Windows x86 SSE2

Fantastic! Thank you. I'll hook my ThinkPad up to Beta a little later and see how it runs overnight.

I would suggest to check SSE vs SSE2 speed before. They both did not tested on target (that is, SSE+SSE2-only) hardware so far. So what is faster - unknown.

Okay, I've got the SSE running on its first task now. AR is 2.729883, so it should run quicker than a normal AR, but I don't have anything to compare it to with v8. On v7, ARs around 2.7 seemed to run about 5.5 hours on SSE2 vs. about 6.0-6.5 hours on stock.

I'll let SSE run overnight, then switch to SSE2 tomorrow.

Okay, the ThinkPad Pentium M has been running Beta for a couple days now, first with SSE, then with SSE2.

I ran 3 tasks with SSE, all with ARs of 2.72+, with run times between 6 hours 27 min and 6 hours 39 min. I got one task for SSE2 with the same approximate AR that ran for 5 hours 47 min. That compares reasonably with the 5.5 hour average RT for similar ARs under SSE2 on v7, as noted in my earlier post, and is certainly much better than SSE.

I then got a couple "normal" ARs (0.414+) for SSE2, which ran in 13 hours 50 min and 13 hours 43 min. That's a bit longer than the 12-13 hours that a similar AR would take on v7, but much better than the first (and so far, only) v8 task I've run with the stock app on Main. That one had an AR of 0.448057 and took 18 hours 53 min. (I suspended a similar task on Main in order to run these Beta tests. Progress on that task is at 51% after 10 hours 19 min.)

I've got a VLAR runnning now with SSE2 on Beta, which probably won't finish until tomorrow morning, at which time I'll switch the ThinkPad back to Main. All in all, I think SSE2 will be the way to go, just as it was in v7.

Mac OS X 8.05 CPU apps, both i686 and x86_64, do not use AVX and show error for sse1 chirp functions.

Optimal function choices:
--------------------------------------------------------
name timing error
--------------------------------------------------------
v_BaseLineSmooth (no other)

v_GetPowerSpectrum 0.000106 0.00000 test
v_vGetPowerSpectrum 0.000048 0.00000 test
v_vGetPowerSpectrum2 0.000066 0.00000 test
v_vGetPowerSpectrumUnrolled 0.000047 0.00000 test
v_vGetPowerSpectrumUnrolled2 0.000068 0.00000 test
v_avxGetPowerSpectrum not supported on CPU
v_vGetPowerSpectrumUnrolled 0.000047 0.00000 choice

v_ChirpData 0.006124 0.00000 test
fpu_ChirpData 0.012480 0.00000 test
fpu_opt_ChirpData 0.006985 0.00000 test
v_vChirpData_x86_64 0.042303 0.00000 test
sse1_ChirpData_ak 0.005734 0.94721 test
sse1_ChirpData_ak8e 0.004785 0.94721 test
sse1_ChirpData_ak8h 0.005001 0.94721 test
sse2_ChirpData_ak 0.005522 0.00000 test
sse2_ChirpData_ak8 0.004694 0.00000 test
sse3_ChirpData_ak 0.005457 0.00000 test
sse3_ChirpData_ak8 0.004428 0.00000 test
avx_ChirpData_a not supported on CPU
avx_ChirpData_b not supported on CPU
avx_ChirpData_c not supported on CPU
avx_ChirpData_d not supported on CPU
sse3_ChirpData_ak8 0.004428 0.00000 choice

v_Transpose 0.008110 0.00000 test
v_Transpose2 0.005504 0.00000 test
v_Transpose4 0.003012 0.00000 test
v_Transpose8 0.004995 0.00000 test
fftwf_transpose 0.002315 0.00000 test
v_pfTranspose2 0.004646 0.00000 test
v_pfTranspose4 0.002924 0.00000 test
v_pfTranspose8 0.004991 0.00000 test
v_vTranspose4 0.002553 0.00000 test
v_vTranspose4np 0.002678 0.00000 test
v_vTranspose4ntw 0.007978 0.00000 test
v_vTranspose4x8ntw 0.003384 0.00000 test
v_vTranspose4x16ntw 0.001036 0.00000 test
v_vpfTranspose8x4ntw 0.007640 0.00000 test
v_avxTranspose4x8ntw not supported on CPU
v_avxTranspose4x16ntw not supported on CPU
v_avxTranspose8x4ntw not supported on CPU
v_avxTranspose8x8ntw_a not supported on CPU
v_avxTranspose8x8ntw_b not supported on CPU
v_vTranspose4x16ntw 0.001036 0.00000 choice

FPU opt folding 0.001247 0.00000 test
ben SSE folding 0.000640 0.00000 test
AK SSE folding 0.000578 0.00000 test
BH SSE folding 0.000606 0.00000 test
JS AVX_a folding not supported on CPU
JS AVX_c folding not supported on CPU
AK SSE folding 0.000578 0.00000 choice

But at the same time the fftw wisdom shows avx codelets are in use :

(fftw-3.3.4 fftwf_wisdom #x9e7d4dee #xdb14fed1 #x34bf76a4 #xeb6e8fdf
(fftwf_codelet_n2bv_16_avx 0 #x11bdd #x11bdd #x0 #xbbdb1c2f #x6c107f3e #x361c5939 #x3317902a)
(fftwf_rdft_rank0_register 4 #x11bdd #x11bdd #x0 #xd8d84312 #xb42e4b70 #x33692be2 #x09c677d4)
(fftwf_codelet_t2bv_8_avx 0 #x11bdd #x11bdd #x0 #x8a2f4f6f #xf1259a8c #x4f6913c8 #xd0b630ce)...

---

_\|/_
U r s

A lot of gpu wus without errors, only 1 failed:
21648156

What version of NVIDIA driver is required here?

I use:
Coprocessors NVIDIA GeForce GTX 750 Ti (2048MB) driver: 347.88 OpenCL: 1.1

I like it and it have been working sooooo well on the main project.
But I guess one change means another has to come.

What version of NVIDIA driver is required here?

I use:
Coprocessors NVIDIA GeForce GTX 750 Ti (2048MB) driver: 347.88 OpenCL: 1.1

I like it and it have been working sooooo well on the main project.
But I guess one change means another has to come.

The OpenCL application is set to require driver >350 (and some of the even higher ones seem to be giving problems again).

The CUDA applications, once ready, will have the same driver restrictions as at Main.

What version of NVIDIA driver is required here?

I use:
Coprocessors NVIDIA GeForce GTX 750 Ti (2048MB) driver: 347.88 OpenCL: 1.1

I like it and it have been working sooooo well on the main project.
But I guess one change means another has to come.

OpenCL NV requires 350.xx+ drivers.

---

News about SETI opt app releases: https://twitter.com/Raistmer

@Eric there are several reports on main of the Mac app not working properly. e.g. host 4153607

report on main (quoting earlier reports) and f. from user Paris (host linked above).

I suppose that's one for Charlie but I don;t know if he reads here.

---

A person who won't read has no advantage over one who can't read. (Mark Twain)

What cards are this app for? 8.05 (opencl_ati5_cat132)

Or am i just mixing up the ATI HD5 numbers....ATI 5xxx and above.

This is just about as minor as minor can be, so even though it's off-topic I'm putting it here as it's not worth starting a new thread over:

The Beta 1% badge file appears to have gone missing so is showing as a broken link/missing image. Path is https://setiweb.ssl.berkeley.edu/beta/img/pct_1.png
(the pct_5.png and pct_25.png in the same path display normally.)

This is just about as minor as minor can be, so even though it's off-topic I'm putting it here as it's not worth starting a new thread over:

The Beta 1% badge file appears to have gone missing so is showing as a broken link/missing image. Path is https://setiweb.ssl.berkeley.edu/beta/img/pct_1.png
(the pct_5.png and pct_25.png in the same path display normally.)

I see same thing and some more pictures missing in the forum and the picture on team page for the profile link as well....

Probebly permissions on the picture library.

I'm missing the unread and unread/locked thread icons, at least on my Windows machines,
and the Cross-project stats icons on the my 'Your computers' page.

Claggy

It looks like since after the Tuesday's outage this is missing inter alia [with BBC's url - img]:

[Windows, Firefox]

http://setiweb.ssl.berkeley.edu/beta/img/pct_1.png -

http://setiweb.ssl.berkeley.edu/beta/img/pct_5.png -

http://setiweb.ssl.berkeley.edu/beta/img/pct_25.png -

http://setiweb.ssl.berkeley.edu/beta/img/head_20.png -

http://setiweb.ssl.berkeley.edu/beta/img/post.png -

http://setiweb.ssl.berkeley.edu/beta/img/unread_post.png -

http://setiweb.ssl.berkeley.edu/beta/img/locked_post.png -

http://setiweb.ssl.berkeley.edu/beta/img/unread_locked.png -


http://setiweb.ssl.berkeley.edu/beta/img/boincstats_icon.png -

http://setiweb.ssl.berkeley.edu/beta/img/freedc_icon.png -

---

It looks like since after the Tuesday's outage this is missing inter alia [with BBC's url - img]:

[Windows, Firefox]
...

Hmmm, no problem on

[Linux, Firefox]

all looks normal. Maybe fixed in the meantime.

---

_\|/_
U r s

Yup all the badges are back and also the thread icons in the forum. Yay!

I was just looking at my inconclusives still hanging around from last night's tasks and I noticed that 2 of the 3 have their disagreements with the same wingman, host #61132. Looking at that host's task list, it appears that it's generating a high number of Invalids with opencl_atiapu_sah and opencl_ati5_nocal, although it's getting plenty of validations, too.

All SETI@home v8 tasks for computer 61132
	
State: All (856) · In progress (2) · Validation pending (42) · Validation inconclusive (79) · Valid (616) · Invalid (117) · Error (0)
Application: All (16031) · AstroPulse v7 (4245) · SETI@home v7 (10930) · SETI@home v8 (856) 

I have no idea if that's a host issue or an app issue, but might be worth looking into.

Yes, I went through that same process of checking the inconclusives on my host 12316 this morning, and found I had three against that same 61132. There's only one now, and no new errors except those caused by my own finger-fumble last night, so they must have validated. And if they were only weakly similar, they wouldn't be passed into a science database as canonical.

But it adds time to the testing processes, and does suggest that either the host or the application should be taken back into the shop for an overhaul.

I'm a bit surprised with opencl_nvidia_sah comparisons on my 2 computers.

Intel quad 2.9, 750TI
AMD dual 4200+, 750TI

While the app is a CPU hog (I'm running 1 task each, no cores reserved) my AMD is out performing my Intel, very marginally @ 08:30 vs 08:40 run times. I never thought I would see that happen.

EDIT: PS I'm running error free :D

EDIT2: With AP opencl_nvidia_100 my AMD has runtimes about 250% of my Intel. AMD (both cores free) Intel (no cores free)

Yes, I went through that same process of checking the inconclusives on my host 12316 this morning, and found I had three against that same 61132. There's only one now, and no new errors except those caused by my own finger-fumble last night, so they must have validated. And if they were only weakly similar, they wouldn't be passed into a science database as canonical.

But it adds time to the testing processes, and does suggest that either the host or the application should be taken back into the shop for an overhaul.

I've noticed that 61132's counts are almost always higher than his wingmen's on the Inconclusive/Invalid tasks. Even on tasks that get validated, his seem to be higher. A case in point is WU 7864192 where his task 21667347 shows:

Work Unit Info:
...............
Credit multiplier is :  2.85
WU true angle range is :  2.727001
Used GPU device parameters are:
	Number of compute units: 44
	Single buffer allocation size: 128MB
	Total device global memory: 3072MB
	max WG size: 256
	local mem type: Real
	LotOfMem path: no
period_iterations_num=50
Autocorr: peak=17.99226, time=73.82, delay=5.5154, d_freq=1421065842.88, chirp=18.826, fft_len=128k
Spike: peak=50.57866, time=6.711, d_freq=1421064118.68, chirp=-29.752, fft_len=128k
Autocorr: peak=92.54395, time=6.711, delay=0.83886, d_freq=1421064253.46, chirp=-29.752, fft_len=128k
Spike: peak=39.70731, time=33.55, d_freq=1421063595.71, chirp=-29.752, fft_len=128k
Autocorr: peak=61.7743, time=33.55, delay=3.7749, d_freq=1421063454.82, chirp=-29.752, fft_len=128k
Triplet: peak=8.401951, time=79.38, period=0.03277, d_freq=1421068651.33, chirp=45.2, fft_len=128 
Triplet: peak=7.507046, time=80.44, period=0.1638, d_freq=1421067514.04, chirp=96.858, fft_len=128 

Best spike: peak=50.57866, time=6.711, d_freq=1421064118.68, chirp=-29.752, fft_len=128k
Best autocorr: peak=92.54395, time=6.711, delay=0.83886, d_freq=1421064253.46, chirp=-29.752, fft_len=128k
Best gaussian: peak=0, mean=0, ChiSq=0, time=-2.122e+011, d_freq=0,
	score=-12, null_hyp=0, chirp=0, fft_len=0 
Best pulse: peak=2.144019, time=46.2, period=0.11, d_freq=1421061706.54, score=0.9031, chirp=0, fft_len=32 
Best triplet: peak=8.401951, time=79.38, period=0.03277, d_freq=1421068651.33, chirp=45.2, fft_len=128 


Flopcounter: 14674750258566.727000

Spike count:    2
Autocorr count: 3
Pulse count:    0
Triplet count:  2
Gaussian count: 0

where my task 21667348 has:

Work Unit Info:
...............
Credit multiplier is :  2.85
WU true angle range is :  2.727001
Used GPU device parameters are:
	Number of compute units: 7
	Single buffer allocation size: 128MB
	Total device global memory: 2048MB
	max WG size: 1024
	local mem type: Real
	FERMI path used: yes
	LotOfMem path: no
period_iterations_num=50
Autocorr: peak=17.99223, time=73.82, delay=5.5154, d_freq=1421065842.88, chirp=18.826, fft_len=128k
Triplet: peak=8.40195, time=79.38, period=0.03277, d_freq=1421068651.33, chirp=45.2, fft_len=128 
Triplet: peak=7.506961, time=80.44, period=0.1638, d_freq=1421067514.04, chirp=96.858, fft_len=128 

Best spike: peak=22.60275, time=100.7, d_freq=1421067464.97, chirp=-22.441, fft_len=128k
Best autocorr: peak=17.99223, time=73.82, delay=5.5154, d_freq=1421065842.88, chirp=18.826, fft_len=128k
Best gaussian: peak=0, mean=0, ChiSq=0, time=-2.122e+011, d_freq=0,
	score=-12, null_hyp=0, chirp=0, fft_len=0 
Best pulse: peak=2.144019, time=46.2, period=0.11, d_freq=1421061706.54, score=0.9031, chirp=0, fft_len=32 
Best triplet: peak=8.40195, time=79.38, period=0.03277, d_freq=1421068651.33, chirp=45.2, fft_len=128 


Flopcounter: 14674750258566.727000

Spike count:    0
Autocorr count: 1
Pulse count:    0
Triplet count:  2
Gaussian count: 0

In particular, his Spikes seem to be totally out of line with mine.

I imagine this one was initially Inconclusive until Mr. Kevvy's host came along and sided with mine, but then all 3 got validated. You're right, though, about 61132 not getting the canonical result, even though it reported first. That honor(?) went to my host.

Host has many computation error, invalids and sanity check abortions:
ERROR: Possible wrong computation state on GPU, host needs reboot or maintenance

With all 3 flavours of ATi app. Seems like unstable hardware.

In case it's software issue:

Name: Hawaii
Vendor: Advanced Micro Devices, Inc.
Driver version: 1800.8 (VM)
Version: OpenCL 1.2 AMD-APP (1800.8)

---

News about SETI opt app releases: https://twitter.com/Raistmer

I'll see if we have other Hawaii's

Here are a few.

http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=61132
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=61446
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=62217
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=63626
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=63970
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=73969
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=73998
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=75861
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=76155
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=76953
http://setiweb.ssl.berkeley.edu/beta/show_host_detail.php?hostid=77441

Some of them are running with GPU crunching turned off.

---