Hi folks

After a long sabaticle from seti I have just dedicated a machine to the cause again :-)

it has 4 280X GPU and runs only GPU
however the CPU is pegged at 90-100%
this is not a worry as the machine is a dedicated headless cruncher

I'm just pondering if the CPU is actually preventing maximum RAC potential or if this is pretty normal

I havent used GPU for boinc before so it has many unknowns for me

Cheers

and well done for keeping the project running for 15 years !
i have only crunched for 14ish

:-)

---

63. (1) (b) "music" includes sounds wholly or predominantly characterised by the emission of a succession of repetitive beats

This is only a 2-core Celeron driving 4 GPUs and 12 tasks:
http://setiathome.berkeley.edu/show_host_detail.php?hostid=7405247

Are you sure the best production of this system will be with 3 tasks/GPU ?

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

yep its only a lowly celeron

I thought the same about the number of instances on each gpu but even on a count of 1 the cpu pegs >90

the big question is , just how much of an impact on the actual crunching does this have ? if its simply adding small delays but the gpus are running an order of magnitude faster, does it really make a lot of real world difference ?

thanks for the input

As i suspected when i checked your webpage. Those that have been running coinmining are soon returning to distributed computing as it's not sustainable anylonger to mine due to powerbill is higher than the relative income. :)

Welcome with your rig at S@H :)

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

The impact is when the workunits requires more cpu to Crunch, if i recall blanking is done by cpu.
And ofcourse the initials of every wu is made by the cpu. So the stronger cpu the more utilisation on the gpu and higher productivity.
Best thing is to monitor with gpu-z how much gpu utilisation you have on average if you are in the 95% range almost all the time then no need to bother.
More in parallell is often better as you have set it than fewer until the time when bus/cpu becomes bottleneck.

---

_________________________________________________________________________
Addicted to SETI crunching!
Founder of GPU Users Group

indeed

i have no heating in my house so in winter i run computers instead.

last year i mined and made money and the heating was paid for :-)

the profitability of GPU mining stopped back in May here in the UK, due to energy prices

it was a good run though made quite alot of money and still have a couple of machines left to heat the house this winter, shame i have to pay for it this time

needless to say the celeron was more than up for the job of script mining

Welcome back Adrian :)

---

great thanks for your input vyper

i will have to wait for the RAC to even out and then do some comparisons

i had no idea that some processing was done on CPU as well, i thought all the CPU did was hand the units out.. not sure of the mechanism of this. any suggestions as to what to read to further understand the process ?

cheers thomas :-)

In your case the motherboard is the bottleneck.
The GPU`s are running fine and the run times are nice for a 280x.

CPU time for work units are pretty normal.

---

With each crime and every kindness we birth our future.

ah ha, the voice of experience Mike, thanks for taking the time to go have a look.

i'm not going to spend on a new cpu so i will have to tune it over the following weeks, its all part of the fun

those 280x are undervolted too to save energy, doesn't appear to make much difference in performance

any idea is GPU Memory clock speed has much impact on overall crunching performance ?

i had no idea that some processing was done on CPU as well, i thought all the CPU did was hand the units out..

Only for AstroPulse v6 the blanking is done on CPU
Not the case for AstroPulse v7 or SETI@home v7

The CPU load is due to the new drivers (> Catalyst 12.1) way of sync

See if:
-cpu_lock -instances_per_device 3

... put in files (similar names, may not be these exact on your system):
ap_cmdline_win_x86_SSE2_OpenCL_ATI.txt
mb_cmdline_win_x86_SSE_OpenCL_ATi_HD5.txt

... will improve (lower) the CPU load


Also try if SetiPerformance will help to check the optimal number of tasks/GPU
http://setiathome.berkeley.edu/forum_thread.php?id=73524&postid=1453245#1453245

---

 


- ALF - "Find out what you don't do well ..... then don't do it!" :)
 

i have already run SetiPerformance, thats why i was running three threads/GPU

thanks for the info about cpu instances, i have done as you said and it looks like a little pressure has been relieved, still bouncing 80-90.

it will be interesting what RAC is possible with this rig, somewhere in the top 40 i hope, although its all nvidia up there

3 instances is perfect for 280x.

No GPU memory speed is not that important for seti.
My last card was highly overclocked but times were almost same.

---

With each crime and every kindness we birth our future.

i have already run SetiPerformance, thats why i was running three threads/GPU

thanks for the info about cpu instances, i have done as you said and it looks like a little pressure has been relieved, still bouncing 80-90.

it will be interesting what RAC is possible with this rig, somewhere in the top 40 i hope, although its all nvidia up there

For MB work the NV is most efficient and for AP ATI is most efficient, but both will run the jobs just fine.
If you wanted to start running AP on your GPUs. Then you may want to wait until we are fully switched over to AP v7. Due to your lower powered CPU. As AP v6 does blanking on the CPU.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Hal9000 i have a question as i am gob smacked at how he can run 4 GPU's on a 2 core celeron so i looked up the card .
My question is this .

Is what AMD call stream processing units simular to CUDA cores , as this card he has says it has 1792 stream processing units , so i'm thinking that would be like saying 1792 CUDA cores ?

Without getting to technical about the differences between the 2 types

---

Hal9000 i have a question as i am gob smacked at how he can run 4 GPU's on a 2 core celeron so i looked up the card .
My question is this .

Is what AMD call stream processing units simular to CUDA cores , as this card he has says it has 1792 stream processing units , so i'm thinking that would be like saying 1792 CUDA cores ?

Without getting to technical about the differences between the 2 types

Before Nvidia stated using the term "CUDA cores" they and everyone else used the term "Shaders". ATI used the term "Stream processors".
If you look at the list for ATI or Nvidia on Wikipedia you will see the "Core config" column. Which lists the Unified Shaders, Texture mapping units, & Render output units for each GPU. If you compare the Titan Z with the Radeon R9 295X2. Then you will se two cards with a similar core config but two very different results.

How the software, firmware/drivers/apps, make use of that hardware is what really separates the two.

---

SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[

Thank's Hal9000 that's what i thought it meant .

---