Nekochan Net

I suspect I might have asked before, but it didn't come up in search.

I have a Taligent/Sun TAAC-1 board. It's a stack of 9-U VME boards that has a VLIW processor and framebuffer on it and you could offload graphics or processing applications to it if you wrote the software for it. So far I don't have any of the docs or dev stuff - has anyone here seen any of them?

Second question - it hasn't been very useful so far, do you think someone like the CHM might be interested? It sort of had the same ideas as VLIW and GPGPU if you stretch a bit...

_________________
Damn the torpedoes, full speed ahead!

Systems available for remote access on request.

(single-CM)

From what I've read this product never worked as advertised and was recalled from service. The one you have is likely to be passing rare.

_________________

_________________
"Was it a dream where you see yourself standing in sort of sun-god robes on a
pyramid with thousand naked women screaming and throwing little pickles at you?"

Yep, one of them. Neat idea, but I don't have the dev software, docs, or a justification to spend a whole bunch of time figuring out how to use something that was discontinued in 1988. Also, my host box is a cobbled together Sun-4/330, which takes forever to boot and might be faster than the TAAC. That's why I was a bit interested in the CHM, but then again I'm not sure if I want to spend lots of money mailing it down to Mountain View.

Makes me wonder why SGI didn't pick up on this and make some of their graphics hardware accessible for other tasks (the TAAC could be used for graphics or the data could apparently be exported back out somehow for other use).

_________________
Damn the torpedoes, full speed ahead!

Systems available for remote access on request.

(single-CM)

There was an old paper on using the VGX boardset for general purpose fp computations. I can't seem to find it though, it's waaaay before my time.

Apparently one of the main problems with the SGI design was that I/O was extremely asymmetrical, i.e. the system could pump a lot of data into the graphics pipe, but the gfx subsystem really was not designed to return a lot of information back to the host. And why would it, as a graphics processor its primary target was the display generator not the host.

_________________
"Was it a dream where you see yourself standing in sort of sun-god robes on a
pyramid with thousand naked women screaming and throwing little pickles at you?"

This sounds similar to the problem that the folks over at NCSA were having when they were trying to use the Sony Playstation 2 for general purpose computing. They could get it to crunch a ton of data, but getting the results back out was very problematic. At least, this is what I heard when I was chatting with them a while back when I still worked for U of I

_________________
And gobs of other hardware...

That would have been exciting to have an OpenCL like API in the early 90s for SGI, that was something the PC world didn't get at least in a generic form until 2008.

_________________
33mhz R3k/48mb/XS24 150mhz R4400/256mb/XL24 600mhz R14kA/2gb/V10 8x1.4ghz Itanium 2/8GB 32x600mhz R14kA/24GB 4x700mhz R16k/8GB/V12/DCD/SAS/FC/DM5 2x700mhz R16k/4GB

I don't think SGI ever offered anything similar to what one can do now with OpenCL or CUDA.


There were some products that exposed the compute capabilities of the GFX pipeline (via OpenGL extensions) to accelerate image processing algorithms, but that is about it. SGI exposed as little as possible of the GFX to the programmer, mainly because that was the whole point behind GL/OpenGL. Those 3D subsystems were too unbalanced to make it worth the while anyway.

_________________
"Was it a dream where you see yourself standing in sort of sun-god robes on a
pyramid with thousand naked women screaming and throwing little pickles at you?"

I heard a similar story about CERN from someone visiting our campus at the time that they had the compute power but not the I/O bandwidth.



I did a little post grad work in the 1990s and I had a 8 CPU power challenge at my disposal in the mid 1990s and my friend had a Cray(*). GPU computing wasn't a concept back then... I had PVM and MPI on SGIs and 486s and Pentiums in the early to mid 1990s. pThreads, PVM and MPI was what SGI was pushing back then, at least on our campus.

Bare in mind that the set of (SIMD) problems solvable by CUDA is a subset of the set of problems solvable by a MIMD approach.

I guess the alternative back then was transputers and OCCAM but they blew up the power supply on ours before I got the chance to play with our set up.

R.

* The Cray stuff was SIMD but you stuck pragmas around loops to tell the compiler to try to vectorize this section of code. It didn't always work and wasn't always easy to code for.

_________________
死の神はりんごだけ食べる

アレゲはアレゲ以上のなにものでもなさげ -- アレゲ研究家


J5600, 2 x Mac, 3 x SUN, Alpha DS20E, Alpha 800 5/550, 3 x RS/6000, Amiga 4000 VideoToaster, Amiga4000 -030, 733MHz Sam440 AmigaOS 4.1 update 1. Tandem Himalaya S-Series Nonstop S72000 ServerNet.

Sold:

Cortex ---> http://www.facebook.com/pages/Cortex-th ... 11?sk=info
Minnie ---> http://www.facebook.com/pages/Minnie-th ... 02?sk=info
Book ----> http://pymblesoftware.com/book/
Github ---> https://github.com/pymblesoftware
Visit http://www.pymblesoftware.com
Search for "Pymble", "InstaElf", "CryWhy" or "Cricket Score Sheet" in the iPad App store or search for "Pymble" or "CryWhy" in the iPhone App store.

Well, in certain contexts MIMD can be viewed as a superset of SIMD, but as general purpose programming models they can solve the exact same set of problems. They differ though in the efficiency with which the solution can be implemented. Efficiency usually defined in terms of speed, which is what I assume you meant by "being solvable."

There is a lot of overlap between Streaming, SIMD, and Vector programming models, which makes things blurry on the data parallelism side of things like CUDA.

_________________
"Was it a dream where you see yourself standing in sort of sun-god robes on a
pyramid with thousand naked women screaming and throwing little pickles at you?"

Theory of computation text books at 20 paces, is it...?
Taken offline. PM sent.

R.

_________________
死の神はりんごだけ食べる

アレゲはアレゲ以上のなにものでもなさげ -- アレゲ研究家


J5600, 2 x Mac, 3 x SUN, Alpha DS20E, Alpha 800 5/550, 3 x RS/6000, Amiga 4000 VideoToaster, Amiga4000 -030, 733MHz Sam440 AmigaOS 4.1 update 1. Tandem Himalaya S-Series Nonstop S72000 ServerNet.

Sold:

Cortex ---> http://www.facebook.com/pages/Cortex-th ... 11?sk=info
Minnie ---> http://www.facebook.com/pages/Minnie-th ... 02?sk=info
Book ----> http://pymblesoftware.com/book/
Github ---> https://github.com/pymblesoftware
Visit http://www.pymblesoftware.com
Search for "Pymble", "InstaElf", "CryWhy" or "Cricket Score Sheet" in the iPad App store or search for "Pymble" or "CryWhy" in the iPhone App store.