Posts by HAL9000


log in
21) Message boards : Number crunching : A journey: iGPU slowing CPU processing (Message 1543959)
Posted 4 days ago by Profile HAL9000
Hal, I suspect the Haswell "smart cache" implementation may be at least part of the problem. SaH v7 tasks use a lot of cache when doing various operations, and having that single 6 MiB cache shared between all processing units may thrash it.

It might also be interesting to see how AP on the iGPU interacts with CPU processing. There's an open Beta at Crunchers Anonymous for AP6_win_x86_SSE2_OpenCL_Intel_r2180.
Joe

I was just wondering if the cache might be the issue & perhaps that could be why I am not seeing the same behavior on Silvermont/Bay Trail-D. It uses a separate 1MB cache for each pair of cores. I am not sure how the iGPU is allocated cache in that configuration. The block diagram here may be helpful in some way.

I did try the iGPU AP app a while back. I even pumped out 12 tasks on this machine. It was dreadfully slow. Around 12-13 hours per AP IIRC. However I was running 4 MB tasks on the CPU. At the time I just figured the iGPU must not be good at doing AP & moved on.

The next step is obviously to run AP on the iGPU alone & see how it responds.
22) Message boards : Number crunching : A journey: iGPU slowing CPU processing (Message 1543734)
Posted 5 days ago by Profile HAL9000
Forgot to list the MB CPU apps previously:
akv8c_r2202_x64_avx
akv8c_r2202_x64_ssse3

I ran several more tests.
Configuration: 3 CPU AP + 1 iGPU MB
Configuration: 2 CPU AP + 1 iGPU MB
Configuration: 1 CPU AP + 1 iGPU MB

In each test the results were:
CPU times 7.5-8 hr, iGPU times ~1 hr

Now it seem less like a memory bottleneck to me.
The iGPU seems to be doing something that is causing the AP tasks to run nearly twice as slow.

Giving a test with -cpu_lock to see what happens. If the AP CPU times are still slowed with that setting I will probably just configure to only run iGPU when running MB is running on all 4 CPU.
23) Message boards : Number crunching : Some questions about a laptop (Message 1543725)
Posted 5 days ago by Profile HAL9000
Sorry, didn't see that, thx Hal!
So yeah, your temps are MUCH lower. I guess that's because your machine ain't a laptop. So maybe I have to wait for winter to see my machine run in burst;-)

I suppose you could but blocks of dry ice to set the laptop on. That should keep it nice and cool. :)

The heatsink on my system is quite large. It looks like it might be twice the size of the one on my Core 2 Duo T7200 system. Which i on an old Abit il-90mv motherboard.
The C2D T7200 is the system that the Celeron J1900 is replacing. At least it will when I am done testing it. Then get around to getting the rest of the parts for the system so that it isn't laying sprawled out across a table.
24) Message boards : Number crunching : Some questions about a laptop (Message 1543703)
Posted 5 days ago by Profile HAL9000
Perhaps it is to warm to run in Burst mode? Your cores temps are well below the designed limit of 100ºC.

Power settings on the laptop will also limit the CPU (and GPU) maximum frequencies.

Do you mean the cpu settings in advanced energy options? Those are set to 100%.

BTW: Not getting any tasks for GPU may have a simple reason: I used the lunatics installer and simply may not have any app info for Intel GPU. I will try to fix that at evening when I'm back home.


@Hal9000: Would be great if you could check your temps/clock speed!

Linked to an image in this post.
http://setiathome.berkeley.edu/forum_thread.php?id=75172&postid=1543438#1543438
25) Message boards : Number crunching : Some questions about a laptop (Message 1543700)
Posted 5 days ago by Profile HAL9000
Perhaps it is to warm to run in Burst mode? Your cores temps are well below the designed limit of 100ºC.

Power settings on the laptop will also limit the CPU (and GPU) maximum frequencies.

True. I always configure all of my machines for High performance. Then adjust the settings for when on battery a bit lower than maximum.
The laptop manufacture could also limit the functionality to a greater amount than Intel designed it in order to maximize battery life.
26) Message boards : Number crunching : A journey: iGPU slowing CPU processing (Message 1543505)
Posted 5 days ago by Profile HAL9000
Initially I didn't think anything of my AP times on my i5-4670K as they were faster than any other system I have run previously. After finding older generation i5's, with lower clock speeds, running AP tasks faster than my two i5-4670K systems I did some observations.

Apps used:
ap6_win_x64_avx_cpu_r2163
ap6_win_x64_sse3_cpu_r2163
mb7_win_x86_sse_opencl_intel_r2170

Times are based on:
MB tasks with an AR near 0.42
AP tasks with blanking between 0-5%

Configuration: 4 CPU MB + 1 iGPU MB
CPU times ~2 hr, iGPU times ~1 hr

Configuration: 4 CPU MB, No iGPU running
CPU times ~2 hr

Configuration: 4 CPU AP + 1 iGPU MB
CPU times 7.5-8 hr, iGPU times ~1 hr

Configuration: 4 CPU AP, No iGPU running
CPU times 4-4.5 hr

Currently I am testing
Configuration: 3 CPU AP + 1 iGPU MB

If that provides fruitful I am going to test.
Configuration: 3 CPU AP + 1 CPU MB + 1 iGPU MB

From these observations it looks like running 4 AP tasks with MB on the iGPU might possibly be hitting the memory bandwidth limit. Given the iGPU was not slowed with the 4 AP tasks. It looks like the iGPU might be given priority to the memory.

I am also testing this on my Silvermont based system. It does not show a change in speed when not using the iGPU. Perhaps 4 AP tasks are maxing it out alone.
27) Message boards : Number crunching : Some questions about a laptop (Message 1543438)
Posted 5 days ago by Profile HAL9000
Even if the two apps I'm using don't agree on the exact temps. The temps on my machine are much lower. The ambient room temp is about 82ºF/28ºC.
28) Message boards : Number crunching : Some questions about a laptop (Message 1543387)
Posted 5 days ago by Profile HAL9000
I soon figured out that I don't have the time and nerves right now to learn linux, so I had to install Windows. I chose 8.1 because I couldn't find any Intel HD drivers for Windows 7.

Any ideas why I don't get work for GPU?

You can always get their drivers from http://downloadcenter.intel.com
In your SETI@home preferences is Use Intel GPU set to Yes?

Yes, GPU use activated in both, seti account settings and client. The GPU is recognized, but no work for GPU is requested.

Some other things that I discoverd within the last hour:

- According to RealTemp the CPU is always running at 2.166 GHz but never goes into Burst (2.42 GHz). HAL9000 said that his CPU runs with Burst-Frequency all the time, so I wonder why mine doesn't?!?

- I suspended Seti and resumed it fter a few minutes and it took very long until the tasks showed progress again. They were stuck at the same percentage for about 10 minutes. I never saw something like that on my desktop.

Perhaps it is to warm to run in Burst mode? Your cores temps are well below the designed limit of 100ºC. I'll have to check mine and see what the temps are and where its cut over temp is for Burst.
29) Message boards : Number crunching : New to SETI@home, have some questions. (Message 1543074)
Posted 5 days ago by Profile HAL9000
Hal900 I noticed your stats don't have Einstein so I'm asumeing you havn,t done it and it's been a while since I did all I can remember they did take a long time just can't remember how long so he mite not have a problem wish I could member how long they took .

Your get more credit per unit as they do take a while to complete my Rac doubled I think when I did it and why it's at 1.9 mill and seti is only just got passed 1 mill

So if HT isn't the prob maybe he don't have a prob now I think back AP's mite just take some 36hrs the Gpu I think took 9hrs oh well

Change to Seti and then he can get a better comparison

I have not run Einstein. It is just one I haven't chose to play around with yet.

As I have a machine rather similar to the OP's machine I can see there SETI@home performance is a bit lower than expected. So I suggested changes for improving it. Having and older OpenCL runtime with a newer driver can be a source of problems. Changes to improve their SETI@home performance I imagine would carry over to Einstein.

Lunatics 64-bit SSSE3 app & HT enabled, while running 8 tasks, provided 31.2% greater output & consumed 11.1% more power. Compared to just running 4 tasks.


Hal9000 I am thinking that I'm not that far off about HT as you said you only get a 30% increase but your doing 100% more work so maybe it's just the fact your doing so many that gives a increase at all .

I would expect there to be at least 80% increase not only 30% which is fine for the better chips but a 2 core I3 with 4 threads mmmmm

Your expectations for HT are to to high then. The normal expected gains from HT are 15-35%. When using HT tasks run a bit slower, but running twice as many = more work done over time. This machine Will run 8 AP at a time with times around 16 hours vs running 4 at a time in around 12 hours. So in 48 hours HT on = 24 tasks & HT off = 16 tasks. When running a reduced number of tasks with HT on times are somewhere in the middle.

So for SETI@home and what others have said about running Einstein on the GPU. It would seem they would want to run 3 CPU tasks & 1 GPU task to improve their performance.
30) Message boards : Number crunching : wu's pending validation (Message 1543058)
Posted 5 days ago by Profile HAL9000
i have wu's from the begining of june that still have not been validated
was just wondering what was going on

http://setiathome.berkeley.edu/results.php?userid=9995198&offset=0&show_names=0&state=2&appid=11

As was said this is not uncommon. Luckily I only have tasks going back to March that are still pending.

Also your link is to your user task list is only accessible to you. Here is mine & you will get "Unable to handle request" message.
http://setiathome.berkeley.edu/results.php?userid=792
If you want to direct others to something specific you have to do it by machine task list. Such as this.
http://setiathome.berkeley.edu/results.php?hostid=7310554&offset=0&show_names=0&state=2&appid=
http://setiathome.berkeley.edu/results.php?hostid=7313906&offset=0&show_names=0&state=2&appid=
http://setiathome.berkeley.edu/results.php?hostid=7340582&offset=0&show_names=0&state=2&appid=

Also good news. The tasks seem to have been validated. As your oldest current pending was returned July 11th.
31) Message boards : Number crunching : New to SETI@home, have some questions. (Message 1542921)
Posted 6 days ago by Profile HAL9000
Hal8000 the chip he is running I think is 1st gen I3 so HT mite not be too good.

Still not used to this AMD but it does have 8 cores and 8 threads but I only see 8 not 16 and I have not seen how to turn HT on the M/B

Gigabyte 970a-d3p m/b so I'm not shore if it has it as I havn't found it yet any help would be most grate full and then I can try it and see if I get much of a improvement

There is the heating prob he has maybe it's just getting to hot if you reackon HT doesn't do anything ...me not so shore but i'll take what you said on board and leave my judgement till I can try myself

Their i3-2100 is a 2nd generation. I have a first generation i3-390M.

Hyper-Threading is an Intel technology, but not to worry. Your chip has 8 cores, but only 4 FPUs. With 1 FPU being shared between 2 cores. So it has a similar effect to having a 4 core processor with HT.

HT does increase heat but their temps are within normal limits. Also their HD6450 is running a bit cooler than my HD6370 is right now pumping out AP work.
32) Message boards : Number crunching : The current top cruncher (Message 1542904)
Posted 6 days ago by Profile HAL9000
So he possible doesn't have even a Hackintosh but runs OSX from a VM mmmmmmmmmmmmmmmmmmmmmmm

Don't listen to me I'm just a DUMB DUMB like our PM hahahahahah joke guys

Unless someone asks them, or they post what they are running. It is all just speculation fun.
"And now it is time for everyones favorite game. Guess how much their electric bill might be!"
33) Message boards : Number crunching : New to SETI@home, have some questions. (Message 1542903)
Posted 6 days ago by Profile HAL9000
Hal I had a m/b with HYPER thread and I didn't know at the time when machine started up it looke like 2 cores on the task manager so I thought great till it started doing seti and was so dam slow.

So after a bit of investigation I found out it was only a Pentium chip 1 core but the m/b had HYPER THREAD and when I turn'd it off it speed up so I have never used HYPER THREAD for this reason as you said it's a Virtual core and it will slow things down

I think you may have had another issue with your system slowing down. I've been using HT since it came out on the Pentium 4 without any issues. I recently just compared the efficiency of the current apps with HT on and off on one of my i7-860 systems. Using the Lunatics 64-bit SSSE3 app & HT enabled, while running 8 tasks, provided 31.2% greater output & consumed 11.1% more power. Compared to just running 4 tasks.
I have not run this same test of the 20 or so other machines I have that are running HT, but I would expect a similar result. Possibly with the older generation hardware showing less of a performance to power usage gain.
There are some types of processing that do not gain any benefit from HT. If there are BOINC projects that are like this I would limit the concurrent number of those tasks running at once.
34) Message boards : Number crunching : New to SETI@home, have some questions. (Message 1542887)
Posted 6 days ago by Profile HAL9000
Robert as I have already told you

TURN OFF HYPER THREADING YOU ONLY HAVE 2 FREAKING CORES ON THAT CHIP MATE

it's that freaking simple you can only do a total of 2 units on the CPU NOT 4 or 6
CPU on seti uses 0.04% of cpu to feed the GPU

THE FACT YOUR OVERHEATING IT AND RUNNING WAY TO MANY UNITS LOOK at my score for Einstein nearly 2 million mate

You have to turn off the HYPER THREADING Einstein units do take longer to do but not that long

The setting should be on the M/B TURN IT OFF

I have no idea where you are getting them running 5 or 6 tasks on the CPU, or that their system is overheating. There is no reason for them to disable Hyper-Threading.

When running Astropulse on a GPU a "core" per AP task should be reserved. Which they can easily do by setting their processor usage to 75% of the processors. It doesn't matter if it is a real core or a virtual HT core. Running 3 MB on the CPU & one free for the 1 AP GPU task. Which is how I configured my i3 system. Even before I reserved a "core" the AP tasks never ran as long as the tasks on that system.
35) Message boards : Number crunching : New to SETI@home, have some questions. (Message 1542798)
Posted 6 days ago by Profile HAL9000
My HD6370M in my notebook runs about 6 hours for a SETI@home AP task with about 10-15 minutes of CPU time. With your GPU being rated at 200-240 GFLOPS vs mine at 120GFLOPS I would imagine your AP tasks would run much faster and use less CPU time.

Perhaps the driver set you are using is not the most ideal. Which could also be true for the Einstein work.

The Cat 13.1 set is bad as it has an OpenCL runtime compiler bug. It might apply to all projects that run OpenCL. I am not sure, but probably.

Looking at the tasks you have complete. You have a version of drivers anywhere from Cat 12.6 to 13.4 & a really old OpenCL runtime from 11.6.

I would suggest shutting down BOINC. Uninstall the video package and clean out your old driver with some tools to get rid of the old junk. Display Driver Uninstaller has been suggested before as a good tool to use.

Then freshly install drivers. The current Cat 14.4 should be OK.
36) Message boards : Number crunching : The current top cruncher (Message 1542782)
Posted 6 days ago by Profile HAL9000
Can someone elaborate on "Pulling a NEZ" ? That one is over my head!

http://setiathome.berkeley.edu/forum_thread.php?id=56411
37) Message boards : Number crunching : The current top cruncher (Message 1542777)
Posted 6 days ago by Profile HAL9000
Well, I found some circumstantial evidence in my databsse that seems to support the "mostly Mac" theory. I found that over 90% of the host IDs added to my DB between 7330981 and 7331846 are owned by Anonymous, 159 of them, all added to S@h on July 6. (A host gets added to my DB the first time it serves as a wingmate on a WU that one of my hosts received.) I checked a handful of them and all appeared to be 16-processor Darwin boxes, though I saw a couple different Intel chips listed.

As I said, it's just circumstantial, but the addition of these machines seems to coincide with our rising star. I'm sure there are a lot more Anonymous machines in there that I just haven't happened to be paired with (yet), such as the one Juan mentioned.

Maybe one of the stat sites lists all of the hosts by ID. That would be helpful. That would be faster than downloading the stats dump. With the hosts file being about 600MB compresses it isn't something I want to parse.

Given the Xeon X5675 is a 6c/12t chip that does support the idea that the OS might be in a VM.
38) Message boards : Number crunching : The current top cruncher (Message 1542771)
Posted 6 days ago by Profile HAL9000
I want a Hackintosh..lol..but I'm betting he's running true Macs..

I thought I wanted a Hackintosh, 'til I found you had to use someone's hacked OSX image. I wasn't impressed. Then I had problems with the 2008 Intel board I bought to use as a Hack. Seems Intel 'fixed' their board to where it wouldn't post when a video card with a Mac Rom was used for the main monitor. Bad Intel, BAD! Then I had Ram problems with the same board so I gave up on it. Now I have corrected the Ram problem and have a worthy PC video card. But you still have to use someone else's hacked system. Dunno, maybe someday.

We didn't get very far trying this at work either. We were even using Optiplex 745 machines that were listed on the hackintish site as fully compatible.

The last true Xserve boxes I am aware of were only dual Core 2 Quad based Xeons, but I'm not a huge Apple person. Do you think Apple still working with companies to produce Xserve setups?
39) Message boards : Number crunching : The current top cruncher (Message 1542763)
Posted 6 days ago by Profile HAL9000
Or they are just running the linux machines they mentioned.

This problem only exists with the OS X Seti client... If there is no fix, I'll have to switch away from implementing Seti in our datacenters across the board.

Posts by Charles Long
Sounds like a bunch of Macs to me. Someone can't read that? He made 3 posts about getting all 16 cores on his 16 core OSX machines to work. Then mentioned his Datacenter machines that are also having the same problem. Seems pretty convincing to me.

He was asking about a problem he as having on the MAC systems, but also stated Linux boxen were in the mix.
Here ya go. http://setiathome.berkeley.edu/forum_thread.php?id=74994&postid=1530999
In addition, my Linux hosts, which use the same global settings, do not have a 8 task limitation and use the all available cores; I have tested multiple systems with different numbers of cores and this is always the case. That and other projects on OS X, such as asteroidsathome, which use the same global settings, they also use all available cores. This limitation is exclusive to the setiathome OS X client.
40) Message boards : Number crunching : The current top cruncher (Message 1542696)
Posted 6 days ago by Profile HAL9000
he might be running it down when I checked a hour or 2 ago at free-dc it looked like he mite be backing off

grrr bloody hope so man 18 mil a day ............................picks jaw up off the floor

The stat sites only update about once a day from the dump that the servers do. Yesterday they had a lower number because of maintenance. Since yesterday they have already gained over 14,000,000 credit.


Previous 20 · Next 20

Copyright © 2014 University of California