[quote]As to cache, the article I linked earlier discusses how the processing stops when the data being accessed is not in the cache and the system has to run out to memory to get it. And how this can impair your CPU by 80%. So, given that example, if you want more then half of your processor time doing useful work, then you need sufficient cache for the application's use of data.[/quoted]

cheers man


[quote]
Robust compilers for the programming language(s) of your application, and enough spare development hours to devote to the effort.[/quoted]

i was thinking more about theoretical posibility of development on GPU, some major problems. But you right, practicly you will need bucket of development time;)

Thanks a lot.

What about making rosetta an open source project and let someone else do the dirty job;)? There are always people with lots of spare time.
You dont need very deep understanding of protein prediction in order to translate code on a different platform. Core of the program with all the formulas can be "copied" in new language/platform.

I believe it wont be so hard to test program by using known results.
What you think?

What about making rosetta an open source project and let someone else do the dirty job;)? There are always people with lots of spare time.
You dont need very deep understanding of protein prediction in order to translate code on a different platform. Core of the program with all the formulas can be "copied" in new language/platform.

I believe it wont be so hard to test program by using known results.
What you think?

it is open source to people who request the code and can demonstrate that they can be of use. It's a huge program though - guys from AMD and Intel both looked at optimising it (search for Mats Petersson or Who?) - I think Who? optimised a version of SETI for SSE4, but Rosetta is a much bigger beast.

---

First, let me again thank DB for taking a few moments from his busy schedule to post and provide some feedback to the donors who are interested in this topic.

I note that the comment specifically addressed the MS xBox, and made no mention of the Sony PS3.

While imho I believe the PS3 to be a superior gaming console for purposes crunching for science, I also understand that there may be legitimate reasons why any potential app may be for the xBox. Rosetta and MS are both in Washington, and perhaps there are more/deeper personal/professional relationships with MS than Sony because of this. I also understand that the Gates charity, legally separate from MS, donated financial resources that were ultimately received by Rosetta.

That being said, and with the explicit intent that this not turn into a PS3 vs xBox trashing thread, I will note the following, which is yet just one more reason why developing an app for the PS3, which can then run on the Cell BE's of IBM's blade servers, should be a goal:

Toshiba Plans to Equip Multimedia Laptops with SpursEngine Processor

Toshiba plans to equip some of its multimedia-oriented mobile computers with its multimedia processor that can accelerate various applications, including graphics, physics, video and so on. Toshiba did not disclose how much its customers should pay for the part.

On the 8th of May Toshiba disclosed its growth strategies for investors for the year 2008, one of which was integration of the SpursEngine processor into advanced media center AV Qosmio notebooks. According to Toshiba, the Cell-derivative processor will upscale standard picture quality to high definition resolution. While the manufacturer did not disclose into details, it claimed that the notebook would be available in 2008.

It will depend on software support whether special-purpose accelerator from Toshiba brings any benefits, but recently the company announced that its chip is supported by such software companies as Corel and Cyberlink.

SpursEngine is a co-processor that integrates four of Cell high-performance RISC core SPEs, half the number of the full configuration, hardware dedicated to decoding and encoding of MPEG-2 and H.264 video, XDR memory interface as well as PCI Express interface. By combining the high level, real time processing software of the SPEs with the hardware video codecs, the SpursEngine realizes an optimized balance of processing flexibility and low power consumption. The prototype of SpursEngine operates at a clock frequency of 1.5GHz and consumes power at 10W to 20W.

A slightly stripped down version of the Cell BE, but when the cpu for the PS3 was originally speculated and rumored about, who would have thought that it would find its way into IBM blade servers, and now mainstream laptops?

In no way am I discouraging the Project from investigating the potential offered by the xBox.

What I am doing however, is encouraging them to take a second look at the Sony PS3 and the Cell BE.

In spite of potentially limited cache/ram (does this limitation also apply to the xBox? I honestly don't know enough about the xBox's architecture), perhaps there still could be some use made from the large and varied installed base of the Cell BE's.

3. I will need to re-read some threads and articles about gpu's, and the problems specifically related to nVidia (there were reasons why up until present that F@H only used ATI and not nVidia), before I comment futher, other than noting what F@H has already stated:

The GPU client is still the fastest, but it is the least flexible and can only run a very, very limited set of WUs. Thus, its points are not linearly proportional to the speed increase. The PS3 takes the middle ground between GPUs (extreme speed, but at limited types of WU's) and CPU's (less speed, but more flexibility in types of WUs).

So, gpu's may be very fast, but there are "very, very limited" instances where a dc project may be able to take advantage of that speed, and it may just be that the very nature of the Project does not allow it to have the "very, very limited" types of wu's which can benefit.

Folding@home does not use CUDA. They are trying to build a one app fits all for the GPU which currently is not possible due to their differences. I remember reading somewhere that until Nvidia support and return accurate results with this method they use they will not use Nvidia GPU's.

Folding@home's biggest limitation is the use of directx for their GPU clients. This causes the need for 1 core/cpu to be used to supply the GPU.

~BoB

---

3. I will need to re-read some threads and articles about gpu's, and the problems specifically related to nVidia (there were reasons why up until present that F@H only used ATI and not nVidia), before I comment futher, other than noting what F@H has already stated:

The GPU client is still the fastest, but it is the least flexible and can only run a very, very limited set of WUs. Thus, its points are not linearly proportional to the speed increase. The PS3 takes the middle ground between GPUs (extreme speed, but at limited types of WU's) and CPU's (less speed, but more flexibility in types of WUs).

So, gpu's may be very fast, but there are "very, very limited" instances where a dc project may be able to take advantage of that speed, and it may just be that the very nature of the Project does not allow it to have the "very, very limited" types of wu's which can benefit.

Folding@home does not use CUDA. They are trying to build a one app fits all for the GPU which currently is not possible due to their differences. I remember reading somewhere that until Nvidia support and return accurate results with this method they use they will not use Nvidia GPU's.

Folding@home's biggest limitation is the use of directx for their GPU clients. This causes the need for 1 core/cpu to be used to supply the GPU.

~BoB

F@H has just announced support for Nvidia GPU's utilising CUDA wtih the new GPU2 Client and should be beta testing within a few weeks.

-Lucus

---

F@H has just announced support for Nvidia GPU's utilising CUDA wtih the new GPU2 Client and should be beta testing within a few weeks.

-Lucus

WHAT?!?!!? Wow thats good to hear!

Finally I'll be able to use my GPU for something other than graphics/games!

~BoB

---

F@H has just announced support for Nvidia GPU's utilising CUDA wtih the new GPU2 Client and should be beta testing within a few weeks.

-Lucus

WHAT?!?!!? Wow thats good to hear!

Finally I'll be able to use my GPU for something other than graphics/games!

~BoB

YES that is sad.... for Rosetta, Already one of my core in q6600 goes to folding@home for a GPU Client.... so Rosie has 4 cores now 3.....

If rosetta were to support CUDA, buying this 1 terraflop beast would be in the realm of possibility:

http://www.amazon.com/TESLA-C1060-COMPUTE-BOARD-RETAIL/dp/B001H5V6LO/ref=pd_bbs_sr_1?ie=UTF8&s=electronics&qid=1230525736&sr=8-1

http://www.amazon.com/TESLA-C1060-COMPUTE-BOARD-RETAIL/dp/B001H5V6LO/ref=pd_bbs_sr_1?ie=UTF8&s=electronics&qid=1230525736&sr=8-1

If rosetta were to support CUDA, buying this 1 terraflop beast would be in the realm of possibility:

http://www.amazon.com/TESLA-C1060-COMPUTE-BOARD-RETAIL/dp/B001H5V6LO/ref=pd_bbs_sr_1?ie=UTF8&s=electronics&qid=1230525736&sr=8-1

Just a note, CUDA is alive and well and operational for SaH and GPU Grid I have run both and have had successful work completed.

Cautionary note, this is new technology and you will need a later version of BOINC (I am using 6.5.0 which is "beta"), a CUDA capable card, and the latest drivers from Nvidia.

On my 9800 GT SaH tasks take about 9 miniutes each to run ...

On GPU Grid the run time is variable but the norm is about 17 hours on the 9800 GT ... later this week I will be able to tell you what it takes on a faster card ...

There have been problems getting work on both projects on GPU Grid because of server issues and on SaH because they flat ran out of work and on other occasions the servers have gotten behind.

Again, this is early days for CUDA, expect problems ...

Several of the identified problems are in design resolution though I am not that convinced that the proposed solution(s) will serve us well ...

Before you leap you may want to look at the discussions in NC at SaH and the Nvida forum in GPU Grid. As I understand the science GPU Grid is doing something akin to what Rosetta and the others are doing with just a different approach. If someone else knows better feel free to correct me ...

I am over there to play with the technology and learn while waiting for projects that I have major interests in releases an GPU client...

---

I noticed that nvidia is selling a 1U rack mount server that packs 4 terraflops of computing power, and iirc sells for around $8000 each. I suppose this may be a better investment than purchasing many "traditional" servers. you need CUDA enabled applications to use it.

http://www.nvidia.com/object/product_tesla_s1070_us.html