Steady Decline? :-(

[Stefano Vaj]

I am a little concerned for the decline in numbers of active processors, not to mention TFLOPS, reported by the statistics.

Especially worrisome seems to me the figures related to Linux users, which might play an actual role with regard to SMP but above all reflect or anticipate some Net trends...

This in turn could have something to do with the relative lack of communication and/or news from the Pande Group. Or are there other reasons I am not aware of?

[RAH]

Well, several reasons.
Mac - which there are fewer users - the old machines will not work with the newer core.
Linux - was used mainly because it gave better output then windows, which as since gotten
more equal with the A3 core. Making it easier to run the GPU2 client also.

They rest - some due to the waiting for a new client (ie ATI/A3/Nvidia), being that most
people are Doctors, they have run out of patients.

[patonb]

Plus its warming up. Summer always sees a decline due to the heat associated with high processor loads.

[7im]

Why is Pande Group always to blame? They post a lot of things, and no one ever reads it. Is that really a communication problem?

Any way, please remember that MANY people formerly running Linux VMs to get higher points on the Linux SMP client are now all switching back to a Windows SMP client because of the new v6.29 client and the new A3 work units that run on both Linux and Windows.

And with the bonus, they get more points from running a single SMP client instead of multiple SMP clients. Both of those might be part of the active linux client numbers decline, don't you think? Could also account for a very slight drop in the over all number of active clients as well.

As for flops decline, the biggest reason is most likely as they said above, summer is coming. It's also a result of some people choosing to run only SMP clients instead of both SMP and GPU clients. With the new bonus system, and with a few types of fast SMP machines combined with older slower GPU cards, they can earn the same points just running the SMP client faster, and use less electricity. So the flops go down.

Let's not forget the economy is still in the toilet, in which EVERYTHING is in decline. And that's certainly not PG's fault.


I'd be curious to see the rate of decline in other projects as compared to fah, over the same time period. That might better indicate a trend, yet not give the complete answer.

[ElectricVehicle]

A large portion is the decline in PS3's from a high of 1.5 TFLOPS native / 3.0 x86 equivalent. I used to run 2 PS3's now I only run one and I'm tempted to shut it down since the SMP and GPU clients use maybe twice as much power but produce 4 - 30 times as much science based on the points, which aren't perfect, but largely align with the scientific value. If the points are every totally out of whack then Pande group can change the points systme to align with the science.

I'm fascinated by the TFLOPS also and from a computing pinnt of view, it's nice to see 10 TFLOPS (x86 equiv) and growing, but that doesn't correlate exactly to the science.

Think of ot more as having somewhat less TFLOPS now, but each of of those TFLOPS is much higher quality, producing more science per TFLOP.

The CPU clients have relativley few TFLOPS, are slow compared to GPUs, but produce high quality simulations.

The GPU clients have higher TFLOPS, and run the science very fast but aren't very flexible, so they produce moderate quality simulations but can simulate longer time sequences than is possible on CPUs, within several months, or several years of a grad student's career!

A small effect is the lack of a Nvidia GTX 470/480 compatible client. That will come soon. Then this next generation architecture can start contributing to FAH. It won't be a complete revolution, but it will be sginificant, with FAH perfomance at least equal to a GTX 295, probably at least 20% better, maybe a factor of two or more, but we have to wait and see until their is a released client for the Nvidia GTX 470/480.

[7im]

Don't sell your PS3s just yet. Turn them off if you like (assuming you replace them with higher scoring hardware).

[road-runner]

I'd be curious to see the rate of decline in other projects as compared to fah, over the same time period. That might better indicate a trend, yet not give the complete answer.

Here is a comparison WCG chart from here... http://stats.free-dc.org/stats.php?page=proj&proj=bwcg

[7im]

I think we'll have to wait for Stefano to define what time period was being considered before we start making more graphs, or at least decide on a time frame, like YTD, 1 year, etc. I think we should use a time frame longer than 1 month to help normalize seasonal trends.

[road-runner]

Well this one goes back to 2005... http://boincstats.com/stats/project_graph.php?pr=wcg



Stanford really needs to get it together...

[mdk777]

Well, from memory over the last year:

ps3 folders down = 10,000
ati down = 2200
nv down = 7000

However total cpu up 34,000

Hence, the total number of folders has actually stayed the same or even increased. However, the loss of GPU and PS3 folders has reduced the flops total from the high 8000 to the current 6000 range.

This trend has been going on for around a year and not just with the start of summer.

[ElectricVehicle]

Don't sell your PS3s just yet. Turn them off if you like (assuming you replace them with higher scoring hardware).

I still fire them up for Blu-Ray and the kids do the Lego Batman! One PS3 is mostly off, the other PS3 is 24x7 folding. I used to have them both running, gloating in how awesome 2,000 PPD was!

The latest folding computer with an i7-980x, once the GTX 480 client is out, should do well over 50,000 PPD - 50 times the PS3 on 3 times the power (550W vs. 200W for the 60GB PS3). Higher scoring hardware on a semi-economical basis, serving multiple purposes, within the power limits of the outlets I have available is always the goal!

[7im]

@ road-runner, how do they define "total credit" ?

Is that the same as active clients? Or is that a measure of production, like FLOPS? Or is that just total WU count that climbs like in every other project.

Let's try this... RAC charts for a few of the top Boinc projects (last 60 days) (recent average credits)

Einstein@home, trending down, but appears to be recoverying...


Rosetta, starting to decline...


SETI, declining...


All BOINC Combined, toppped out, and appears to be declining...



My crystal ball broke a while back, so I can't read minds any more. This, to me, looks like summer coming. But who knows? It's all just guesswork.

[k1wi]

This, to me, looks like summer coming. But who knows? It's all just guesswork.

It's still a good 6 months away here... (I know, I know, most users are in the Northern Hemisphere... but still)

If we're just talking about credits (which != flops) then surely it wouldn't be too hard for some of the third-party sites to graph that? For example:
(who lost the 40billion points? )

My own personal experience: When VMware Player 3.0.0 came out I converted my 3x2core VMs into 1x8thread VM: loss = 2 machines. When I switched to bigadv I turned off my ATI GPU: loss = 1 client. Now that I'm earning 20,00+ ppd on one computer, I have to look at the other clients I'm using around the house & decide whether it's worth folding on them (returning ~100-300ppd each) One's a p4 3.2 which I'm looking to phase out and the other is a Centrino 1.83 laptop, so it's probably likely I'll stop folding on them. Loss = potentially 2 machines.

Total loss (assuming I do shut down my two inefficient folders): 5. (Pre consolidation: 8 clients, post consolidation: 3 clients.)

Also had a problem with my university computer where they implemented a new web proxy that blocked my clients for about 3-4 months, so for a while I was down an additional 2 machines. Just managed to get them up and running again.

I'd say that there are possibly a lot of people like me. There are also a large number of people who are not like me, who fold on a p4, and join up, fold a bit, then decide it's not for them, perhaps for electricity costs (as computer's idle draw decreases, this is more pronounced).

What I'd find interesting would be some stats such as the median/quartile ppd or TFLOPS of clients/users over time, so we can see how much science is being done by seriously powerful (and say purpose built machines, like those owned by people who are likely to frequent this forum) and how much is being done by 'average computer users'.

[codysluder]

Think of ot more as having somewhat less TFLOPS now, but each of of those TFLOPS is much higher quality, producing more science per TFLOP.

This is a very important concept. If anybody considers the change made by the SMP bonus plan, they'd realize that the Pande Group is telling us emphatically that the best science doesn't come from the most TFLOPS but rather from a combination of good TFLOPS and faster WU returns. The age-old question about running two classic clients on a P4 with HT or the newer question about running multiple SMP clients in multiple VMs has been answered. Those used to be ways to maximize TFLOPS and PPD. Now it no longer maximizes PPD so get with the program and run slightly lower TFLOPS to get higher PPD.

[road-runner]

@ 7im I have no idea what any of those are except seti, most folks I know want to cure cancer, aids, or other diseases instead of looking for aliens which is one of the reasons I also do folding, I could care less about ET. With that being said here is the WCG RAC graph, the one most folks go to when they leave folding...