2012 slump?

[TonyStewart14]

I noticed on Wikipedia a graph by Jesse V that showed the number of contributors to F@H over time.



As you can see, participation has plummeted this year. Perhaps it is due to a rise in the popularity of tablets and mobile devices over desktop and laptop computers. In any case, it is concerning that it would drop so much. Let me know if there is any other cause or theory for this. It's just one more reason to keep folding!

[mdk777]

discussed here:

viewtopic.php?f=16&t=21367
viewtopic.php?f=16&t=21317

I haven't seen any changes since these discussions...some improvement in seasonal, but nothing firm yet.

NVDA did release a CUDA version for Kepler today, so perhaps a core for Kepler could come out soon.

[k1wi]

The graph comes about from the discussion here: viewtopic.php?f=16&t=21317, as mdk777 beat me to. To be fair the slump in the graph coincides with a particularly hot Northern Hemisphere summer and daily logging of the official stats, which does not exist for previous summers.

As you might see from the source (a google doc), it does appear that overall production has started to increase again, albeit still around early July levels.

[Jesse_V]

We can't exclude the possibility that the FAH statistics site may be inaccurate in some way. That's an underlying assumption of this conclusion, just saying.

In any case, I approve of this thread.

[7im]

People are selling their folding pharms (lot's of dual cores and quads) and upgrading to higher density 8 core machines or higher.

Hottest summer on record didn't help.

[bruce]

A rather basic question: Does a dual quad (or whatever) count the same as a Uniprocessor in that data?

[Joe_H]

I don't recall seeing any post from PG that clarified how they count CPU's for the official stats. Though from other places they list CPU counts, I would infer that one client running on a "CPU", whether that is single core, multi-core, or a dual quad, counts as one. The graph I watch more than number of clients is the TFLOPS by platform and the total.

[7im]

No post needed when it's explained right on the stats page. Each System ID + Machine ID = 1 active client. So in v6 clients, if you fold on 2 GPUs, they each had a unique Machine ID, so 2 active clients.

I still have no idea how V7 slots get counted, if at all...

[patonb]

So yha, the 4p AMD replacing the 4 boxen and multi gpu setups are the cause.

You can see the decline starts as Linux takes off.

[kiore]

No post needed when it's explained right on the stats page. Each System ID + Machine ID = 1 active client. So in v6 clients, if you fold on 2 GPUs, they each had a unique Machine ID, so 2 active clients.

I still have no idea how V7 slots get counted, if at all...

According to my stats I have 4 'machines' which is accurate in that I have a 2 GPUs an SMP and 1 uniprocessor in v7 , previously the number reported in v6 seemed to make no sense in relation to what I had running. This could be a reporting bias that is exaggerating these stats. The uptake of portable devices may have had an impact as well but I would have thought this offset by the advances in processing power that have also occurred.

[mdk777]

let's leave core/unit counts to software vendors trying to maximize their per seat revenue.

Unfortunately, no matter how you count the participation, or number of processors; both the total FLOPS and the Total native 86 FLOPS are almost half of the previous high.

In two years you would expect the opposite just from Moore's law(2x the FLOPS from the same number of users if they all upgraded).

The trend is not good. Focus really has to be on solutions and not on excuses.

[7im]

Solutons like an easier to install V7 client with a nicely upgraded viewer?

[alancabler]

kakaostats is listing only 48K+ lines for active donors (w/in past 50 days)...

[mdk777]

Solutons like an easier to install V7 client with a nicely upgraded viewer?

Looking at the big picture, really too little too late.

The enthusiast market, (those using multi-core CPU and massive GPU) is getting smaller everyday. The V7 Client addresses a 4 year old problem. Now the problems have changed.

The discrete GPU market, along with the desktop market, continues to shrink. IGP are taking market share in the remaining desktop market.

Anyway, we all know the trends. The question remains how to utilize the remaining processors that are in the installed base.

NVDA says they have sold 350,000,000 CUDA capable processors.
Currently, 6000 of those are FOLDING.

WHY??

I think we can safely say that the weather is not the root cause.

There has to be other reasons.

As I have said a hundred times before, the solution is taking a customer centric, rather than developer centric approach to the problem.
Every day, people try the V7 client and conclude that the GPU takes up too much power/temperature/noise/and responsiveness.

Say the default client used the GPU at 25% capacity. Say you had 6million people willing to fold at that reduced GPU utilization rate.

I come up with something like a 250x increase in FLOPS for FOLDING.

Yes, yes, I know that total FLOPS is not the end all be all, and the GPU FLOPS are not necessarily the most flexible....

My point is that we all know what is holding the project back, and it isn't the weather. The decline is due to users demanding more flexible and current software support, and leaving when they exhaust their patience waiting for it.

Blame resources, difficulty in programing and delay in industry driver support if you want. The customer, the end user just can only care so long.

[7im]

Any answers more than simple environmental, social, or economic issues are better suited for the "What's holding FAH Back" thread, not this one.