Posts by Thamir Ghaslan
log in
1) Message boards : Number crunching : Combining extremely different (CUDA) GPUs? (Message 8320)
Posted 2780 days ago by Thamir Ghaslan
Sounds more like a feature addition that boinc developers can work on.

Apart from disabling specific GPUs when launching boinc which is already implemented, and really not sure if you can disable a specific GPU when "while computer is in use", I'm not aware of any function to specifically assign a GPU to a specific project.

You could try your luck in http://boinc.berkeley.edu/dev/ message boards.

2) Message boards : Number crunching : Optimized Applications (Message 4988)
Posted 2940 days ago by Thamir Ghaslan
i have a problem with the new 2.07 application, my machine: i920, GPU0 = 5870, GPU1 = 4890. (Win7-64)

All WU's on GPU1 crashing with:
<message>Unzulässige Funktion. (0x1) - exit code 1 (0x1)</message>
(illegal function call)

WU's on GPU0 are running well.
[..]
Boinc reports:
ATI GPU 0: ATI Radeon HD5800 series (Cypress) (CAL version 1.4.515, 1024MB, 2720 GFLOPS peak)
ATI GPU 1: ATI Radeon HD 4700/4800 (RV740/RV770) (CAL version 1.4.515, 1024MB, 1360 GFLOPS peak)

but if i look at my "Computer Details" for ID 3461, i see:
Coprocessors: [2] ATI ATI Radeon HD5800 series (Cypress) (1024MB) driver: 1.4.515
what is wrong.

So i assume, that the application tries to run the 4890 with the code for 5xxx cards?!

Yes, that appears to be the case. The application isn't able to handle such mixed configurations yet. Guess I have to build a new version which takes this into account. Fortunately, in this case it can be handled by the application itself (there is a similar issue over at MW when mixing cards capable of DP calculations and non-capable ones). But basically it points to an underlying general flaw of the GPU handling by BOINC.
That is one of the reasons it is good to have a larger beta test of new applications as such issues never come up on the development systems.

But even as this issue is easy to fix, I doubt a bit the new version will be ready today ;). I hope you can wait for the new year!


Heh!

There goes my grand scheme to mix my current 4870x2 with a 5970!

Not even sure how it'll behave if I crossfire, though not a fan.

Possible boost to dx9, 10, 11, with or without tessellation?

Contrary to popular belief, 48xx support dx11 with a few missing features, as far as my test run with uniengine heaven went.

I'm confident you'll be able to resolve it.
3) Message boards : Number crunching : Optimized Applications (Message 4876)
Posted 2941 days ago by Thamir Ghaslan
One of the local shops in Saudi have a 5970 for sale.

For 800$ !

This is almost a 300$ premium over US sales.

Online shopping shows always out of stock, and even if they get it, international shipment restriction applies.

This whole 40 nm fabrication troubles affecting ATI and Nvidia might get cleared up by q2 2010!

Bottom line, as hesitant as I am, I might grab one in the next few days before it disappears!
4) Message boards : Number crunching : ATI 5970 (Message 3752)
Posted 2983 days ago by Thamir Ghaslan
I've postponed buying any 5800 card until an x2 solution comes out!

And here it is, which is earlier than I expected!

http://www.tomshardware.com/reviews/radeon-hd-5970,2474.html

Any scapegoats who tried it or will try it? :P

Word has it that some 5800 owners are disappointed!
5) Message boards : Number crunching : Rising pending (Message 3501)
Posted 2992 days ago by Thamir Ghaslan
for some reason i cant edit! but adding to the previous post, i notice an occasional message from the server stating that it did'nt need my uploaded results because they were already submitted!?
6) Message boards : Number crunching : Rising pending (Message 3500)
Posted 2992 days ago by Thamir Ghaslan
maybe i'm one of the lucky few, but I've always had 1800 - 2000 worth of pending wether my dsl line gets disconnected for a few days thanks to an unreliable phone company, or wether collatz server hickups!

7) Message boards : Number crunching : Proposed new BOINC credit system (Message 3392)
Posted 2997 days ago by Thamir Ghaslan


once again..... people are leaving Seti@home for a reason,, lol and it

aint the credits.

its all summed up in 1 word

DICTATORSHIP!!!


I've left it years ago because I think searching for alien porn is silly!
8) Message boards : Number crunching : Proposed new BOINC credit system (Message 3391)
Posted 2997 days ago by Thamir Ghaslan


DA is a computer scientist...period! The credit system in BOINC (or any other DC project) is a fundamentally social/economic entity. The only way to design a reasonable "currency" and update that "currency" to match technological innovation is to get the input of someone with expertise in currency systems!


Does this mean we are on a verge of a global credit crisis melt down?

I'm glad we can't borrow credit or borrow from the borrower!
9) Message boards : Number crunching : Increased crunchtime (Message 3386)
Posted 2997 days ago by Thamir Ghaslan

So... why did they take longer for the same credit? You happened to get a WU that isn't "GPU friendly". The GPUs process in parallel. Anything not capable of being processed in parallel leads to wait states where one thread has to wait for the others to catch up.


Thanks for the explanation, for a few days I've noticed that one task takes 10 minutes, and the next task with an increment of one takes 15 minutes on my 4870x2.

Not complaining either, but I thought it was due my driver or card!
10) Message boards : Number crunching : credits are going down (Message 3357)
Posted 2997 days ago by Thamir Ghaslan


The next thing is that over at SAH, there's an obscure "script" involved that is responsible for "adjusting" credits. Part of the formula is the time taken by the app. Unfortunately (for us), this script is looking at the "CPU time" instead of the GPU time that a WU takes...


I've always wondered, since when did SETI become the overlords of boinc?

does'nt make sense for a decentralized concept either in projects, resources, or administration!
11) Message boards : Number crunching : credits are going down (Message 3313)
Posted 3001 days ago by Thamir Ghaslan
And here is a possible new entry for the 500 list, 5000 intel cpus and 5000 ati GPU, China's 1 peta flop beast to compete with road runner.

http://www.tomshardware.com/news/China-Supercomputer-Tianhe-Petaflop-Intel,8966.html
12) Message boards : Number crunching : credits are going down (Message 3311)
Posted 3001 days ago by Thamir Ghaslan
road runner, the fastest computer in the world that broke the peta flop barrier, is made up largely of modified cell processors, aka Play Station 3s.

As far as I'm concenred, its one big GPU with great performance in terms of flops and flops per watts.

Top500 never whined or came up with flops adjusments and it runs scientific computations just as its non cell counter parts.

to hell with seti.


13) Message boards : Number crunching : 128 bit windows 7 & 8! (Message 3128)
Posted 3007 days ago by Thamir Ghaslan
Guys.. this is really nothing to get excited about. To quote from the Inq article:


Thanks for the lengthy clarification and I could not agree with you more!

Problem is I couldn't explain it as eloquently as you've put it!
14) Message boards : Number crunching : 128 bit windows 7 & 8! (Message 2990)
Posted 3012 days ago by Thamir Ghaslan

x64 is (!) useful, but not for all tasks may be, and keep in mind that some of programmers are lasy enough to optimize their code for x64 and debug it. Remember, that long time ago multicore CPUs were slower than single-core ones - the case was the same - lack of optimization for parallel data processing. So without changes in code and compiling it wi wil no get any benefits even of 256-bit CPUs.


"128bit" is kind of a broad statement. It could be memory bus, data bus, such as the ones many modern GPUs use, or it could be used to address memory sizes with 40 zeros infront of it, or it could be used to calculate quadrouple precesion floating points.

Bottom line, More's Law currently seems to favor how many cores you can add and how small you get the die size, and not how many bits you can add.
15) Message boards : Number crunching : 128 bit windows 7 & 8! (Message 2763)
Posted 3020 days ago by Thamir Ghaslan


I can only thing of two contexts in which the announcement could make sense: The first is if the actual effort is to clean the code base so that in a few decades the transition to x128 will be less painful than the x64 one was. I don't find this credible because even if Midori doesn't replace windows the normal rewrite cycles will end up junking a good chunk of the codebase anyway.

The second is if some of the 128bit SSE instructions are being used in windows to accelerate FPU math involved in video playback, or the like.

EDIT: hit post by accident before finishing.


There are speculations that this is Larabee! Maybe a 128 GPU while the rest of the CPU is 64?
16) Message boards : Number crunching : 128 bit windows 7 & 8! (Message 2762)
Posted 3021 days ago by Thamir Ghaslan

"any significant performace"? Have you tried Collatz with both 32 bit and 64 bit? Collatz is twice as fast with 64 bit operating systems because it uses all 64 bit integers for performance.


Not not really. Maybe I'm a bit outdated, but I did try 32 and 64 bit boinc projects a year or two back. On 64 bit Linux but cant remember all the non truly optimized 64 bit projects. It was unanimously concluded back then that full memory gains would be the first obvious benefit as opposed to performance gains.
17) Message boards : Number crunching : 128 bit windows 7 & 8! (Message 2732)
Posted 3021 days ago by Thamir Ghaslan

Rummor mill is increasing to the point of certainty! Just google it! It all started with a microsoft employee leaking info in his linked in profile, something in conjuction with Intel's future 128 bit CPU.

Question is, we have'nt seen any significant performance increase between 32 and 64 bit code, and, the world is still at 32 bit and 64 is'nt wide spread, so 128 bits should be a wait and see, like Nvidias next GPU. :P



18) Message boards : Number crunching : NVIDIA vs ATI (Message 2720)
Posted 3021 days ago by Thamir Ghaslan

So how fast is a ATI 5870?
My 920i7 with a 4870x2 is doing 2 WUs every 9.20mins with credit av of 560 approx

So a 5870 single card would have to do a WU in 50% of that time to keep up.

What is the results out there at the momenent?

Do 5870s do 2 WUs per cycle?

Ross


I've seen early benchmarks with Collatz, milkyway, and a few games that set the standards for frames per second benchmarks.

a 5870 is equal to or slighlty faster (10 - 20% lead) than a 4870x2.

Which is preciesly why I'm waiting for a 5870x2 before I jump from my current 4870x2.
19) Message boards : Number crunching : NVIDIA vs ATI (Message 2709)
Posted 3022 days ago by Thamir Ghaslan
We will know more soon, but if it will be soon enough to keep AMD from gaining a lead in the enthusiast market has yet to be seen.


I've switched between ATI and Nvidia over the past few years I've lost track of all the model names. :P

The criteria was obviously which board gives me the most bang for the buck?

There was a time when most enthusiasts picked AMD over Intel, until Intel changed the playing field by introducing dual cores some 3 to 4 years ago!

Now everyimte AMD plays catchup Intel goes a step further.

As concluded, if ATI will dominate and continue to dominate everytime Nvidia plays catchup will remain to be seen!

Love the competition and would'nt mind seeing more than 2 competitors! :)
20) Message boards : Number crunching : NVIDIA vs ATI (Message 2704)
Posted 3022 days ago by Thamir Ghaslan
Interesting read.

http://www.pcper.com/article.php?aid=791&type=expert


Next 20

Main page · Your account · Message boards


Copyright © 2018 Jon Sonntag; All rights reserved.