Folding Forum

bruce wrote:

new08 wrote:I have downloaded the latest whql driver for a later trial anyway.[/'quote] if you downloaded it from WindowsUpdate, you may have trouble with OpenCL. If you downloaded it from nVidia, you'll have a choice.

Does the GPU list allow all memory size versions of a card?
The 3MB version GTX1050 ti is on there- took a bit of finding, as right at the end!

When someone buys a new GPU, I recommend they get the model with the smallest VRAM. They're cheaper, use less power/generate less heat, and I've never seen a performance difference.

Actually,
On GeForce cards, larger ram usually goes paired with more cores; like,
The GTX 1050 with 2GB of RAM has 640 cores, while the 3GB card has 768 cores (like the 1050 TI, it also has 786 cores).

The larger ram cards sometimes have names, like 'ti' or 'oc' added.
I presume, and just speculating here; that the 'oc' cards are 'failed', or lower performing 'ti' cards, sold as regular or 'oc' cards, but are overclockable to almost the same speeds as the TI cards.

In any case, graphics cards with more ram, are usually tuned with more performance to make use of the extra ram.
Few cards have high horsepower, but not the RAM to back it up. It would cause performance issues.


As far as folding on GPU, I'd recommend the 1060. It seems to be the best bang for the buck.
If you're concerned about power usage, the 1050 ti with CPU folding will probably get around 200kPPD tops, using 75W + 65W (on average for PCs) = 140W,
Folding on the 1060 by itself, no CPU, will get 330+k PPD, at 120W + about 25W of system idle power = ~145W.
Same power usage, 1/3rd or more better folding.
price difference is minimal between the 1050ti and 1060 (especially if you get them second hand).

ProDigit wrote:Actually,
On GeForce cards, larger ram usually goes paired with more cores; like,
The GTX 1050 with 2GB of RAM has 640 cores, while the 3GB card has 768 cores (like the 1050 TI, it also has 786 cores).

The larger ram cards sometimes have names, like 'ti' or 'oc' added.
I presume, and just speculating here; that the 'oc' cards are 'failed', or lower performing 'ti' cards, sold as regular or 'oc' cards, but are overclockable to almost the same speeds as the TI cards.

Actually this is false as often as it is true. In the example you post, the 3 GB VRAM model 1050 was introduced about 18 months after the 2 BG model. It has a 96-bit memory bus to that RAM as compared to a 128-bit bus on the 2 GB model released earlier.

The "ti" cards are higher performance versions of the card and use a different chip. The "oc" cards indicate ones that have been factory overclocked compared to the reference design cards by nVidia. They are usually sold by other card makers using chips from nVidia. How stable they are doing folding varies.

I settled on the 1050ti as it was cheap new and matched the power supply of the HP Elite.
I just wanted to prove concept at the cheapest price.
There is a power lead option with the 1050 but I don't think it has to be used for stock speeds.
I worked out the cost of folding on the beefier cards over the years and it does add up.
The old P/S used to roar quite a lot, too! The new rig is pretty quiet - so banking on a quiet graphics card.
As to production, 3 years ago before stopping I was doing 20K ppd, without solid folding, so getting up to 10x that would be fine with the 1050, at a lot less taken power.

Joe_H wrote: Actually this is false as often as it is true. In the example you post, the 3 GB VRAM model 1050 was introduced about 18 months after the 2 BG model. It has a 96-bit memory bus to that RAM as compared to a 128-bit bus on the 2 GB model released earlier.

The "ti" cards are higher performance versions of the card and use a different chip. The "oc" cards indicate ones that have been factory overclocked compared to the reference design cards by nVidia. They are usually sold by other card makers using chips from nVidia. How stable they are doing folding varies.

I can only vouch for my EVGA 1050 OC card, it runs very stable, with a 270Mhz overclock on CPU (from 1518Mhz to 1786Mhz), and 900Mhz on ram to 4.4Ghz; at close to the stock power requirements of the card (35W vs 30W stock), before throttling starts.
The EVGA app has a hardware monitor, that makes it very easy to see when the card is being pushed too hard.
Usually within 10 seconds, the hardware monitor can tell me when the card is failing, well before the WU fails.
Only with RAM overclocking do I need to be careful, as RAM overclocking gives no warning.
WUs immediately fail, when RAM OC is too high.
Thankfully finding that limit usually takes less than 10 minutes per card; and then it's just a matter of creating a safety margin.

For instance, my RAM, I run it at 50-100Mhz below the max stable result during testing.
CPU I run 20-50Mhz below max stable result.
So far, all WUs have come out good; and no card failure for the past 36 hours.

new08 wrote:I settled on the 1050ti as it was cheap new and matched the power supply of the HP Elite.
I just wanted to prove concept at the cheapest price.
There is a power lead option with the 1050 but I don't think it has to be used for stock speeds.
I worked out the cost of folding on the beefier cards over the years and it does add up.
The old P/S used to roar quite a lot, too! The new rig is pretty quiet - so banking on a quiet graphics card.
As to production, 3 years ago before stopping I was doing 20K ppd, without solid folding, so getting up to 10x that would be fine with the 1050, at a lot less taken power.

Take a look at MSI Afterburner, or EVGA's Precision XOC.
You can safely overclock your card to increase PPD, and as long as you're not increasing the voltage, power consumption will be nearly identical.
It is also recommended to use the power plug when you can.
The GTX 1050 TI uses the max power, a PCIE slot can provide. Over time ti could wear out the caps on your motherboard. Just using the plug, or an adaptor plug from a hard drive or SATA connector, could balance out the voltage and currents a lot better. Adaptor plugs cost next to nothing. A SATA bridge should provide with about 75W of power; and a SSD drive would probably use less than 5W of this. All the remaining can be routed to the Graphics card; which means less load for the motherboard to deal with. (they should deal with it, but still...)
I'd also recommend not to use a hard drive, if you're concerned about power usage. An SSD has much lower idle power draw than hard drives.

Thanks for tips on power lead. Worth checking out- I have some spare SATA lead outs, so hopefully will match up.
At least I know adapters are available for the job & should cope.
Not too bothered about o/c, till 1050 is installed.
I used those utils on EVGA cards before -but they don't seem to work on other makes now.
On the power front, running at half previous take will be fine.
I knew, a way back, that really lower power rating chips were under way, just hadn't got into it till now.
A big improvement!

My EVGA app, safely overclocks MSI, EVGA and other cards, that are based on the MSI GeForce chipsets.
It may no longer be the case that you can't overclock from other brands. It works on mine.

I've used them before ok and will try on the new card later. They are handy utils.

On second thought, my cards still seem to error, even after overclocking them mildly on a very stable overclock.
I want to see if memory can be overclocked, instead of GPU, because GPU overclocks give the most errors.
At least on my GTX 1060. On my GT1030 it seems to run fine with a mild overclock (~50Mhz).
But I've come to the conclusion that your PPD/WATT ratio remains largely the same when overclocking.
In other words, you'll be paying more electricity. In many cases it makes the whole event less efficient, especially when you consider that overclocking can damage the graphics card.

I agree that in this case I'll be happy to run at stock speeds on both PC and graphics.
There's so much improvement on the efficiencies now,that it hardly seems worth the hassle - unless there's a bet involved

Yeah, so in the app, a stable overclock on most of my GPUs is anywhere between 100-200Mhz.
But even at 50Mhz it still errors on some.

The memory on the other hand, can be overclocked by 600-1000Mhz on my cards. But with a 400-600Mhz overclock I use about 5 watts more power (+1.5%) and my PPD points go up by about 5-7%. The results come in stable.
Generally it's not worth doing, unless you want to spend several hours of your time tweaking a balanced setting.
Though once set up, it works well for systems that fold 24/7

ProDigit wrote:Yeah, so in the app, a stable overclock on most of my GPUs is anywhere between 100-200Mhz.
But even at 50Mhz it still errors on some.

If you are still getting some errors, then your overclock is NOT stable for folding.

The official position of FAH is that overclocking is (A) not supported and (B) not recommended. There are a number of overclocking support forums and THIS IS NOT ONE OF THEM.

If your overclocking increases fAH's throughput by 1% it's a lot of work for a very small return -- and it only applies to a very few machines. If FAH improves their efficiency by 1% that change applies to EVERYBODY so it is a significant improvement to science for all of FAH except for those few individuals whose machine cannot tolerate the highly efficient code being distributed by FAH. (and they have a choice about removing their overclocking. The loss of an occasional unstable WU should be enough of a disincentive to convince them that overclocking FAH is not a good idea.

Joe_H wrote:

ProDigit wrote:Yeah, so in the app, a stable overclock on most of my GPUs is anywhere between 100-200Mhz.
But even at 50Mhz it still errors on some.

If you are still getting some errors, then your overclock is NOT stable for folding.

Indeed. Though it is stable enough for gaming and bitcoin mining.