Posts by FalconFly
|
41)
Message boards :
Number crunching :
Lunatics Windows Installer v0.42 Release Notes
(Message 1555181)
Posted 11 Aug 2014 by  FalconFly
Post: Forgot to ask : Is there a table that lists known optimum/good cmdline_options for certain GPUs on the optimized AP/MB applications ? So far I set -sbs 256 on all my ATI hardware, but I have no clue what specific settings to use for my new NVidia cards (GTX 750TI & GTX 780)... Not wanting to screw anything up, I left that untouched so far. I'm normally an AMD guy and just aquired this CUDA 5.0 hardware today ;) PS. Installation of v0.42 over existing v0.41 installs worked like a charm, only had to reset <count> to run multiple tasks/GPU. |
|
42)
Message boards :
Number crunching :
Lunatics Windows Installer v0.42 Release Notes
(Message 1555056)
Posted 11 Aug 2014 by FalconFly
Post: As I've read from the 0.41 release, it seems you guys can't handle tons of download traffic. If it's okay for you, I've uploaded them onto my site as well. Directlinks : http://www.falconfly.de/downloads/Lunatics_Win32_v0.42_setup.exe http://www.falconfly.de/downloads/Lunatics_Win64_v0.42_setup.exe My server can handle quite alot of traffic, so if you like an additional mirror feel free to spread the download links, files will remain there for download. |
|
43)
Message boards :
Number crunching :
Problems with running 1 wu per AMD Radeon gpu , app_config.xml
(Message 1554030)
Posted 9 Aug 2014 by FalconFly
Post: I still wonder if the GPU tasks are getting enough priority when the CPU is also loaded with CPU tasks. Plus, the default numbers for typical GPUs are like <avg_cpus>0.04 and <max_cpus>0.2 That always seemed a little low, I prefer a fixed 0.5 avg_cpus/max_cpus for 2 tasks/GPU, respectively 1.0 avg_cpus/max_cpus for a GPU running a single task. In all cases, reserving one CPU core for each GPU running seemed to yield far better performance. Must say though, compared to the older BOINC versions (where anything below 1.0 would cause GPU tasks to just hang idle due to almost never getting CPU time), things have improved alot with the current BOINC version. |
|
44)
Message boards :
Number crunching :
Setting up A-10 6700 with stock seti
(Message 1551841)
Posted 3 Aug 2014 by FalconFly
Post: As far as I know and based on my own experiences with that APU, the architecture simply isn't a high performance one. There's little you can do (outside existing optimized Clients), and as soon as the GPU swamps the limited RAM bandwidth you'll see the typical impact of a shared memory GPU on the rest of the system. I was very disappointed of the performance and sold the APU shortly thereafter. I think the best performance can be extracted from the APU by either exclusively running its GPU or at very maximum run one single CPU task besides it. Having 3 CPU tasks (or god forbid even 4) compete for the limited Cache AND the limited RAM bandwidth with a GPU task running will reduce overall output quite a bit and there's nothing you can do about that fundamental issue. PS. I've heard AMD plans to slap high-performance RAM onto its future APUs to yield better performance and migitate the shared memory impact. But that won't come before 2015 the earliest. The way it is now, they repeated the same shared memory mistakes from the days of the earliest iGPUs on the motherboards. An APU is a budget compromise solution really that by defaults accepts that bottleneck, but they do make for cheap (and somewhat limited game-able) desktop solutions. |
|
45)
Message boards :
Number crunching :
Waiting for response from Seti
(Message 1551815)
Posted 3 Aug 2014 by FalconFly
Post: That's a bit weird indeed. For network testing, I'd do the following : - check the Hosts file, which should be clean unless used by special software (if it's infested with data, time to check for malware otherwise) - any new network adapter or -Driver?, maybe a default maxMTU of 1500 on it with an ADSL line? (at least on pre-Win7 systems this causes packet fragmentation an re-routing delays while browsing etc. ; not sure how other OS like Linux/Mac or other internet uplinks handle that issue) - for kicks, testing another DNS server also helps, if there's a rare issue with DNS propagation (i.e. and OpenDNS server or use the NSA *cough* sorry use Google with 8.8.8.8 for a test) - otherwise, a deep antivirus scan (using an other than already installed tool, ideally boot from a dedicated antivirus scan CD) is always a good idea - some browsers seem to get slower over time, despite not really being extended in features while in use (I like to reinstall FireFox once a while and just import the favorites again plus install the plugins needed) - from a command prompt, do a traceroute to websites you know are always up and stable - and to the seti servers of course (if there's a "clog" or weak link in your routing, you'll see it there; it's rare nowadays, but possible that the ideal routing goes over a node that suffers packet loss for some reason) Sometimes it can happen when ISPs and broadband IXPs can't get along and fail to negotiate conditions they can both be happy with. Cutting them off or limiting routing bandwidth is amongst their arsenal to increase their negotiation posision. In the worst case, you're affected by the recent net neutrality violations of some US ISPs and your provider is possibly throttling your traffic to prioritize more lucrative traffic... Might be worth checking out, as you wouldn't be the first one. |
|
46)
Message boards :
Number crunching :
Another question : CPU & GPU vs. total CPU usage
(Message 1551655)
Posted 2 Aug 2014 by FalconFly
Post: Hm, okay, so I think I'll continue to go with use xxx% of CPU cores, as it's easy to switch on the fly... |
|
47)
Message boards :
Number crunching :
Another question : CPU & GPU vs. total CPU usage
(Message 1551542)
Posted 2 Aug 2014 by FalconFly
Post: Hm, I've been pondering that question but couldn't come to a definitive answer. Most know that by default, the BOINC Client will load up all CPU cores AND the GPU. Of course, most users optimizing their system free up CPU cores to maintain optimum loadout of the entire system. But which of the two methods I'm aware of is best? - On multiprocessors, use at most xxx% processors (general computing preferences or computing preferences of individual BOINC Client in local preferences) OR - modify <avgcpus> and <maxcpus> to higher number as to reserve CPU cores As I'm seeing the CPU load from all tasks distributed across all CPU cores, I wondered if using the first option might actually leave unused performance (with both CPU and GPU tasks i.e. just hopping around a now limited number of cores by the Windows management and effectively just generally lower CPU load). Thinking about it, it would seem more reasonable to use the <avgcpus> and <maxcpus> value. But in end, I'm not sure if both methods might actually yield identical results or one being a better performance option ? |
|
48)
Message boards :
Number crunching :
Question regarding ATI GPU optimized clients
(Message 1550663)
Posted 31 Jul 2014 by FalconFly
Post: I recently ran a cleanup on the app_info.xml while implementing the latest MB optimized Versions for both CPU and GPU. First of all, I found the Lunatics 0.41 installer that I had previously installed had all <platform> entries double, one for 32bit and one for 64bit. - Running a 64bit machine, I removed all 32bit platform app_versions, cutting app_info.xml size nearly in half But then I also ran across multiple instances of the same client <plan_class>, such as ...ati_opencl_sah ...opencl_ati_sah ...ati_opencl_100 ...ati13ati ...opencl_ati5_sah ...opencl_ati5_cat132 ...opencl_ati_cat132 I can only assume this will cause the client to operate in ways as to create a binary compatible result with a wingman running other/older hardware ? Now the 1000$ question is : Should I re-enter all those <plan_class> variations or stick only to one that should run fastest on my Tahiti GPU ? Which one would that be anyway ? PS. Since that app_info.xml is still rather hughe, I left only ati_opencl_sah and opencl_ati_sah in it - system is crunching along just nice. (sorry to all wingmen for unintentionally abandoning workunits as their x86 platform got lost over my app_info.xml cleanup :p ) I must say, things have become much more complicated since the early days of optimized SETI clients. Seems best to empty the cache before modifying the app_info.xml, that rings some old memories from the past. |
|
49)
Message boards :
Number crunching :
Puzzled, once more
(Message 1550627)
Posted 31 Jul 2014 by FalconFly
Post: On my setup (HD7970), the GPU and VRAM clocks weren't reduced - but the GPU utilization dropped some ~12% plus some VRAM was being freed (while 2 GPU tasks running). Hard to tell how much performace got lost that way but I can imagine it must have been a chunk. |
|
50)
Message boards :
Number crunching :
Puzzled, once more
(Message 1550554)
Posted 31 Jul 2014 by FalconFly
Post: Well I'll be damned.... HAL9000 is absolutely right. I just ran GPU-Z logging during the period of having the monitor shutdown (HD7970 on Win7). The result : GPU load consistently dropped from 99% to 87% despite still running 2 GPU tasks. PS. So far I thought the only source of significant loss of crunch performance came from forgetting that BOINC by default also installs and sets its screensaver, sucking performance while running (if one forgot about that detail after installing it). |
|
51)
Message boards :
Number crunching :
Puzzled, once more
(Message 1550177)
Posted 30 Jul 2014 by FalconFly
Post: Weird, I would be surprised to see any influence on the GPU clocks of a loaded GPU (i.e. by BOINC). So far I never noted any difference even in non-monitor equipped multi-GPU setups under 24/7 load with BOINC. |
|
52)
Message boards :
Number crunching :
Some questions about BOINC for Android...
(Message 1549836)
Posted 30 Jul 2014 by FalconFly
Post: I can only add my experience with my own Galaxy S4 : Running 2 cores permanently with SETI for me is only possible in winter basically, right now I'm forced to limit it to 1 core due overheating issues otherwise. Since the casing is basically plastic, placing it onto a cooler metal don't do anything for me. Also keep in mind its display is an AMOLED which lifespan will suffer if permanently exposed to high temperatures. Running 3 or even all 4 cores on the S4 would be downright impossible for me, short of submerging it into cold water inside a thin waterproof trashbag in a fridge or something... These things just aren't built for permanent full CPU load. |
|
53)
Message boards :
Number crunching :
Radeon 7990 opinions?
(Message 1549628)
Posted 30 Jul 2014 by FalconFly
Post: I would think there's no reason they shouldn't. Based on my experience with Einstein concerning PCIe bandwidth usage and not exactly knowing how bandwidth-limited SETI is exactly... I'd just make sure to use it on a PCIe 3.0 platform and reserve a good part of the avilable CPU power just to feed it to the max. (i.e. check the GPU loads with GPU-Z or a similar tool and experiment with the BOINC maximum CPU usage to see how many running CPU cores still allow for maximum GPU loading). Based on my experience with PCIe 2.0 and AMD platforms, I found out at least at Einstein, the maximum PCIe bandwidth of the CPU/Chipset was already limiting a single HD7970 GPU somewhat (albeit running 2 or 3 tasks on the GPU in parallel). So seeing the HD7990 has to feed 2 potent GPUs over a single PCIe slot, I'd definitely prefer a PCIe 3.0 platform to make sure you don't run into an unexpected bottleneck. Of course, you'll have to own a power supply capable of safely handling the beast and a potent cooling setup to keep it from overheating. Plus, I do hope the energy cost in your region is a low one. That setup will sure draw alot of juice from the plug. |
|
54)
Message boards :
Number crunching :
Astropulse 601 - when to give up?
(Message 1549608)
Posted 29 Jul 2014 by FalconFly
Post: My old 1.5GHz Pentium M notebook does an AP in about 20 hours with Windows 7 x86. Using ap6_win_x86_sse2_cpu_r2137 optimized app. That's not too shabby. I'm expecting to bring mine back to life within the next 48hours and see what it gets me. |
|
55)
Message boards :
Number crunching :
SETI@Home Wow!-Event 2014
(Message 1549500)
Posted 29 Jul 2014 by FalconFly
Post: And I might add : The challenge got excellent stats - it's already fun to watch them and the actual challenge hasn't even started :D |
|
56)
Message boards :
Number crunching :
Astropulse 601 - when to give up?
(Message 1549444)
Posted 29 Jul 2014 by FalconFly
Post: Have you guys used optimized clients ? They're almost mandatory in order to get decent performance, especially from older CPUs. I just ran an Astropulse on an old notebook (Pentium M 1.86GHz) and it took it some whopping 61 hours to complete, using the stock linux client. I'll dump some old Vista32 onto it and see what it can do with the latest optmized clients. |
|
57)
Message boards :
Number crunching :
SETI@Home Wow!-Event 2014
(Message 1548928)
Posted 28 Jul 2014 by FalconFly
Post: I think I missed this last year! Yep, that's all there is to it, no switching or fiddling with the account. It instantly starts counting you in and after the next full hour passed, you'll see yourself added. |
|
58)
Message boards :
Number crunching :
Some questions about a laptop
(Message 1548657)
Posted 27 Jul 2014 by FalconFly
Post: I'm running a Notebook with Core i3-3110M (DualCore w/ HT), an NVidia GT 720M and the integrated HD4000 iGPU since a few days. I've set it up to use both GPUs and one logical CPU core for crunching, which seems to yield pretty much optimum overall performance. Since it got very hot with both GPUs exceeding 80deg C, I placed it onto 4 spacers that give it about an inch of free space beneath it - and have a pretty good USB Fan blow right beneath the bottom. Not ideal but for this weekend of testing/tweaking a suitable makeshift solution. Basically and especially with GPUs being used for crunching as well, getting a good active laptop cooler of some type is absolutely mandatory. The notebook cooling systems really aren't made for 24/7 100% stress. |
|
59)
Message boards :
Number crunching :
Need app_info.xml for optimized intel GPU clients
(Message 1548565)
Posted 27 Jul 2014 by FalconFly
Post: Cool, thanks alot :) (Nu kann ich der iGPU endlich wieder Feuer unter'm Hintern machen :D ) -- edit -- Well crap, they all got aborted, so that didn't work :( I didn't have all additional files in the Directory, so I hope that was the error. -- edit2 -- Now up & running :D |
|
60)
Message boards :
Number crunching :
Need app_info.xml for optimized intel GPU clients
(Message 1548542)
Posted 27 Jul 2014 by FalconFly
Post: Hi there, after the Lunatics 0.41 installer did give me the optimized clients for both CPU and NVidia GPU, it unfortunately also disabled a intel HD4000 GPU present in that system. I found optimized clients for the intel GPU here : http://www.arkayn.us/forum/index.php?action=tpmod;dl=cat105 However, I found out these do not come with any info how to incorporate them into my existing app_info.xml :( I don't want to fool around without exactly knowing how to correctly implement them, as I know this trashes lots of workunits - especially since they come with several dependency files. Last time I edited app_info.xml files was years ago and comparably easy since GPU clients with dependencies did not exist yet. If anyone has the lines I need to add (both V7 MultiBeam and V6 astropulse), that would be appreciated. Should there be better optmized Clients for the intel GPU, feel free to point me to those and their respective app_info.xml lines. |
Previous 20 · Next 20
©2025 University of California
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.