With AVX, speedup for AP6 is around 15%, so it will pay better, tho.
I measured 4 hours per WU with AVX against 4:30 hours without AVX. Since there is no AVX MB application, AP6 worths more for the moment, I think.

---

Note, XP 32 (or any 32 bit OS) should have no problem with 4GB RAM, 4GB is it's max IIRC. It will not recognize MORE than 4GB.

I think we are talking at cross-purposes.

You said, "I just swapped a really old celeron, and 1G ram, for that p4 and 4 gigs. Let's just not tell the wife I'm gonna be buying some parts..."

To which I replied that SETI@Home on a 32-bit OS using less than 4GB of RAM on a P4 is just fine (and here's the important bit) at pushing a CUDA capable card to a usefully high RAC.

I just didn't want that last part to be lost in some form of misunderstanding.

We were talking about the superiority of GPU crunching to CPU crunching, low-power crunching, and low-cost crunching. My point was that a relatively inexpensive video card crunching on a system you already own will improve your number-crunching per dollar and watt far, far more than new motherboards, RAM, CPUs and all the implied expense that goes with that.

But as others have warned you, once you start down that road, it is really hard to know when to say when.

I "grew" two more computers recently because I was involved in something else. That lead to three more video cards than I had even imagined I would purchase. That lead to getting close to a "RAC" number that my warped brain told me I just HAD to achieve, which had me rationalize about power-consumption and heat, which lead to two expensive video cards being purchased, which lead to another computer's being taken out of mothballs, which lead to another video card purchase rationalizing that the computer was "wasted" unless I used it.

I can easily see how, if they had the money, a person could easily find themselves with several $10,000+ computers. In fact, money-no-object computers could get close to $20,000.

I was able to leave my setups alone until things changed and what was easily possible changed.

I think I've reached another plateau and can stop being OCD about "just needing to fill this empty PCIe slot." I think I have. That is, I hope I have. Which is to say, I should probably stop for a while.

Good luck with whatever you choose to do. The more we crunch, the faster the data is sifted, the sooner we learn if we've heard something from way out there somewhere.

I'm all in favor of each of us crunching as much he/she can as our personal circumstances allow.

I understand what you are saying. And I totally agree a video card is simply the cheapest way to add value to that p4 box.

But I could (and will) also upgrade to a cheap 64bit processor, just because the stuff I run for myself (desktop applications etc.) on that machine runs like garbage. ;-)

Processor upgrade @70$, not bad

(funny this little system was upgraded one thing at a time through the years, and is now containing none of it's original hardware- not even it's original case, but that new-ish mainboard is the reason I can do a 64 bit proc, so I'm thankful the system can still be brought (somewhat) up to date with yet another single part upgrade)




For the crunching:
What's the best pci-e video card I can get at or under 100$?

---

#resist

I "grew" two more computers recently because I was involved in something else. That lead to three more video cards than I had even imagined I would purchase. That lead to getting close to a "RAC" number that my warped brain told me I just HAD to achieve, which had me rationalize about power-consumption and heat, which lead to two expensive video cards being purchased, which lead to another computer's being taken out of mothballs, which lead to another video card purchase rationalizing that the computer was "wasted" unless I used it.

So its not just me then, they do talk about `green computing` :¬)

Why should you're CPU only 'adventure', be worthless in the first place?!
It 'sounds' negative , while it isn't.
Any valid contribution is welcome and valued.
I started with a P4, whithout HT, sold it and bought a second hand P4Dual and
later I added an GTX9800GTX+.
The beginning of CUDA, where a lot of people, better 'coders', spend a lot
of time to get it working. (Sorry for not mentioning their familiar names!)
Also started using my 'work pc', a C2Q6600 with an 8500GT, which I swapped with
a GTX250.
Then I started to build 2 rigs, used 2 ASUS P5E mobo's, Coolermaster Case, 650Watt PSU, 1st got an Q6600 and a GTX470. I bought a GTX480, too,
payed €300 for both cards, if I provided the CUDA possebillities and helped him
setup a similar system, but used for something different.
I also could buy an QX9650 for €350,00 Two 'test-samples', 1 for me,1 for the shop-owner, interrested for OC possebillities and custom build systems
(€900,00 at that time, a piece!) This was 3 years ago..................

A year ago I did put a Core i7-2600 and 2 (ASUS)EAH 5870 GPU, on an INTEL DP67BG, together after trying an ATI 5770. This was the start of OpenCL crunching, Raistmer and others, did put quite
a lot of time in trying to produce reliable apps.(Alex Kan, Josef Segur, Richard Haselgrove, Jason Gee, JD Whale, Raistmer and others).
CUDA was matured and FERMI capable. OpenCL had yet to prove, it's usefullness
in SETI MB and AstroPulse crunching, in which he, with help of other 'coders',
active on these forums, succeeded. Other projects followed, Collatz C., MW, GPUgrid, WCG, Einstein, etc..

Tests were done at SETI Bêta and still are.
I like the view of (high) credits, but valid results is my primary concern and
goal.

And concerning GPUs, better 1 GPU with 15 CUDA-Cores, then 2 with 7,
or using nVidas 200 series GPUs (Non-FERMI). Due to too many faults,found
triplets in a row, etc., I swapped them, 1 at a time for 400 series (FERMI)
GPUs. The same for ATI GPUs, I smoked out a 4850 GPU, GT8800 and a GTX9800.
(Sounding like a low flying Star-Fighter, when fully loaded and clearly not
made for 24 x 7 x 365 use loaded > 90%)
But I'm still trying out cmd-line-settings on my OpenCL host, trying to find a balance between CPU and GPU times and GPU-load, whithout
creating a lag, due to wrong (ATI)GPU settings.

If I had more money to spend on this, I'd probably build a house of computers
and 'live', in between ;-)
Electronics and computers always made my curiosity grow bigger then my purse
allowed, there's also something like having a life/wife/children etc.

I'll better stop, before I write a paper back............. ;-)

---

Lol. Spoken like a true cruncher.


And my original statement of worthless should be redacted. ;-)

I'm now doing somewhere around 10k/day with no GPU. This I can no longer consider worthless. (EDIT: maybe more like 8k/day, hard to tell with a building cache of "awaiting validation", but I've earned about 20,000 in the past 2 days)


I too like you, would have a house of computers if I had but the money. The amazing part is how powerful a modern machine it compared to what I grew up with. Even small machines can do a whole lot. I would spend all of my time with my computers. (Oh wait I do that too much already)

(on that note I still never optimized my windows laptop.. Off to figure out how to install optimized apps in windows. (side note from a Linux geek: as I do more with Linux, it keeps getting harder to do things in windows, and much easier to do them in Linux)


:-D

Happy crunching.

---

#resist

(on that note I still never optimized my windows laptop.. Off to figure out how to install optimized apps in windows. (side note from a Linux geek: as I do more with Linux, it keeps getting harder to do things in windows, and much easier to do them in Linux)


:-D

Happy crunching.
____________
-Dave (#2)

I'd wish, being more familiar to work with a LINUX build, already having a hard
time trying to 'understand' CUDA & OpenCL (BROOK+)!

By Trial and error, I do understand a bit more, being 60+ and having surgery to many times, missing an eye, takes his toll, too............

---

This might interest you:


http://www.newegg.com/Product/Product.aspx?Item=N82E16814150612

XFX FX-775A-ZNP4 Radeon HD 7750 Core Edition 1GB 128-bit GDDR5 PCI Express 3.0 x16 HDCP Ready Video Card $109.99

edit: And now I see this
http://www.tomshardware.com/reviews/geforce-gt-640-review,3214.html

Yea I gotta say, I wont be using any ATI/AMD GPU here. Only Nvidia. I'm sure everyone will concur.

---

#resist

umm, no.

Yea I gotta say, I wont be using any ATI/AMD GPU here. Only Nvidia. I'm sure everyone will concur.

Past performance is no guarantee of what is yet to come....

                                                                   Joe

I'll remember that then and not limit myself.

---

#resist

Yea I gotta say, I wont be using any ATI/AMD GPU here. Only Nvidia. I'm sure everyone will concur.

I concur.

The first 90 "top computers" are all NVIDIA cards. There's not a single ATI/AMD card among them until you get to computer #91.

That #91 computer is mine. That ATI card is a 6670 I have doing AP tasks because it is better at AP tasks than the CPU is and because there is no released CUDA AP application.

In the top 100 computers the *only* ATI card is mine and it is in a three video-card machine with the other two being GTX 670s.

The next computer with an ATI/AMD card in it is at #188 and it, too, has an NVIDIA card in it, also.

The next AMD card to show-up is at #208, another of my computers, two NVIDIA cards and a Radeon also. Again, the ATI is crunching AP-only and doing it in half the time of an i7-880 3GHz.

So...yeah, unless you are going to crunch AP-only, I concur: NVIDIA.

Am I saying that NVIDIA will always trump AMD? No. I'm just saying that with the exception of the AstroPulse application, the NVIDIAs tend to trounce the Radeons.

I have no idea how a GTX 550Ti will crunch compared to the new GTX 640, and I have no idea how a GTX 550Ti will crunch compared to a HD 6670 or an HD 6770 (or the superior 57xx mentioned in the Tom's Hardware discussion of the GTX 640). I do intend to get experience with that, though.

An aside: As far back as I can remember, I don't remember my ASUS GT 240 being anything but the most reliable cruncher I have ever owned. I honestly don't ever remember its ever, ever, throwing an error, ever. I hope the GTX 670s will do as well. The 560Ti cards have not been the most reliable things in the world.

I'm just saying that with the exception of the AstroPulse application, the NVIDIAs tend to trounce the Radeons.

Are there any head to head statistics that actually show this? I'm interested to see what the real performance difference is. I've always assumed that there aren't many ATI cards being used in the top computers because they aren't accessible to the project without the use of optimized apps. I haven't seen any performance based statistics that show NVIDIA cards crunch faster than ATI cards so I would like to see if there are actual numbers that back this up.

---

I'm just saying that with the exception of the AstroPulse application, the NVIDIAs tend to trounce the Radeons.

Are there any head to head statistics that actually show this? I'm interested to see what the real performance difference is. I've always assumed that there aren't many ATI cards being used in the top computers because they aren't accessible to the project without the use of optimized apps. I haven't seen any performance based statistics that show NVIDIA cards crunch faster than ATI cards so I would like to see if there are actual numbers that back this up.

This is a test run of the v7 apps that are being worked on.

v7 run on the FX-4100 against the GTX460 and HD7750.

------------
Quick timetable

WU : PG0009_v7.wu
JG_setiathome_6.98_windows_intelx86.exe -verb -nog :
Elapsed 888.546 secs
CPU 869.004 secs
setiathome_6.98p03_windows_intelx86.exe -verb -nog :
Elapsed 953.890 secs
CPU 931.654 secs
Lunatics_x41x_win32_cuda42.exe -verb -nog :
Elapsed 189.030 secs, speedup: 80.18% ratio: 5.05
CPU 9.656 secs, speedup: 98.96% ratio: 96.48
MB7_win_x86_SSE3_OpenCL_ATi_r1218.exe -period_iterations_num 20 -instances_per_device 1 -sbs 128 :
Elapsed 168.023 secs, speedup: 82.39% ratio: 5.68
CPU 19.906 secs, speedup: 97.86% ratio: 46.80

WU : PG0395_v7.wu
JG_setiathome_6.98_windows_intelx86.exe -verb -nog :
Elapsed 1064.990 secs
CPU 1046.408 secs
setiathome_6.98p03_windows_intelx86.exe -verb -nog :
Elapsed 1079.015 secs
CPU 1061.868 secs
Lunatics_x41x_win32_cuda42.exe -verb -nog :
Elapsed 57.151 secs, speedup: 94.70% ratio: 18.88
CPU 9.516 secs, speedup: 99.10% ratio: 111.59
MB7_win_x86_SSE3_OpenCL_ATi_r1218.exe -period_iterations_num 20 -instances_per_device 1 -sbs 128 :
Elapsed 124.381 secs, speedup: 88.47% ratio: 8.68
CPU 15.975 secs, speedup: 98.50% ratio: 66.47

WU : PG0444_v7.wu
JG_setiathome_6.98_windows_intelx86.exe -verb -nog :
Elapsed 892.774 secs
CPU 879.222 secs
setiathome_6.98p03_windows_intelx86.exe -verb -nog :
Elapsed 923.222 secs
CPU 909.891 secs
Lunatics_x41x_win32_cuda42.exe -verb -nog :
Elapsed 50.337 secs, speedup: 94.55% ratio: 18.34
CPU 8.502 secs, speedup: 99.07% ratio: 107.02
MB7_win_x86_SSE3_OpenCL_ATi_r1218.exe -period_iterations_num 20 -instances_per_device 1 -sbs 128 :
Elapsed 112.813 secs, speedup: 87.78% ratio: 8.18
CPU 16.193 secs, speedup: 98.22% ratio: 56.19

WU : PG1327_v7.wu
JG_setiathome_6.98_windows_intelx86.exe -verb -nog :
Elapsed 774.325 secs
CPU 762.548 secs
setiathome_6.98p03_windows_intelx86.exe -verb -nog :
Elapsed 761.987 secs
CPU 755.700 secs
Lunatics_x41x_win32_cuda42.exe -verb -nog :
Elapsed 69.916 secs, speedup: 90.82% ratio: 10.90
CPU 14.492 secs, speedup: 98.08% ratio: 52.15
MB7_win_x86_SSE3_OpenCL_ATi_r1218.exe -period_iterations_num 20 -instances_per_device 1 -sbs 128 :
Elapsed 145.215 secs, speedup: 80.94% ratio: 5.25
CPU 11.263 secs, speedup: 98.51% ratio: 67.10

---

I'm just saying that with the exception of the AstroPulse application, the NVIDIAs tend to trounce the Radeons.

Are there any head to head statistics that actually show this? I'm interested to see what the real performance difference is. I've always assumed that there aren't many ATI cards being used in the top computers because they aren't accessible to the project without the use of optimized apps. I haven't seen any performance based statistics that show NVIDIA cards crunch faster than ATI cards so I would like to see if there are actual numbers that back this up.

You know, it's really hard to point-to facts that support my assertion. Since I can't make the same work unit crunch locally on multiple cards, I'm always at the mercy of accidentally decent-looking comparisons and that's a problem on a lot of levels.

Oh... that's clear as mud, isn't it?

For clarity: When comparing the times of the WUs done by the 6770 and 6670 with the same WU reported by NVIDIA cards, while still faster than the same work unit done on a CPU, I was sufficiently disappointed that I confined the ATI cards to AP tasks; that being their highest utility.

On non-AP tasks the 6450 was not-good at all, and waiting for the 3450 to finish a work unit was just marginally better than falling into the gorge of eternal peril and waiting for the fall to abruptly stop (which it never seems-to, the peril being eternal).

I think it is a fair bet that most of the top 100 hosts are running optimized applications and so while you have a point about the power of really huge numbers and who might make-up the average users, I don't believe that explains this specific imbalance.

I'm just saying that with the exception of the AstroPulse application, the NVIDIAs tend to trounce the Radeons.

Are there any head to head statistics that actually show this? I'm interested to see what the real performance difference is. I've always assumed that there aren't many ATI cards being used in the top computers because they aren't accessible to the project without the use of optimized apps. I haven't seen any performance based statistics that show NVIDIA cards crunch faster than ATI cards so I would like to see if there are actual numbers that back this up.

This is a test run of the v7 apps that are being worked on.

v7 run on the FX-4100 against the GTX460 and HD7750.

------------
Quick timetable

(lots of numbers)

Nice! Those numbers just make me want more hah. It seems the NVIDIA card finishes the tasks at about twice the speed of the ATI/AMD card and actually slower on one of the tasks (hmmm). Online research shows the GTX460 costing more and the HD7750 uses almost 1/3 of the energy? It just makes me want more numbers from more cards.

I'm curious as to how you got those numbers and if there is an easy way to run benchmarks on specific WU's then why isn't there a ton of people running them so we can get more info as to how different cards perform? It would help greatly for people looking to make purchases based on crunching performance.

---

Thanks for that Arkayn. (Nice to see you back around here)

---

#resist

I'm just saying that with the exception of the AstroPulse application, the NVIDIAs tend to trounce the Radeons.

Are there any head to head statistics that actually show this? I'm interested to see what the real performance difference is. I've always assumed that there aren't many ATI cards being used in the top computers because they aren't accessible to the project without the use of optimized apps. I haven't seen any performance based statistics that show NVIDIA cards crunch faster than ATI cards so I would like to see if there are actual numbers that back this up.

This is a test run of the v7 apps that are being worked on.

v7 run on the FX-4100 against the GTX460 and HD7750.

------------
Quick timetable

(lots of numbers)

Nice! Those numbers just make me want more hah. It seems the NVIDIA card finishes the tasks at about twice the speed of the ATI/AMD card and actually slower on one of the tasks (hmmm). Online research shows the GTX460 costing more and the HD7750 uses almost 1/3 of the energy? It just makes me want more numbers from more cards.

I'm curious as to how you got those numbers and if there is an easy way to run benchmarks on specific WU's then why isn't there a ton of people running them so we can get more info as to how different cards perform? It would help greatly for people looking to make purchases based on crunching performance.

The task that the HD7750 does better on is a vlar.

Lunatics has a benchmark program that we use to test out new apps.

Thanks for that Arkayn. (Nice to see you back around here)

I am almost always here, I just don't post a whole lot.

---

I am almost always here, I just don't post a whole lot.

Maybe not, but we're almost always enlightened when you do post.

Thanks for your effort.