win32-GPU_XP-631.zip)
Console 6.31 For Windows Vista/7: (
http://www.stanford.edu/~friedrim/.Folding@home-Win32-GPU_Vista-631.zip)
Before Using GPU3 BETA Client:
Users of GPU2 Client, wanting to check out GPU3 BETA Client, are advised to use the -oneunit flag. After the Client exits, you can then proceed with the installation of GPU3 BETA Client. The reason why it is recommended to upgrade the Client after the WU has been completed, is to avoid any potential problems that the user may experience from incompatibilities between the FAHCore and the WU. Also it won't cause loss of points if any problems occur. Ensure that you are meeting the minimum requirements for running the GPU3 BETA Client. Usage of Nvidia 197.45 Driver is recommended. If you use drivers prior to this, it may or may not work. Note that GPU2 Client will not use the FAHCore_15 to process the WU.
Some users complain that after they installed the new driver, they are having problem(s) in running F@H or other GPU related applications. Most of this can be avoided by making a clean installation of the driver. However, you can use this freeware application Driver Sweeper to remove any "stubborn" file(s) which may be causing problem(s).
NVIDIA GeForce Forceware 2XX.XX WHQL + GPU3 BETA Client: (UNOFFICIAL)
So far, some users can fold without problems while others are facing problems. If you are using any of these Driver Versions, please report your experience in the appropriate thread:
Nvidia GeForce Forceware 257.21 WHQL
Nvidia GeForce Forceware 258.49 WHQL
Nvidia GeForce ForceWare 258.19
Installation Of Systray Version:
The Client is installed in C:\Program Files\Folding@Home and you have to start the application from there. This will cause a screen to appear where you can configure the Client. Restart the Client and if everything is properly configured, it will then download an appropriate WU along with the required FAHCore.
Installation Of Console Version:
The installation requires you to simply unzip the file in the folder of your liking and run "Folding@home-Win32-GPU" which will allow you to configure the Client. Restart the Client (use Ctrl + C OR use the X button) and if everything is properly configured, it will then download an appropriate WU along with the required FAHCore.
Upgrading The Console Version: (Needs more data)
Download the correct GPU3 BETA Client version and after you unzip the file, copy/paste the files into the pre-existing GPU2 Client folder. If asked to overwrite the file(s), answer Yes. The already created shortcuts will work.
If you encounter problems, please report it.
Installation Of Multiple Instances:
Initially install the first Client and make sure that it is running smoothly before setting up multiple instances. More detailed instructions can be found here and here. To find the information about the GPUs that are installed in your system along with the driver information, you can use GPU-Z. For obtaining the X value in -gpu X, you can use CUDA-Z. Note that the value of X starts with 0 and not 1.
Using GTX 400 Series GPU & GTX 200 Series GPU In One System: (Needs more data)
There are reported problems when in a single system; GTX 400 Series GPU and GTX 200 Series GPU, are both being used for folding at the same time. A possible solution is the usage of -forcegpu nvidia_g80 flag with the GTX 200 Series GPU ONLY. This will result in the correct detection of the GPU by the Client.
If you are still having problems with the setup, then you can use these arguments with the correct GPU:
GTX 400 Series GPU (Fermi): -forcegpu nvidia_fermi
GTX 200 Series GPU (Non-Fermi): -forcegpu nvidia_g80
There is a single report where in a single system, FAHCore_15 WU would not fold on a GTX 285 when GTX 480 is folding. GTX 285 folds FAHCore_11 WUs without any problems.
If you have a similar setup, please report your finding in this thread: (Folding on 4xx GPU and 2xx GPU?)
Using Dual GTX 400 Series GPU In One System: (Needs more data)
So far there is only a single instance were using GTX 480 and GTX 470 to fold on the same system will cause the latter to slow down by roughly 27% At this stage, it is unknown if it is a bug or not as there is too little data to come to a conclusion. Some users are suggesting that the CPU may be the bottleneck in this situation. A possible solution is to run one instance of Systray Version and one instance of Console Version.
If you have a similar setup, please report your finding in this thread: (Mixed Fermi results)
Notes: (General)
1) ATI users may be eager to try out the Client but this current release isn't for ATI yet. At this moment, there isn't any advantage and you will be exposing yourself to new bugs: (Link)
2) Distribution of WUs are like this:
GTX 400 Series GPU (Fermi) = GPU3 Client using only FAHCore_15
8000 Series GPU, 9000 Series GPU, GTX 200 Series GPU (Non-Fermi) = GPU2 using only FAHCore_11 OR GPU3 using FAHCore_11 and FAHCore_15
3) GPU3 Projects are: (FAHCore_15)
A) 10626
B) 10627
C) 10628
D) 10629
E) 10630
F) 10631
4) Do note that the FAH Folder, for either the Console Version or the Systray Version, isn't located in the Program Files. If it is, it is advised to change it to a simpler location like C:\F@H due to the UAC implemented in Windows Vista and Windows 7 which may conflict with the Clients and causes unexpected results.
5) If you are using 8800 Series GPU or 9800 series GPU, you may encounter bugs. The Development Team is aware of it and are working for a fix: (Link)
6) Some users complain about desktop lagging when performing basic tasks. An effective method to make the lagging less apparent is to enable Aero if you are using Windows Vista or Windows 7.
7) To make your FAHLog more detailed, you can add this flag -verbosity 9 which will be helpful when troubleshooting your Client.
8) The Client is in BETA Stage so users can expect instability or crashes.
9) The GPU3 Client will be starting with a fresh Project and won't be porting any of the GPU2 Project.
9) The GPU3 is in BETA stage so the Credits haven't been finalized yet: (Link)
Notes: (Only GTX 400 Series GPU (Fermi))
1) Usage of flag -forcegpu nvidia_g80 will cause problems with the Client.
2) If the Client isn't having enough CPU Cycles, the drop in PPD is significant.
3) If you are getting CoreStatus = 63 (99), you may want to:
A) If you were initially tweaking the system for CUDA apps, you may have added this Environment Variable "CUDA_FORCE_PTX_JIT=1" Deleting this will allow the GPU3 Client to work: (Link)
B) Check if the GPU is completely securely fixed inside the PCI-E Slot: (Link)
Observations: (GPU3 BETA Client)
1) Using -advmethods flag in GPU3 Client on a Non-Fermi GPU (8000 Series GPU, 9000 Series GPU, GTX 200 Series GPU) will result in user obtaining FAHCore_11 WUs. In order for the user to obtain FAHCore_15 WUs, remove the -advmethods flag. This may change in the future: (Link)
2) The usage of Nvidia Environment Variable haven't yet been tested with GPU3 Client. (Folders report that there isn't any need as the GPU temperatures are safe enough to begin with)
3) WU result file size is roughly 3MB with FAHCore_15 and <1MB with FAHCore_11. Will need more data to get a conclusive pattern. Meanwhile users having a transfer cap may want to adjust their calculations accordingly.
4) Users are reporting that GPUs using FAHCore_15 operate at a lesser temperature than FAHCore_11. Will need more data to reach a conclusion.
5) Usage of Flag -forcegpu nvidia_g80 isn't really necessary as nearly all latest GPUs from Nvidia and ATI are automatically being recognized.
6) Under some circumstances, you may get an error about cudart.dll so you may want to post your FAHLog here so we can figure out the problem. Coping the file may or may not work.
If you would like to contribute information, please post it and I will add it along with giving the user credit. If I have made any mistakes, please correct me.
P5-133XL = Method For Systray and Multiple Instances.
stevehat1 = Upgrading The Console Version.
grunion = Using Dual GTX 400 Series GPU In One System.
ikarppin = Testing with GTX 400 Series GPU and GTX 200 Series GPU in one system.
Sidicas = Provided point A for CoreStatus = 63 (99).
rbpeake = Provided point B for CoreStatus = 63 (99).
Michael_McCord,_M.D. & ElectricVehicle = Thread for Nvidia 257.21 WHQL
I had another GTX 260 to make a set of 3 for my MSI K9A2 rig. While putting it back together I decided to change over to GPU3. To my surprise I didn't get a FAHCore_15. Running the FAHCore_11 I was getting around 7800 PPD. It wasn't until I read the GPU3 Guide that I discovered my error. You currently can't use the -advmethods flag with the GPU3 Client using the 8000 Series GPU, 9000 Series GPU or GTX 200 Series GPU and receive a FAHCore_15.
