Not so much from nvidia directly but many of the articles talk about the 25% value.

This one is the most recent one I could find
https://devtalk.nvidia.com/default/topic/992502/why-is-cl_device_max_mem_alloc_size-never-larger-than-25-of-cl_device_global_mem_size-only-on-nvidia-/

Well, I did some googling and came up with this:

why is CL_DEVICE_MAX_MEM_ALLOC_SIZE never larger than 25% of CL_DEVICE_GLOBAL_MEM_SIZE only on NVIDIA

and GeForce GTX 1070, 2048/8192 MB allocatable

So it seems to be true. I never would have known. But it explains why my stderr.txt output never showed more than 2148 MB allocated to the SBS buffer even though I had specified -SBS 3072 as buffer size.

I see that we were googling at the same time. Interesting to find out that there isn't that restriction on Intel or AMD graphics in the OpenCL 1.2 spec.

So help me understand, so then the use of 1024 in both places is correct?

Depends on the operating system you use, I see. Mac OS X uses 1000, Windows uses 1024. Personally for memory calculations I always use 1024, thinking of doing so because memory uses bits, and there's only 2 outcomes when using bits. So everything for memory related cases is a power of 2.

From https://en.wikipedia.org/wiki/Kilobyte#1024_bytes: They are also used for random-access memory capacities, such as main memory and CPU cache sizes, due to the binary addressing of memory.

---

Thanks Ageless.

@Keith, Yeah I would never have thought that either but it does explain somethings, but makes me scratch my head on others.

Checking the BOINC source code, I see they use CL_DEVICE_GLOBAL_MEM_SIZE which returns Size of global device memory in bytes. So perhaps that the application builder should use that as well?

Edit: Although newer versions of BOINC do report the memory on Nvidia devices wrong when they have more than 4GB of memory, but that's because the value is read as an int (32bit), not as a long long (64bit)

---

I knew that there was something wrong with the BOINC code because it can only show 4096 MB of graphics cards memory even though there is more on my current cards. I never thought that was the reason that the -SBS 3072 never worked.

I haven't checked yet on the phone in the bedroom which is slowly working away on two tasks, but is it just the web server that's down, or everything?
I did send out an email to Eric at 8:22am but haven't heard back yet.

---

Yes, it's currently so early, but at the time I sent my email it was just 23:22 in California. I doubt Eric goes to bed at 10pm.

---

... is it just the web server that's down, or everything?

Just the web server - work distribution (uploading, downloading, reporting, fetching) is working normally, as is validation and credit updating. Nothing to worry about.

I also doubt that Eric is online dealing with Seti matters at home at 11.22pm at night either!!! In case you didn't know he happens to have an attractive wife that he prefers to spend quality time with after a days work.

Nothing to worry about.

Indeed, but people are people ....

Yes of course the same Eric who went in to fix a problem on Thanksgiving.

Still as Angela often posts early AM PT she may have already let Eric know all is not well.

I also doubt that Eric is online dealing with Seti matters at home at 11.22pm at night either!!! In case you didn't know he happens to have an attractive wife that he prefers to spend quality time with after a days work.

Nothing to worry about.

Indeed, but people are people ....

But Angela may have been trying to post around 9-10 PM when I noticed it was down.

<ed>Maybe Eric wanted Angela to come to bed?

---

...or by me either in this house :-))

Continue checking the meters. As has been found here in The Netherlands, there are meters on the market that measure wrong. Despite me having two 'smart' meters, I continue to check them and keep their meter readings in a spreadsheet.

Some 750,000 ‘smart’ electricity meters installed in Dutch homes over the past few years may not work properly, according to researchers at Twente University and Amsterdam’s hbo college.

They found that in some cases, the electricity usage measured by the devices in laboratory tests was wrong by almost 600%.
‘The inaccurate readings are due to the energy meter’s design, together with the increasing use of modern (often energy-efficient) switching devices,’ Twente University said in a statement.

‘The energy meters we tested meet all the legal requirements and are certified. These requirements, however, have not made sufficient allowance for modern switching devices,’ Twente professor Frank Leferink said.

---

Well, when we're back up - I'm going to look at my reported GPU tasks. See if they're doing what my Einstein ones did. Then I might have a chat about which incantation would be best.

Waiting will give me more time to drip the cloudburst off first.

I said hello down here ---v and here ------------------------------> here
in case anyone
was wondering
:)

---

The forums are back, and have been for at least 35 minutes at the time of me posting this.

---

Well, when we're back up - I'm going to look at my reported GPU tasks. See if they're doing what my Einstein ones did. Then I might have a chat about which incantation would be best.

Waiting will give me more time to drip the cloudburst off first.

I said hello down here ---v and here ------------------------------> here
in case anyone
was wondering
:)

What was the problem with your task Anne?? Hello Anne!

Hello Zalster :)

Computation errors :/ which is a bit perplexing. Seti GPU tasks are completing and validating though. Einstein on the other hand is coming out with this:

read_checkpoint(): Couldn't open file 'LATeah0030L_1012.0_0_0.0_1273825_0_0.out.cpt': No such file or directory (2)

I'm assuming that's not what I should be seeing? (link https://einsteinathome.org/task/651453465)

I'll try find the time over the weekend to relocate the dust to my face. I've not opened this case before, but I have found a manual which explains the lack of screws quite well. Hopefully that's all that's needed :)

Couldn't open file '....cpt': No such file or directory

is OK if it happens right at the beginning of a task: it's Einstein's way of working out if you're starting a new task from scratch, or picking up in the middle of something you were doing yesterday. Either is possible, and both work fine.

Good to know :)

*start looking for another tree to bark up*

It bangs on about it a lot here at the end :\

Error in computing index of fft input array, i:959971328 pair:3988605
ERROR: prepare_ts_2_phase_diff_sorted() returned with error 2288728
20:28:09 (5760): [CRITICAL]: ERROR: MAIN() returned with error '1'
FPU status flags: PRECISION

to the point of shouting :)

Oh, while I'm here - thank you for corroroborating corroborating the blueness of those minions, Chris :) was any throttling involved... briefly...? ;)

Annie - we had thunder and lightning and a mini rain storm, not much else.

Oh.

anniet wrote:

*start looking for another tree to bark up*

@anniet there was a similar problem-gtx1070-upgrade-570 and whilst the OP never published a solution it looks like a driver upgrade / change may have helped (no longer getting errors) - at least worth a try or asking the OP.