PainIs4ThaWeak1 wrote:Could anyone elaborate on what exactly happened with BigAdv? I haven't been folding for the past calendar year, due to being out of the country... So I haven't paid much attention to the happenings in the F@H community.
From what I'm gathering here, it sounds that the benefit of running a 4P system has since been completely/largely diminished? (Due to no awarding of BigAdv points?)
I had been running a 3930k and 2x GTX 680's for some time, and an E8400 & GTX 295 before that. My intention was to soon invest(/throw money away) on a 4P system, now that I've returned state-side, but if it now makes more sense to put that money into GPUs, then why spend the $1200-$2000 on a complete 4P system? (Or is my logic flawed here?)
Kougar wrote:... GPUs now require 1-2 dedicated CPU cores apiece to maintain that level of performance...
BIF wrote:Kougar wrote:... GPUs now require 1-2 dedicated CPU cores apiece to maintain that level of performance...
This may have been true for the older work units, but is not true for F@H x17 work units. At least not on AMD cards using OpenCL, which I can vouch for.
On my old Q6600 system's old cards, what you say was true. Each X16 WU running on a GPU would use a CPU core and with only 4 to spare (and no Hyperthreading in a Q6600), I would run out of cores very quickly.
But almost every newer GPU can do x17 WUs. It may even be an efficient use of your money to upgrade to the lowest R9 or Nvdia equivalent card just for the ability to run those x17s and free up a CPU core.
I can't speak for Nvidia's CUDA cards; but I suspect they too don't reserve a CPU core when processing x17 WUs.
BIF wrote:I can't speak for Nvidia's CUDA cards; but I suspect they too don't reserve a CPU core when processing x17 WUs.
Kougar wrote:BIF wrote:I can't speak for Nvidia's CUDA cards; but I suspect they too don't reserve a CPU core when processing x17 WUs.
x17, and x18 cores both consume a full CPU core on NVIDIA cards, and the F@H client auto-configures a CPU core in reserve for them. It would be sort of ironic if AMD cards don't because it would mean NVIDIA and AMD have swapped places.
Flying Fox wrote:Since when? I remembered last year when I got x17's on my 660 the CPU usage dropped quite a bit. It was the x15's that needed a full core.
npore wrote:I don't know if there is already a thread for this, but I thought it'd be kinda cool to see what kinda hardware everyone is folding with - number of machines/specs/cooling etc.. I imagine there's quite a range of gear folding for the team... ancient - new, low end - high end, server stuff..
BIF wrote:I haven't seen an x16 or any other type of GPU WU since the beginning of the year. As far as I can tell, they have the pipeline fully packed with x17 WUs.
Edit: I should also add that errant x17 WUs seem to be fewer in number of late, too. In the first few months of the year, I was seeing an occasional series that would fail and checkpoint-resume many times before my system would finally finish them. I would lose an average of 2% progress on each resume, but would still make decent forward progress. Only one time I had a truly bad one that wouldn't even post back to the mothership, but that was many months ago. Reliability has been high in the last 6-8 months (or more).
Kougar wrote:Flying Fox wrote:Since when? I remembered last year when I got x17's on my 660 the CPU usage dropped quite a bit. It was the x15's that needed a full core.
Then you're mixed up!
I'm looking right now at my systems. x15 project on a 750 Ti requires <1% of the CPU. An x17 project on my 480 FTW requires 15% of a 4771, which is roughly 1.2 cores. x17 has ALWAYS required CPU overhead... the last time that was true for NVIDIA folding was back around 2007-2008 when F@H first began using CUDA on NVIDIA GPUs and Vista was just out. And yes, x18 also requires CPU overhead but if I recall AMD cards can't use that core.