Posts by archae86

1) Message boards : Projects : News on Project Outages (Message 89371)
Posted 27 Dec 2018 by archae86
Post:
Now the top level at einstein.org responds with an outage page, saying the website is down for maintenance and that they hope it will be back up soon. I'll take hope that at least someone knows there was a problem.
2) Message boards : Projects : News on Project Outages (Message 89370)
Posted 27 Dec 2018 by archae86
Post:
Gary Charpentier wrote:
Einstein@home... the website is MIA. Any news?

This outage looks different from the usual to me.
The server status page at https://einsteinathome.org/server_status.html responds normally.
a ping to einsteinathome.org gets a normal response.
but using http://einsteinathome.org in the browser address bar gets:
Proxy Error
The proxy server could not handle the request GET /.
Reason: Error during SSL Handshake with remote server

Apache/2.4.10 (Debian) Server at einsteinathome.org Port 443

So I'll speculate there is an Apache or Debian configuration error, not an actual down server.
3) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82186)
Posted 23 Oct 2017 by archae86
Post:
No one touched this yet, but what make and model PSU do you use?
And how old is it?
Nexus Value 430 purchased August 2010.

As of this morning I've given up on working this problem. I removed the dual 750Ti cards and put a single GTX 1050 in the primary slot, and resumed production work.

In summary, it appears that the combination of Windows 10 Fall Creators Update, the most recent Nvidia driver, and my system configuration somehow lead to a situation in which some, but not all, recognition of the graphics cards installed does not work properly for at least three distinct combinations of cards I've tried. Clinfo and BOINC both think there is only one card installed, and in all three cases where there were really two cards, it was the card in the secondary slot which BOINC and clinfo reported to be present (and which BOINC actually used). When there is only a single card installed in the primary slot, that is recognized by clinfo and BOINC and used successfully by BOINC.

I reviewed the BOINCstats top Einstein host list, and found more than one top 100 host which runs Windows 10, has installed the Fall Creators Update, runs Nvidia, and runs more than one card. I don't know their install date, so can't rely on their credit history to confirm or deny that they share my pain.
4) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82141)
Posted 21 Oct 2017 by archae86
Post:
Ageless wrote:
No one touched this yet, but what make and model PSU do you use?
And how old is it?
I've turned the machine back over to the primary user (my wife) after hours of labor this morning. I think I'll wait to try to read the power supply model until the next time I open it. Generally I use Seasonic units, running far below their rating. That box, however, quite likely has something else in it, and when the 970+750Ti configuration was still running (less than a week ago) was consuming a higher fraction of nameplate than my custom. If I can read the model number, I can probably find my purchase record. I'll hazard a guess at five years.

However the 970 has not run a single second of Einstein work since the conversion, and is currently not in the machine, which is therefore consuming over 100 watts less than last week.
5) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82139)
Posted 21 Oct 2017 by archae86
Post:
I've flashed the most recent UEFI BIOS available for the motherboard (2.70)
I've run the Intel Chipset Device Software version 10.1.1.42 (commonly called chipset drivers, actually is an INF installer)
My UEFI settings before and after the BIOS reflash both had the Intel onchip graphics disabled
Possibly because of the disabled onchip graphics setting, my running of the Intel Graphics driver installer version Win64_165036.4703 terminated with the advice that "this computer does not meet the minimum requirements for installing this software".
I ran the Intel Driver and Support Assistant Installer 3.0, which announced that "no driver updates were found for your system", but left behind an ever-running process with tray icon (grrr...)

After these updates I tried again a configuration with the 1050 in primary and a 750 Ti in secondary. Clinfo and BOINC only saw and BOINC only used the 750Ti.

Subsequently I did a full uninstall/DDU/hardware change/driver install with two identical model 750 Ti cards.
Clinfo and BOINC see and BOINC uses only one of the two 750Ti cards.

While the Nvidia thread on problems with the Fall Creators Update rattles along, my post there has gotten zero response, and the many complaints don't generally seem to align with mine.

As this configuration is less productive than the single 1050 configuration, I'll probably revert to that if no new clever ideas turn up soon.
I shall try to avoid the Windows update on my other two multi-GPU machines until I hear others have success with BOINC in general, and Einstein in particular.
6) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82129)
Posted 20 Oct 2017 by archae86
Post:
What's the model of the motherboard on that computer? Could there be settings in the BIOS/UEFI that have an effect on the pci-e configuration? Also, if there's a newer BIOS/UEFI version available for that board I definitely recommend to go on and flash.
It is an Asrock Z87 Extreme3. I purchased it in early 2014, and the Asrock support page lists updated release dates later than that for almost everything.

Asrock pretty strongly advises not to update (especially the UEFI BIOS) unless you have a problem (probably because they have experience with users bricking their motherboards). Which is part of my defense for never having done so since I built the machine. But I clearly have a problem, and with each passing hour the lack of reports from other multi-GPU users strengthens the case that something in my particular system configuration contributes.

So I plan to gird my loins and go on an update binge today, prioritizing the BIOS at the top of the list. After I've gone through the Asrock support page for this motherboard, I'll try to see whether the Intel updater wants to do anything. I tried running that already on my primary system, and it professed to find no Intel products to update!?
7) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82116)
Posted 19 Oct 2017 by archae86
Post:
Full cold reboot with fast startup turned off and still running the couple of months-old driver made no difference.

Then I removed the 750 Ti from the secondary slot, leaving untouched the 1050 in the primary slot, but doing a full driver uninstall, DDU, up-to-date driver install.

The 1050 was recognized by clinfo and by boinc, and an Einstein task is running on it as I type.

Over at the Nvidia forum, a user has posted a complaint asserting that he cannot use multiple GPUs. However he makes no reference to issues I recognize as CUDA recognition related, so possibly his problem is materially different from mine.

I've had about enough fun for today, so plan to leave the modern driver, one card, GTX 1050 in primary slot configuration running until tomorrow.

Tentatively, I think tomorrow I shall attempt simply adding back the 750Ti during power down with no driver uninstall, ... Assuming that puts me back where I've been, my next move will be to try a dual 750 Ti card configuration, using full driver clean changeover. As the two 750 Ti's will be the exact same EVGA model ID, if they work it may be that some aspect of non-similarity among multiple cards installed is a factor.

If that does not work, I'll just put the 1050 back in alone and wait to see if someone else learns something I can go to school on.
8) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82114)
Posted 19 Oct 2017 by archae86
Post:
My power option remains the high performance one, and clicking around I could not find anything turned on to allow optional lower CPU states, any form of sleeping or hibernation, and such.

However, clicking the "choose what the power button" does item revealed to me that "turn on fast startup" was checked under the shutdown settings. Maybe the update did that to me, as I think my policy is no to enable it.

I've turned that off. The comment claims it does not apply for restarts, so I plan to do a full cold shutdown before I make another configuration change, just in case that fixes things.

Meanwhile I did install Nvidia's 385.41 driver, a relatively recent one which is successfully supporting a somewhat different two-GPU configuration on another of my systems (not yet OS upgrarded) but is not recommended by Nvidia for this OS update. It made no difference that I could detect. The system works, BOINC runs Einstein, but Clinfo reports only one card, BOINC startup messages only discuss one card, and only one Einstein task runs. (on the 750 Ti).

My next plans are to try a power off reboot. If that does nothing, I'll uninstall the driver, DDU, open the box and remove the 750Ti, so it will be a single 1050 in the primary slot, install the up-to-date driver, and see what I get, If, as I expect, it sees and uses the 1050 in the primary slot, the problem definition remains a somewhat fuzzy scope of failure to get the CUDA environment informed of one of the two cards when I have either 970 + 750 Ti or 1050 + 750 Ti.

As some of our major contributors run husky multi-CPU systems, and some of those are Windows 10, I imagine there will be screaming soon if this is a general problem. However a look at the Einstein top 50 list shows the specific release version of this Windows update (10.00.16299.00) on only one system, which belongs to Mumak, which is a AMD Radeon RX Vega system, which does not exclude the prime possibilities.
9) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82110)
Posted 19 Oct 2017 by archae86
Post:
Juha wrote:
I think you have plenty of GPUs around so what you could try is replacing the 1050 with another 750 to see if the issue is in multi GPU support in general or multi GPU generations.

As it happens, at the transition the top slot was a 970. I swapped in the 1050 after some hours of unsuccessful fiddling, figuring there might be a difference, and that in any case my plan all along had been to run that card in that system.

So the combinations tried so far are:
1. 970 (Maxwell2) in primary slot with 750 Ti (Maxwell) in secondary slot--only the 750 Ti is seen and used by BOINC.
2. 1050 (Pascal) in primary slot with 750 Ti (Maxwell) in secondary slot--only the 750 Ti is seen and used by BOINC.

At one point I had about five total 750/750 Ti cards. I'll poke around and see if I can find one to try swapping in the primary slot. I'm not sure to what degree Maxwell and Maxwell2 cards can be consider same/different generations.

I think I'll try the somewhat older driver option first. If that fails and I don't get a bright new idea, I actually plan to pull the 750 Ti out and check that the 1050 alone is detected and operates correctly, then try adding the 750 without doing uninstall, DDU, install.
10) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82107)
Posted 19 Oct 2017 by archae86
Post:
Juha wrote:
Looks like a driver bug to me. I don't really know where to report it. Nvidia has a thread for Fall Creators Update in GeForce forums but I don't know if that's the right place for computing related issues. There is also Developer forums but no threads about FCU.

I've made a post in the Nvidia Fall Creators thread you mentioned. Possibly they will advise me if there is a better place.

Is it pointless for me to uninstall the (recommended by Nvidia for Fall Creators Edition) 387.92 driver and try 385.41, which is only a couple of months old and works for me on another un-updated system? Or are older Nvidia drivers pretty much guaranteed not to work on an updated system?

Thanks
11) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82104)
Posted 19 Oct 2017 by archae86
Post:
Could you copy-paste the output of clinfo here? If it didn't come with the drivers you can download it from here.

Interesting: this viewpoint only mentions the 750 Ti and not the 1050

Number of platforms: 1
Platform Profile: FULL_PROFILE
Platform Version: OpenCL 1.2 CUDA 9.1.0
Platform Name: NVIDIA CUDA
Platform Vendor: NVIDIA Corporation
Platform Extensions: cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_fp64 cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll cl_nv_d3d10_sharing cl_khr_d3d10_sharing cl_nv_d3d11_sharing cl_nv_copy_opts cl_nv_create_buffer


Platform Name: NVIDIA CUDA
Number of devices: 1
Device Type: CL_DEVICE_TYPE_GPU
Device ID: 4318
Max compute units: 5
Max work items dimensions: 3
Max work items[0]: 1024
Max work items[1]: 1024
Max work items[2]: 64
Max work group size: 1024
Preferred vector width char: 1
Preferred vector width short: 1
Preferred vector width int: 1
Preferred vector width long: 1
Preferred vector width float: 1
Preferred vector width double: 1
Max clock frequency: 1254Mhz
Address bits: 14757395255531667488
Max memory allocation: 536870912
Image support: Yes
Max number of images read arguments: 256
Max number of images write arguments: 16
Max image 2D width: 16384
Max image 2D height: 16384
Max image 3D width: 4096
Max image 3D height: 4096
Max image 3D depth: 4096
Max samplers within kernel: 32
Max size of kernel argument: 4352
Alignment (bits) of base address: 4096
Minimum alignment (bytes) for any datatype: 128
Single precision floating point capability
Denorms: Yes
Quiet NaNs: Yes
Round to nearest even: Yes
Round to zero: Yes
Round to +ve and infinity: Yes
IEEE754-2008 fused multiply-add: Yes
Cache type: Read/Write
Cache line size: 128
Cache size: 81920
Global memory size: 2147483648
Constant buffer size: 65536
Max number of constant args: 9
Local memory type: Scratchpad
Local memory size: 49152
Error correction support: 0
Profiling timer resolution: 1000
Device endianess: Little
Available: Yes
Compiler available: Yes
Execution capabilities:
Execute OpenCL kernels: Yes
Execute native function: No
Queue properties:
Out-of-Order: Yes
Profiling : Yes
Platform ID: 00364200
Name: GeForce GTX 750 Ti
Vendor: NVIDIA Corporation
Driver version: 387.92
Profile: FULL_PROFILE
Version: OpenCL 1.2 CUDA
Extensions: cl_khr_global_int32_base_atomics cl_khr_global_int32_extended_atomics cl_khr_local_int32_base_atomics cl_khr_local_int32_extended_atomics cl_khr_fp64 cl_khr_byte_addressable_store cl_khr_icd cl_khr_gl_sharing cl_nv_compiler_options cl_nv_device_attribute_query cl_nv_pragma_unroll cl_nv_d3d10_sharing cl_khr_d3d10_sharing cl_nv_d3d11_sharing cl_nv_copy_opts cl_nv_create_buffer
12) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82100)
Posted 19 Oct 2017 by archae86
Post:
Ageless wrote:
Your given log says something different:
10/18/2017 4:59:58 PM |  | app version refers to missing GPU type intel_gpu
10/18/2017 4:59:58 PM | Einstein@Home | Application uses missing intel_gpu GPU

If you don't want to use the Intel GPU then don't select it in the Project preferences, at Einstein at https://einsteinathome.org/account/prefs/project

Adding Intel chipset drivers, or in the least making sure that the correct ones are installed, increases the way all parts of your motherboard interact with the operating system, so I would certainly try it. https://downloadcenter.intel.com/

Last for my part for now, what does Microsoft say about it? You have reported it to them as well, haven't you?

I have the "use INTEL GPU" option deselected on all four locations shown for my Einstein preferences, and have for many months. However, as Richard's comment hints, it is true that I tried running Einstein tasks of some type on the Intel on-CPU GPU on one or more of my hosts in the past, and this system is probably one of them.

As I have some lingering hope that some dangling leftover in my BOINC setup may be a part of this, a full reset and re-install is likely something I'll try within the week, if I don't stumble on some resolution sooner. On the way there, I may pull the 750 Ti (the "extra" card in a non primary PCIe slot) out of the box, and check that BOINC is then able and willing to use the 1050 primary card.

I've probably never updated the chipset drivers on this system since I built it, so I'll take a look at the link you provided, and assuming I think I understand what to do, perform an update.

I have not reported this to Microsoft, and have no idea of a suitable reporting point. If you have a suggestion, I'm all ears. At the moment, since all of GPU-Z, HWiNFO, and the Windows Control Panel device list show the GTX 1050, and BOINC seems not to see it, I'm inclined to regard this is at least in part a BOINC problem, and not necessarily a purely Microsoft or Nvidia problem. I don't know whether my starting this thread constitutes reporting this experience to BOINC. If not, is there someplace I should make a report?
13) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82096)
Posted 19 Oct 2017 by archae86
Post:
Ageless wrote:
The one thing I saw was a missing Intel GPU complaint, so advice is to reinstall the Intel (motherboard chipset) drivers as well.

I actively don't want to run BOINC work on any of my Intel GPUs, and as I only run Einstein, I accomplish that with an application preference setting at my Einstein web site account preferences by location. I also don't have a monitor connected to the motherboard graphics output on any of my systems.

Given that, do you advise Intel driver updating as appropriate for me?
14) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82090)
Posted 19 Oct 2017 by archae86
Post:
Here is the early part of the BOINCMgr event log for the system in this condition:

10/18/2017 4:59:56 PM | | Starting BOINC client version 7.8.3 for windows_x86_64
10/18/2017 4:59:56 PM | | log flags: file_xfer, sched_ops, task, http_debug
10/18/2017 4:59:56 PM | | Libraries: libcurl/7.47.1 OpenSSL/1.0.2g zlib/1.2.8
10/18/2017 4:59:56 PM | | Data directory: C:\ProgramData\BOINC
10/18/2017 4:59:56 PM | | Running under account Beth
10/18/2017 4:59:58 PM | | CUDA: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 387.92, CUDA version 9.1, compute capability 5.0, 2048MB, 1689MB available, 1606 GFLOPS peak)
10/18/2017 4:59:58 PM | | OpenCL: NVIDIA GPU 0: GeForce GTX 750 Ti (driver version 387.92, device version OpenCL 1.2 CUDA, 2048MB, 1689MB available, 1606 GFLOPS peak)
10/18/2017 4:59:58 PM | | app version refers to missing GPU type intel_gpu
10/18/2017 4:59:58 PM | Einstein@Home | Application uses missing intel_gpu GPU
10/18/2017 4:59:58 PM | | Host name: Stoll8
10/18/2017 4:59:58 PM | | Processor: 4 GenuineIntel Intel(R) Core(TM) i3-4130 CPU @ 3.40GHz [Family 6 Model 60 Stepping 3]
10/18/2017 4:59:58 PM | | Processor features: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush dts acpi mmx fxsr sse sse2 ss htt tm pni ssse3 fma cx16 sse4_1 sse4_2 movebe popcnt aes f16c rdrandsyscall nx lm avx avx2 vmx tm2 pbe fsgsbase bmi1 smep bmi2
10/18/2017 4:59:58 PM | | OS: Microsoft Windows 10: Core x64 Edition, (10.00.16299.00)
10/18/2017 4:59:58 PM | | Memory: 7.95 GB physical, 15.95 GB virtual
10/18/2017 4:59:58 PM | | Disk: 1.82 TB total, 1.73 TB free
10/18/2017 4:59:58 PM | | Local time is UTC -6 hours
10/18/2017 4:59:58 PM | | Config: GUI RPCs allowed from:
10/18/2017 4:59:58 PM | | 192.168.1.100
10/18/2017 4:59:58 PM | | 192.168.1.101
10/18/2017 4:59:58 PM | | 192.168.1.102
10/18/2017 4:59:58 PM | | 192.168.1.103
10/18/2017 4:59:58 PM | | 192.168.1.104
10/18/2017 4:59:58 PM | | 192.168.1.105
10/18/2017 4:59:58 PM | | 192.168.1.106
10/18/2017 4:59:58 PM | | 192.168.1.107
10/18/2017 4:59:58 PM | | 192.168.1.108
10/18/2017 4:59:58 PM | | 192.168.1.109
10/18/2017 4:59:58 PM | | 192.168.0.100
10/18/2017 4:59:58 PM | | 192.168.0.101
10/18/2017 4:59:58 PM | | 192.168.0.102
10/18/2017 4:59:58 PM | | 192.168.0.103
10/18/2017 4:59:58 PM | | 192.168.0.104
10/18/2017 4:59:58 PM | | 192.168.0.105
10/18/2017 4:59:58 PM | | 192.168.0.106
10/18/2017 4:59:58 PM | | 192.168.0.107
10/18/2017 4:59:58 PM | | 192.168.0.108
10/18/2017 4:59:58 PM | | 192.168.0.117
10/18/2017 4:59:58 PM | | 192.168.0.157
10/18/2017 4:59:58 PM | | 192.168.0.109
10/18/2017 4:59:58 PM | | 192.168.0.160
10/18/2017 4:59:58 PM | | 192.168.0.151
10/18/2017 4:59:58 PM | | 192.168.0.145
10/18/2017 4:59:58 PM | | 192.168.0.170
10/18/2017 4:59:58 PM | | Config: use all coprocessors
10/18/2017 4:59:59 PM | Einstein@Home | URL http://einstein.phys.uwm.edu/; Computer ID 10659288; resource share 99
10/18/2017 5:00:04 PM | Einstein@Home | General prefs: from Einstein@Home (last modified ---)
10/18/2017 5:00:04 PM | Einstein@Home | Computer location: home
10/18/2017 5:00:04 PM | | General prefs: using separate prefs for home
10/18/2017 5:00:04 PM | | Preferences:
10/18/2017 5:00:04 PM | | max memory usage when active: 4069.28 MB
10/18/2017 5:00:04 PM | | max memory usage when idle: 7324.70 MB
10/18/2017 5:00:05 PM | | max disk usage: 60.00 GB
10/18/2017 5:00:05 PM | | max CPUs used: 3
10/18/2017 5:00:05 PM | | (to change preferences, visit a project web site or select Preferences in the Manager)
10/18/2017 5:00:05 PM | | [http] HTTP_OP::init_get(): https://einsteinathome.org/rss_main.php
15) Message boards : Questions and problems : Lost multi-GPU BOINC capability with Windows 10 Fall Creators Update (Message 82089)
Posted 19 Oct 2017 by archae86
Post:
I have run three systems with dual dis-similar GPUs running Einstein applications under BOINC for many months.
Yesterday I was an early adopter after production release of the Windows 10 Fall Creators Update 10.00.16299.00 on one of those systems.

Ever since then BOINC on startup only has messages relating to one of the two GPUs installed in the system, and only runs Einstein work on that one. Oddly it is the less capable GPU, of older generation, in a lower priority PCIe slot, which is not connected to a monitor.

I downloaded from Nvidia and installed the currently recommended driver AFTER I did the Windows 10 update.
At the suggestion of someone on the Einstein forum I also tried the current beta developer's Vulkan driver, with no difference observed.

The system in question is listed here at Einstein.

Operating System: Microsoft Windows 10 Core x64 Edition, (10.00.16299.00)
CPU: Intel i3-4130 (Haswell)
GPU seen by BOINC: GTX 750 Ti in lower PCIe slot, not connected to any monitor or other device
GPU not mentioned or used by BOINC: GTX 1050 in top PCIe slot, driving a monitor
Nvidia driver installed: 387.92

While in my first Einstein post on this topic I claimed that other elements of the system only saw one GPU, that was wrong. The monitoring programs HWiNFO and GPU-Z report on both GPUs. Both GPUs are listed in the Control Panel Device Manager Display Adapter section.

When I have made (several) new attempts at display driver installation, my general practice has been this sequence:
1. run the uninstaller for the current driver as found in the Control Panel Programs and Features section
2. reboot to Windows Safe miminal
3. run DDU in the preferred mode
4. reboot and run the Nvidia installer, ticking the clean install box, and selecting on the graphics driver (no optional components)
5. reboot again and run BOINCMgr

I'm typing these detailed notes on another system, and find that I can't get a stable representation of the startup logging on the system of interest through BoincTasks for some reason, so I'll edit this message after posting or add a reply with that information after I post this.

While it could be that my loss of multi-GPU capability with the Fall Creators Update is coincidence, I'm concerned that for systems as I configure them it might be systematic, in which case I shall soon lose about half my total Einstein contribution capacity. I'd appreciate any ideas for things I might try.
16) Message boards : Questions and problems : Automatic Temperature regulation (Message 33483)
Posted 22 Jun 2010 by archae86
Post:
I got Windows 7 Ultimate in on Saturday
<snip?
But I am still trying to get to grips with this version of Windows. When coming from XP, it's a big difference.
I had a pure XP Pro fleet until a couple of months ago, when I added a laptop with Windows 7 Home Premium.

The one crucial tweak which took W7 from a pain of excess popups to mostly rather nice for use for me was turning off UAC. I already run COMODO, which protects from many of the same things, and, rather crucially, remembers my answers, so I felt fairly safe in turning it off.

Of course your annoyances and safe path may vary. The other thing that troubled me was the rather different way W7 and XP handled shared files on a personal network. I have no easy answers for that one, however.
17) Message boards : Questions and problems : Cores vs Threads ? (hyperthread matter?) (Message 33480)
Posted 21 Jun 2010 by archae86
Post:
Last year I saw the tasks web page of a hyperthreaded computer running 8 CPDN HadAM3P climate models that were not designed for hyperthreading. IIRC it was a decent computer but the models were advancing I think 8 times more slowly than on my C2D 6600. I've never seen a slower speed on any other computer.

The lesson is not to use HP for CPDN models until a type is developed specially for it. This is planned but not soon. Or if you do try it, check what's happening.
I don't know what coding for HT benefit would mean, other than trying to get a smaller working set and other measures of RAM footprint.

If you saw a really dramatic slowdown, and there was not something non-comparable going on, then the most obvious possibility would be that the HT variant, with double the memory demand, pushed the system into heavy enough disk-swapping to slow it severely.

That is not what was going on in the case I observed. My systems generally have substantial RAM relative to the demands of the BOINC projects I've used them on.

But performance degradation when an appreciable amount of memory activity spills down to the next speed tier, whether from cache to RAM, or from RAM to disk, is really severe. So depending on configuration and code, could be a source of HT performance loss well below break-even.

Not that HT is required to get this effect. Someone was running a monster server possessing a large number (at least eight, I think) of the Intel hex-core processor Dunnington chips designed in India on BOINC a while back. The total throughput per core, and the execution time per result were just awful. I think the problem was that the system configuration provided far less RAM bandwidth per core than the smaller also Penryn-generation systems to which I compared it, so the processors spent most of their time waiting for RAM requests to complete.
18) Message boards : Questions and problems : Cores vs Threads ? (hyperthread matter?) (Message 33450)
Posted 18 Jun 2010 by archae86
Post:
IIRC, some of the earlier NetBurst experiments showed much better results with dis-similar tasks - one SETI with one Einstein was a favourite pairing.

< snip>
We've also lost Tony (mmciastro), who used to do similar testing with mainly AMD processors (if I may be permitted to use those letters in this company!)

No objection from me--I as a former employee have more concrete reasons to dislike Intel than most people do, though AMD fans tend to have serious blind spots to flaws and less than competitive aspects of that product set.
So I wonder if the opening question has yet been answered with iN technology and diverse projects?

(posting from a Williamette, as it happens)

I've looked at ap diversification benefit (specifically for Einstein and ordinary SETI) myself, as it happens, and even got the honor of having my results pointed to several times by one Joe Segur! and commented on by you. Those results were observed on a Q6600 (4-core Conroe). But I never looked at the question of application diversity benefit vs. hyperthreading (Conroe does not do HT). You are right in assuming there was nothing but SETI on Mark's system when I monitored it. I've certainly done nothing at all on i7 behavior in the face of ap diversity at all, still less on the HT interaction.
19) Message boards : Questions and problems : Cores vs Threads ? (hyperthread matter?) (Message 33443)
Posted 18 Jun 2010 by archae86
Post:
Would it be fair to say that those earlier comparisons were done on NetBurst-era HT processors? Have you had any chance to repeat them on the Core iN range, or do you know anyone else who has?
I wrote a long answer yesterday in this thread. Not sure if it was moderated away, or whether I failed to click on the post button after previewing it.

I'll recast the text part of my answer: my own personal owned system comparisons were done on a Gallatin, which is the large-cache variant of Northwood, which in turn was the next-process implementation of Willamette (with some appreciable improvement). So, yes, marketing called them all NetBurst, and they all were from a diseased branch of the Intel microprocessor tree--now happily cut off in favor of the vastly better Conroe and Nehalem branches. I don't currently operate any hosts capable of HT.

But my most recent measurement of this kind was on msattler's Frozen Nehi. The first was before it got frozen, and sadly, it also only had one of the three channels of RAM populated at the time, rendering the results of rather limited application. Still, they showed a quite modest hyperthreading productivity benefit in two Angle Ranges which had quite a bit of work at the time, and a slight disadvantage in another Angle Range region.

On the chance that embedded images are forbidden here, but links permitted, I'll include a couple of links this time:

Single RAM channel Nehalem HT comparison by AR

same comparison--expanded view near 0.4 AR


Much later in the Frozen Nehi's life, Mark undertook another comparison--this time on Astropulse, and this time running with RAM channels fully populated with high-performance RAM, overclocked as Mark would. With RAM starvation not getting in the way to nearly the degree seen in those first comparisons, HT but highly consistent productivity improvement--ballpark 10%.

Astropulse comparison on fully populated and overclocked system




20) Message boards : Questions and problems : Cores vs Threads ? (hyperthread matter?) (Message 33412)
Posted 16 Jun 2010 by archae86
Post:
The answer is highly application dependent, and also will vary with the particular CPU implementation.

In direct careful comparisons, I've commonly seen same system net throughput improvement on the order of 10 to 20% in comparing running HT vs. running with HT disabled. But there certainly have been cases well outside that range (including a pathological case in which running HT actually lowered net throughput on one short series of Einstein third-party aps).

Now if, on the other hand you are comparing completely different architectures or generations, then the HT portion of the comparison is of minor importance compared to everything else.


Next 20

Copyright © 2024 University of California.
Permission is granted to copy, distribute and/or modify this document under the terms of the GNU Free Documentation License, Version 1.2 or any later version published by the Free Software Foundation.