Folding Chit-Chat

[Tattysnuc]

I had a raid 5 setup on a ich10r chip and it struggled with the write speeds. I would get a max of around 10-20mb/s on it.

I've now moved to a perc 6/i and 8 x 1tb drives in a raid 5 array and get around 300-350mb/s write speeds.

That's the same symptom that I'm getting, but a bit faster. about 50mb/sec. The raid array still says it was initialising so I guessed that was the reason for the low writes.

I'm looking forward to seeing what the read speed is. 2x300Gb raptors on the OS drive in Raid 0 have made a considerable difference to how it feels. the copy to the 5x1Tb drives should be done for tonight, so I'll see if it's working ok. The perc 6/i certainly looks the part - how does it compare vs the highpoint solutions? I'm fairly new to this Raid stuff....

[watercooled]

The 4+1 VRM on my motherboard gets a bit toasty when folding. I mean it should be fine as my 1055T is a 95W version. I'll try to get a few units in but I doubt it will be anything regular like the team leaders.

The apparent lack of concern for AMD GPUs annoys me. Everywhere I look it seems that the AMD cards should be faster at folding (thanks to FLOPS ratings). A few reasons I've heard why that doesn't seem to be the case is due to all the cores not being utilised or simply that they do perform faster but the points system is wonky. The fact it's been like this for years without so much as an explanation from the developers, let alone a fix, leaves me less than impressed with they effort they're putting into the software, considering how many people spend a fair bit of money folding for them to save their costs on using supercomputers.

[scaryjim]

I'm pretty sure nvidia have put a lot more effort into developer relations for folding than AMD/ATI have. Also, the current ATI card set up is targeted at pixel fill rates, so it's only actually good for certain compute tasks. Complex calculations like the ones in folding get mired up in latencies and queueing.

AMD are already working on a completely redesigned graphics core that is much more like Fermi in approach and should be a lot better for compute purposes. But that won't land for quite a while yet...

[Tattysnuc]

The 4+1 VRM on my motherboard gets a bit toasty when folding. I mean it should be fine as my 1055T is a 95W version. I'll try to get a few units in but I doubt it will be anything regular like the team leaders.

The apparent lack of concern for AMD GPUs annoys me. Everywhere I look it seems that the AMD cards should be faster at folding (thanks to FLOPS ratings). A few reasons I've heard why that doesn't seem to be the case is due to all the cores not being utilised or simply that they do perform faster but the points system is wonky. The fact it's been like this for years without so much as an explanation from the developers, let alone a fix, leaves me less than impressed with they effort they're putting into the software, considering how many people spend a fair bit of money folding for them to save their costs on using supercomputers.

As far as I was aware, it's not to do with the bonus points, it's to do with the rate at processing the data. The code has been kept up to date with the AMD chips. I think that I read that there was a beta client that brought the AMD code up to date, and that it was significantly better, but there was still a strong bias toward NVidia still. Guess that Stanford are just working more focussed on the Green team. Saying that though, the latest changes definitely favours Intel or high core count CPU's, which on the desktop market, means that Intel have a lead as they've got desktop processors that support 12 threads....

[watercooled]

I've not heard anything from the F@H developers though. But I have read what a few programmers have written and they seem to think VLIW should run F@H very well.

@Tattysnuc: It does seem there is a bias. Even if one card type is faster than another, the points shouldn't be so different - look at CPU vs GPU, GPUs do certain units far faster than any CPU could run them but get about the same or less PPD.

[1stRaven]

That's the same symptom that I'm getting, but a bit faster. about 50mb/sec. The raid array still says it was initialising so I guessed that was the reason for the low writes.

I'm looking forward to seeing what the read speed is. 2x300Gb raptors on the OS drive in Raid 0 have made a considerable difference to how it feels. the copy to the 5x1Tb drives should be done for tonight, so I'll see if it's working ok. The perc 6/i certainly looks the part - how does it compare vs the highpoint solutions? I'm fairly new to this Raid stuff....

Look at the price of the card. The perc is a server solution so quite high end but with the performance to match.

I have a highpoint 2 port card and while good, it cant match the perc in terms of ports and bandwidth availability.

[Tattysnuc]

Look at the price of the card. The perc is a server solution so quite high end but with the performance to match.

I have a highpoint 2 port card and while good, it cant match the perc in terms of ports and bandwidth availability.

both are £100-£130 on Ebay so price is pretty similar for 8 port cards which is what I'm after. The Perc says it can go up to 32 devices, but I can't see how...

[watercooled]

Port multipliers?

[Tattysnuc]

I've not heard anything from the F@H developers though. But I have read what a few programmers have written and they seem to think VLIW should run F@H very well.

@Tattysnuc: It does seem there is a bias. Even if one card type is faster than another, the points shouldn't be so different - look at CPU vs GPU, GPUs do certain units far faster than any CPU could run them but get about the same or less PPD.

True, but what is the differential in the WU size? I know I get a shed load more points for doing BIGADV units, but they are 200mb in size, 350mb when expanded - the Fermi GPU ones were smaller @ 50mb (not sure expanded...)

[1stRaven]

ummm not used the highpoint 8 port card but do use the perc card all day long at work so went with what I know. Does the highpoint have battery backup?

[Tattysnuc]

ummm not used the highpoint 8 port card but do use the perc card all day long at work so went with what I know. Does the highpoint have battery backup?

Nope. Been reading around, and technically it's not true hardware raid either.

My array has now finished initialising, and the write speed has grown massively to anything up to 100mb/sec, so it's performing normally now. Looks like it was just becauseit was still initallising in th bios being the reason it wss so slow...

Looks like you got out of the BIGADV folding at the right time. Stanford have just dropped the points for BIGADV's...

http://foldingforum.org/viewtopic.php?f=24&t=19059

Maybe there's some truth in the NVidia favouritism rumours..... It does sound like someone's had a good moan about the points system NOT favouring their platform....

[1stRaven]

Nope. Been reading around, and technically it's not true hardware raid either.

My array has now finished initialising, and the write speed has grown massively to anything up to 100mb/sec, so it's performing normally now. Looks like it was just becauseit was still initallising in th bios being the reason it wss so slow...

Looks like you got out of the BIGADV folding at the right time. Stanford have just dropped the points for BIGADV's...

http://foldingforum.org/viewtopic.php?f=24&t=19059

Maybe there's some truth in the NVidia favouritism rumours..... It does sound like someone's had a good moan about the points system NOT favouring their platform....

Thats going to have an impact! Whats the reaction been like?

[Tattysnuc]

Lots of angry bunnies over on Bit-tech.... It's kinda mooted though because it's all in a good cause. This was just a mechanism to promote folding.

To be fair, I only started doing it because I had an NVidia card. After that, it was about improving my scores, and it became more efficient to fold on the CPU only. Will have to revisit that decision now that BIGADV are not as significant.

[Tattysnuc]

Regarding the stats, I'm very busy with work at the moment. The earliest I think I'll be able to do them will be Monday 11th July. Sorry folks, but life is VERY busy at the moment. If I get chance, I absolutley will have a go, but I'm tied down 7 days a week lately on some big projects which is eating up all my time...

[Tattysnuc]

OK guys, will be working on the stats a little later today.

Back from anniversary weekend and the wife's gone out with a friend leaving me to my mischief

[Tattysnuc]

Yay... only 1.7 weeks till I'm in the Number 2 spot. It's taken me 15months to catch Lowe, which isn't bad by my reckoning...

Should be in Second place by the end of July. Anyone else gunning for a specific position at the moment?