I have a NVIDIA GT730 and four AMD FuryX VGA cards.

It's possible (and 'worth'?) to crunch GPUGRID tasks on this VGA cards?

Thanks.


[EDIT: BTW. GT730 with 347.52 driver and FuryX with Crimson 15.12 (1912.5 driver) - OK?]

---

This project is CUDA based, so you can use NVidia cards here.
There is an AMD app, but the work for these cards is scarce.
The GT 730 is shipped in 3 versions:
1. 96 CUDA cores @700MHz, DDR3 - this version is insufficient here
2a. 384 CUDA cores @900MHz, DDR3
2b. the 2a version with GDDR5
The 2a and 2b is theoretically capable to finish a task in the given deadline (5 days), but you should keep a short (0.1 days) queue to avoid the 2nd task in the queue to miss the deadline, also you should keep your card running 24 hours a day. Running other GPU projects on a GT 730 could also cause missed deadlines (which is pretty bad for a task which takes 2-3 days to complete on that card).

I have the ZOTAC GT730 PCIe x1 (ZT-71107-10L) VGA card...

CUDA cores: 384
Video Memory: 1GB DDR3

SETI/AstroPulse, Einstein, Milkyway is running fine on this VGA card.
BTW. It's a secondary VGA card, no screen connected - just for crunching. The screen is to the Intel HD Graphics (iGPU of Intel Celeron J1900 CPU) connected.

So GPUGRID tasks ('short' and 'long') could finish during the 5 days deadline (yes, the PC is running 24/7)?

During this 5 days deadline the Credits/task will be always the same, or within 1 or 2 days upload/report and a bonus will be granted (after reading old messages in the forum) - this will still happen?
These days how big is the bonus and within which days?


It's difficult to get GPUGRID tasks?
My BOINC is asking for new tasks since over 24 hours, but don't get:

Requesting new tasks for NVIDIA GPU
Scheduler request completed: got 0 new tasks
No tasks sent
No tasks are available for Short runs (2-3 hours on fastest card)
No tasks are available for Long runs (8-12 hours on fastest card)
Project has no tasks available

From the first few messages in BOINC:

CUDA: NVIDIA GPU 0: GeForce GT 730 (driver version 347.52, CUDA version 7.0, compute capability 3.5, 1024MB, 933MB available, 692 GFLOPS peak)
OpenCL: NVIDIA GPU 0: GeForce GT 730 (driver version 347.52, device version OpenCL 1.1 CUDA, 1024MB, 933MB available, 692 GFLOPS peak)

It looks like after the apps page, cuda60 and cuda65 apps are available, the VGA card/driver combi can do cuda70.


Or what is the problem, that my BOINC don't get GPUGRID tasks?


Thanks.

---

If Server Status https://www.gpugrid.net/server_status.php ... shows 0 unsent tasks, then that means none are available. Task availability is pretty sparse here, so be sure you are attached to other GPU projects which also keep your GPUs busy doing science!

Because of my last question...

During this 5 days deadline the Credits/task will be always the same, or within 1 or 2 days upload/report and a bonus will be granted (after reading old messages in the forum) - this will still happen?
These days how big is the bonus and within which days?

So during this 5 days deadline the granted Credits will be always the same?
No bonus because of quicker upload/report - within few hours or days?


My GT730 finished this WU in 3 days + 20 hours + 50 minutes:
e35s18_e33s4p0f5-GERARD_FXCXCL12RX_1153966_2-0-1-RND8448
http://www.gpugrid.net/workunit.php?wuid=11693272
Long runs (8-12 hours on fastest card)

That's OK, or it should be faster?


It looks like the GPUGRID WU (1WU/GPU) made the GPU ~ 5 °C warmer - that could be?
GPUGRID WUs use the VGA card 'harder' than e.g. SETI (2WUs/GPU), Milkyway (2WUs/GPU), Einstein (1WU/GPU)...?


What happens if the result will be uploaded/reported after the deadline - and if the wingmen didn't uploaded/reported in the meantime, the result get Credits granted nevertheless?


Thanks.

BTW.

On the server status page: http://www.gpugrid.net/server_status.php
- is shown now:

Detailed computing status
Application 	unsent 	in progress 	success 	error rate
Short runs (2-3 hours on fastest card)
-- 
Long runs (8-12 hours on fastest card)
ADRIA_2OV5_CONF_ASP		153	740	287	56.71%
GERARD_FXCXCL12RX_1153966_	45	38	428	22.04%
GERARD_FXCXCL12RX_1167531_	1	40	448	22.89%
GERARD_FXCXCL12RX_1497613_	21	59	408	26.09%
GERARD_FXCXCL12RX_1573605_	0	68	415	22.86%
GERARD_FXCXCL12RX_1576145_	24	57	383	28.94%
GERARD_FXCXCL12RX_1579678_	0	56	418	27.18%
GERARD_FXCXCL12RX_2183201_	21	46	420	29.65%
GERARD_FXCXCL12RX_224327_	1	64	408	31.89%
GERARD_FXCXCL12RX_2353730_	0	61	370	28.98%
GERARD_FXCXCL12RX_629461_	52	34	382	26.96%
GERARD_FXCXCL12RX_941195_	22	51	400	29.58%
GERARD_MO_MOR_			137	226	1237	32.99%
GERARD_MO_TRV_			126	216	1211	29.84%
GIANNI_D3C36bCHL		59	433	380	48.23%

What does mean the 'error rate'?
Does this mean 22 - 56 % of the WUs finish with an error and no Credits will be granted for the calculation time?


Thanks.

---

Because of my last question...

During this 5 days deadline the Credits/task will be always the same, or within 1 or 2 days upload/report and a bonus will be granted (after reading old messages in the forum) - this will still happen?
These days how big is the bonus and within which days?

So during this 5 days deadline the granted Credits will be always the same? No bonus because of quicker upload/report - within few hours or days?

These bonus rules are still applied, so +50% for returning the result within 24 hours, and +25% within 48 hours.
The granted credit is the same for every WU from the same batch, there are only two levels of bonus.

My GT730 finished this WU in 3 days + 20 hours + 50 minutes:
e35s18_e33s4p0f5-GERARD_FXCXCL12RX_1153966_2-0-1-RND8448
http://www.gpugrid.net/workunit.php?wuid=11693272
Long runs (8-12 hours on fastest card)

That's OK, or it should be faster?

It's OK for an OS with the WDDM overhead. You could try to set the SWAN_SYNC environmental value (it would make the GPUGrid client to use a full CPU thread/core), but it would have a little effect on a GT730. Also, you can try to suspend CPU crunching to make GPUGrid crunch faster.

It looks like the GPUGRID WU (1WU/GPU) made the GPU ~ 5 °C warmer - that could be?
GPUGRID WUs use the VGA card 'harder' than e.g. SETI (2WUs/GPU), Milkyway (2WUs/GPU), Einstein (1WU/GPU)...?

Exactly that's the reason.

What happens if the result will be uploaded/reported after the deadline - and if the wingmen didn't uploaded/reported in the meantime, the result get Credits granted nevertheless?

Yes, it's a bug in how the scheduler works. Both parties will receive the same credit, so the wingman will receive the credits without bonus if the first one returns the result while the wingman is still processing it.

BTW.

On the server status page: http://www.gpugrid.net/server_status.php

What does mean the 'error rate'?
Does this mean 22 - 56 % of the WUs finish with an error and no Credits will be granted for the calculation time?

Results marked as invalid, and workunits failing during the processing won't receive any credit.
The error rate of the different batches are more likely the error rate of the hosts processing them, not of the workunits themselves. The error usually happens right at the start of the processing, thus there are only failed workunits in the stats until the first valid results are returned (8-10 hours). So that's why the error rate starts high (above 90%) for every batch, then normalizes to ~22%. That ~22% of the hosts are using unsupported GPU, or have bad drivers, or their GPU is overclocked too much. The only known error in the GPUGrid app is if there's a dirty shutdown and the host has a fast GPU then the files needed to restart the calculation get corrupted, and either the WU will crash at restart, or restarts from 0% and the result will be invalid.