RTX 2060 vs RTX 3050 Performance

[hrsetrdr]

The "sweet spot" price wise for me is around the $400 mark, for a GPU folding acquisition . Currently the RTX 2060 and the RTX 3050 fall into this range, but performance numbers are not what I expected, considering the hardware specs.

RTX 2060 specs:

6GB 192-Bit GDDR6
1920 CUDA Cores
PCI Express 3.0 x16

Estimated PPD : 1,724,376 https://folding.lar.systems/gpu_ppd/bra ... e_rtx_2060

RTX 3050 specs:
8GB 128-Bit GDDR6
2560 CUDA Cores
PCI Express 4.0

Estimated PPD : 1,455,823 https://folding.lar.systems/gpu_ppd/bra ... e_rtx_3050

I've always thought it was all about the # of CUDA cores, what else could account for the difference here?

[aetch]

Personally, I don't know but it is something I'm interested in myself.

It's thanks to charts like these that we can compare different GPUs because simply looking at the tech specs doesn't really reveal much.

I have a GTX 1080 Ti and an RTX 2070 Super. On paper the 1080 Ti should be the more powerful and I would expect it to produce the higher PPD.
However, experience has shown the RTX 2070 produces roughly 20-25% higher PPD in folding, all while drawing less power.

[bikeaddict]

The TDP or memory bus width and bandwidth could be factors in folding performance.

My experience with an EVGA RTX 3050 XC Gaming over a month showed it consistently estimated lower PPD than a GTX 1660 Super and 1660 Ti. The 3050 usually estimated 1-1.2M PPD and my 1660 cards have always been 1.2-1.6M PPD.

The RTX 3060 is the better buy in the $400 range, with PPD usually at 1.5-2M. Keep an eye on evga.com for the 3060 to be in stock. They sell out quickly. If you can stretch to a 3060 Ti, that is a huge jump in PPD for slightly more cost. Another way to get a 3060 is to grab an EVGA 3050 and do a step up to a 3060. Not sure if they offer a step up from 3060 to 3060 Ti at the moment. There are 3050 and 3060 cards in stock at the moment of posting this message. Remember to find an associate code for 3% off GPUs at EVGA.

https://www.evga.com/products/ProductLi ... y&Onlyin=1

[hrsetrdr]

The RTX 3060s have 'attractive' PPDs (2,126,824 PPD) according to: https://folding.lar.systems/gpu_ppd/bra ... e_rtx_3060

Newegg has one sub-500$ RTX 3060 currently. No doubt I'm in a hurry to get another GPU online before summer electric bills start, but maybe I should wait for GPU prices to drop a bit, and spend my money on a much needed new system.

[bikeaddict]

The EVGA site occasionally has a 3060 for $369.99 with rebate in recent days, or $429.99 without rebate a week ago. You just have to check the site every hour or two during the day and grab one fast. It helps to stay logged in and have an associate code ready to paste.

[hrsetrdr]

Argh! O.K. I signed up with EVGA, they got the hook set and I am so driven to get the RTX3060, or maybe the RTX3060ti at a price as close to the MSRP as possible.

[MeeLee]

The 3050 has a very low boost speed of below 1500Mhz (~1480Mhz), vs the 1870Mhz boost of a 2060.
More modern 2060s you can run at 2050Mhz boost easily, but the 3050s I don't know.

Even though the 3050 has 33% more cuda cores, it also is operating at ~25% lower frequency. That paired with the slower memory bus, and potential drivers issues in Windows 11.
What you should do, is run a 3050 in Linux, and play a bit around with the wattage, and overclocking potentials.

There's a third issue, in which Nvidia has disabled the full potential for miners; and I'm not sure if this affects FAH. They say no, but it wouldn't surprise me if there's a bug that still lowers performance on other than mining programs.

[DocJonz]

@MeeLee - I have a number of RTX 3070 v2 (i.e. Lite Hash Rate) cards Folding under Linux - I can confirm that they are not hamstrung by miner-deterrent algorithms.

[hrsetrdr]

The 3050 has a very low boost speed of below 1500Mhz (~1480Mhz), vs the 1870Mhz boost of a 2060.
More modern 2060s you can run at 2050Mhz boost easily, but the 3050s I don't know.

Even though the 3050 has 33% more cuda cores, it also is operating at ~25% lower frequency. That paired with the slower memory bus, and potential drivers issues in Windows 11.
What you should do, is run a 3050 in Linux, and play a bit around with the wattage, and overclocking potentials.

There's a third issue, in which Nvidia has disabled the full potential for miners; and I'm not sure if this affects FAH. They say no, but it wouldn't surprise me if there's a bug that still lowers performance on other than mining programs.

Interesting information, I wasn't sure if OCing GPUs yielded actual increase in ppd. All my GPUs are running in Linux, except my Dell 15 7000 7567 laptop, which is folding on it's GTX 1050 in Win 10. The laptop will 'retire' from folding when the parts for a new build arrive & assembled. According to folding.lar.systems/gpu_ppd the GTX1050 mobile averages 255,278 PPD, but FAHcontrol is currently reporting 348,000 PPD.

[aetch]

It's entirely up to you how you run your kit but I personally try to discourage others from overclocking. The card runs hotter, it draws disproportionally more power, the fan spin faster and noisier.

Many of us actually underclock our cards. I normally run mine about 70% power. That gives me something like 80-90% performance and it runs near silent.
For Windows I would recommend MSI afterburner.
For linux you're looking at a commandline component of the NVidia driver called nvidia-smi.

A note on PPD - FAHControl returns an estimate based up on the current work unit being processed. It will rise or fall depending on the project. You will see similar PPD for different work units from the same project. On any given day my 2070 SUPER can swing from ~2.5 to ~3.2million PPD. The average on LARS will be based upon the results of hundreds, probably thousands of work units.

[hrsetrdr]

It's entirely up to you how you run your kit but I personally try to discourage others from overclocking. The card runs hotter, it draws disproportionally more power, the fan spin faster and noisier.

Agreed. The novelty of higher clocks is long faded for me, stock is just fine, less heartaches. Back when posting & sharing benchmarks on MadOnion(Futuremark) was a 'thing', pushing my GeForce 2/3/4 series graphics cards was my 24/7 passion, until I started with f@h.

[bikeaddict]

My brief experience with mild overclocking on a GTX 1650 Super is that you are at increased risk of intermittent BAD_WORK_UNIT errors that will cause the WU to dump at a random point before completion and lose points. The small few percent increase in performance is not worth the risk.

What I do is to "overfan" which is to set the fan to a fixed speed like 70% rather than the default 25-35% to force the GPU to run about 5-10°C cooler and stay at a boost clock that is 15-45MHz higher but doesn't cause bad work unit errors. This works for me since I have machines in a basement where the noise isn't a problem.

[aetch]

What I do is to "overfan" which is to set the fan to a fixed speed like 70% rather than the default 25-35% to force the GPU to run about 5-10°C cooler

I have occasionally done this as well. One of the side benefits is that the fan gives a constant drone, which is easier to tune out.

Thanks for the reminder.

[hrsetrdr]

My brief experience with mild overclocking on a GTX 1650 Super is that you are at increased risk of intermittent BAD_WORK_UNIT errors that will cause the WU to dump at a random point before completion and lose points. The small few percent increase in performance is not worth the risk.

What I do is to "overfan" which is to set the fan to a fixed speed like 70% rather than the default 25-35% to force the GPU to run about 5-10°C cooler and stay at a boost clock that is 15-45MHz higher but doesn't cause bad work unit errors. This works for me since I have machines in a basement where the noise isn't a problem.

Oh yes, the BAD_WORK_UNIT errors.
I set fan speeds at 70%, not too loud and keeps nice and cool.

[hrsetrdr]

Argh! O.K. I signed up with EVGA, they got the hook set and I am so driven to get the RTX3060, or maybe the RTX3060ti at a price as close to the MSRP as possible.

Early this morning I bought an RTX2060(Amazon) but later discovered the MSRP priced RTX3060s from EVGA were available, I cancelled the RTX 2060 and grabbed an RTX 3060 from Evga. I paid the 28 bucks shipping but is only costing $8 more for the rtx3060 than the 2060.