Folding Forum

First off, I must mention that I have no problems with the technical language on the client s/w download page. However, many medical professionals have told me that they find the page confusing. (So I guess medicine really is an "art" rather than a "science")

So here are my suggestions for a few changes:

1) the phrase "Graphical client" should be changed to either "GUI client" or "non-text client".

2) Since "GPU" is so close to "CPU" (as in "GPU-Client" vs "CPU-Client"), "GPU" should be changed to "GPC" (Graphical Co-processor)

Just my 2-cents worth.

Neil Rieck
Kitchener/Waterloo/Cambridge,
Ontario, Canada.
http://www3.sympatico.ca/n.rieck/docs/GHLE.html

An incomplete diagnosis would seem to be the problem. The "types" of clients are explained at the bottom of the download page.

What are the different types of clients? The Consoles run in the background and are launched from a terminal window (command line), to which they output text messages about the simulation's progress. The Graphical client runs in the background and has a graphics window available that allows you to see the protein being simulated. The screensaver runs just like any other screensaver, except that it also runs our calculation in the background.

Feel free to suggest any additions to that language you think might help our poor misguided medical friends. Or maybe that paragraph should be moved up the page, now that the list of clients is so very long. I do see how that might get missed.

I agree with the CPU vs. GPU confusion, but "co-processor" is somewhat inaccurate. GPGPU is more accurate, less confusing, but just too much extra typing to say the same thing. And changing the name kinda kills the CPU and GPU naming symmetry.

The reason I was thinking "co-processor" is this: back in the day, an 80x86 CPU could only do Integer math in hardware. All floating point math had to be simulated in software until Intel introduced the 80x87 FP Co-processor. In later generations this FP logic was moved on die with teh CPU but I digress.

As I understand it, your FAH client software utilizes the ATI card as a vector (or even tensor?) co-processor.

But I get your point.

Neil Rieck

The ATI card does all of the processing for the GPU client. The CPU usage you see with that client is a function of DX9 polling for data. The CPU feeds the GPU with data, but does not process work unit data, so co-processor does not fit the description quite like the 387 co-processor I added to my 386 SX-25.

What do you think about moving that description paragraph, or updating any of that text?

I think your suggestion of moving the description to the top before the list of clients is a very good one, especially since the list is so long. If I'm a new and not very sophisticated user (which I was and still am [not sophisticated, that is]), I'm not going to go very far past the "recommended" downloads, especially when I start seeing betas. I might add a line that the graphical client is more user friendly, especially in configuring it.

7im wrote:The ATI card does all of the processing for the GPU client. The CPU usage you see with that client is a function of DX9 polling for data. The CPU feeds the GPU with data, but does not process work unit data, so co-processor does not fit the description quite like the 387 co-processor I added to my 386 SX-25.

What do you think about moving that description paragraph, or updating any of that text?

Sure. Try moving it first then get a non-technical person (like a medical doctor or dentist) to see if the page is less confusing. If moving some lines doesn't help much then you'll have to tweak some words.

p.s. You poll the card? I just assumed it was all interrupt driven.

neilrieck wrote:
p.s. You poll the card? I just assumed it was all interrupt driven.

One of the fah people sent me a private message stating that polling the graphics card is something that is done by the DirectX-9 API (which means that the polling "feature" was put in by Microsoft). He also mentioned that MS fixed this in DirectX-10 under Windows-Vista (at least for PCI-E cards).

Now I know that fah requires DX9 but I do have the fah-GPU software running on a Vista machine with an ATI card in a PCI-e slot "and the CPU associated with that process is not maxed out at 100%". Just food for thought.