Any recommendations for inexpensive secondhand GPUs?

[bamapolitics]

Mining cards are a great value for F@H since the crypto market crashed and every miner is trying to get their money back. I have a p106 Zodiac card and it runs just fine in my gaming system. I would look for them if you want to save a buck. $50 is a steal, $75 is average.

Gaming while folding depends on the game you are running. Guild wars 2 (is a CPU heavy game) will run fine while folding. low powered games like dead cells, dusk, world of tanks or any game that isn't GPU heavy will run with minimal FPS drops. you just might run into a monitor flicker which is cured by a PC reboot. any game that is GPU heavy will require you to turn off the folding on the dedicated gaming card to play. but you can run your mining card/ secondary card while gaming since its not even touched by games (Unless you use SLI (HA!)). This assumes you have more then 2 CPU cores. most games only use one core. the rest are just sitting there.

one day, I had Black desert online running in the background in 'tray' mode, Folding at home on both GPUs and GTA 5 with average FPS all at once. because I could.
This is running an:
Kaby lake i5 3.70Ghz
32 GB ram
EVGA 1080 TI SC2 hybrid
Zodiac p106
on an EVGA z390 FTW motherboard powered with 850W.

My room was hot that day.

edit: correct name of card. (both GPUs were second hand from a miner.)

What is the PPD on the P106?

[JimboPalmer]

What is the PPD on the P106?

The P106-90 should be similar to a GTX 1050 ti, about 190k PPD (My GTX 1050 ti is low profile and lacks a 6 pin power connector, so may be lower PPD)

a P106-100 should be similaar to a GTX 1060, about 300k PPD.

[DonLuis]

I purchased one a few days ago and have a tracking number will update on how it works and it’s performance against my Gigabyte 1060

[PantherX]

...I assume that's on the smaller work units that it gets. Surely the software only sends smaller units to weaker cards and bigger ones to more powerful ones... is that the case?...

The current situation with the GPU detection is that it can only assign WUs based upon the GPU architecture, not the GPU model. There's no ability to distinguish a GTX 2080 Ti from a GTX 2060. As long as the GPU architecture is the same, it will be assigned a WU that the architecture supports. Thus, you can end up where a large WU is assigned to a low end card within the same family just because it uses the same architecture.

[DonLuis]

I just got the P106-90 in my mail. I'll be installing it in about 4hours (GTX 660 Work unit finishes). It has the PCB a "bowing" a little bit. Hopefully, it will work.

[DonLuis]

...I assume that's on the smaller work units that it gets. Surely the software only sends smaller units to weaker cards and bigger ones to more powerful ones... is that the case?...

The current situation with the GPU detection is that it can only assign WUs based upon the GPU architecture, not the GPU model. There's no ability to distinguish a GTX 2080 Ti from a GTX 2060. As long as the GPU architecture is the same, it will be assigned a WU that the architecture supports. Thus, you can end up where a large WU is assigned to a low end card within the same family just because it uses the same architecture.

Noticed that on my folding rig, I have a GTX 1060 and GTX 660, Some times the 660 gets very large WU and it gets penalized for taking a long time to finish the job. It be good if the WU creators make smaller units, should be in theory the same result with fasters return times.

[kiore]

Is difficult to know what the market for 2nd hand gpu's will be, once a bargain identified the online mob will drive up the price, certainly old mininig cards may be a good opening but also likely they are a bargain because they are thrashed. Open box deals another good source, I recently bought a open box GTX1660ti for what I thought a excellent price for a new card, although wish I'd held out for a RTX2060 for a bit more power.. But such is life and last minute bargains. The 1660ti fits underneath my 1080ti and as is a shorter card doesn't interfere with the cooling of the more powerful card.

[bruce]

...Surely the software only sends smaller units to weaker cards and bigger ones to more powerful ones... is that the case?...

Surely that's what SHOULD happen, but it's still on the wish-list/enhancement list except in rather limited situations.

EDIT:
Oops. I lied. I didn't find it on list, but it is now.
https://github.com/FoldingAtHome/fah-issues/issues/1479

[ZePompom]

I just got the P106-90 in my mail. I'll be installing it in about 4hours (GTX 660 Work unit finishes). It has the PCB a "bowing" a little bit. Hopefully, it will work.

Nice! I look forward for the results

Still waiting for mine

[DonLuis]

Nice! I look forward for the results

Still waiting for mine

It was plug & play. Got detected right away not errors in the log and picked up a WU. Has been running for almost a day.

-It generates about HALF the PPD of my Gigabyte GTX 1060 6Gb. (192968 vs 423183)
-Runs at about 50C temperature where my Gigabyte runs at 77C.
-Memory is clocked by default at the same speed of my Gigabite
-Per MSI Afterburner the 2 GPUs are fully utilized.
-Core is clocked about 70Mhz less than my Gigabyte.

Running on a Gigabyte B75M (1x16PCIe at 16x and 1x16PCIe at 4x) and i3-2120T HT enabled. different box.

EDIT: Searching for a thread about people having Ethereum slow hash rate using P106. Found this comment:

https://www.reddit.com/r/EtherMining/co ... _hashrate/
"these cards has half of the shaders and compute units, they suck at computing, it seems like they were specifically designed to mine the daggerhashimoto algo"

With makes sense, since I'm getting half the PPDs. Next, I'm going to get a new KillaWatt (Wife misplaced mine... into the garbage) and see if the PPD/watt-hour cost makes them worth it.

[ZePompom]

Nice! I look forward for the results

Still waiting for mine

It was plug & play. Got detected right away not errors in the log and picked up a WU. Has been running for almost a day.

-It generates about HALF the PPD of my Gigabyte GTX 1060 6Gb. (192968 vs 423183)

[...]

EDIT: Searching for a thread about people having Ethereum slow hash rate using P106. Found this comment:

https://www.reddit.com/r/EtherMining/co ... _hashrate/
"these cards has half of the shaders and compute units, they suck at computing, it seems like they were specifically designed to mine the daggerhashimoto algo"

With makes sense, since I'm getting half the PPDs. Next, I'm going to get a new KillaWatt (Wife misplaced mine... into the garbage) and see if the PPD/watt-hour cost makes them worth it.

Aw, not cool for the half PPDs

But the P106-90 is based on GTX 1050 Ti and not GTX 1060 (the GTX 1060-like is the P106-100).

1050 Ti have 768 CUDA cores while the 1060 6GB have 1280 of them. I think that's the biggest factor explaining that difference.

If I refer to this post : https://www.overclock.net/forum/55-over ... abase.html
Your PPDs with the P106-90 are indeed on par with the 1050 Ti.



EDIT : Can I ask you a favor?

On your P106-90 can you tell me what the "BUS interface load" (In GPU-Z sensors) is, when the card is folding?

I try to figure out if a PCIe 8x 1.1 would be enough or if it would bottleneck and need a full 16x.

(My old MB have two 16x slots, but it splits it in two 8x is both are used, so it would help me to know if these 8x would be enough for an hypothetical dual P106-90 config)

Thank you in advance!

[kiore]

Usually PCIE x 4 is good enough for folding, I use a mobo with a 16 and an 8 lane which is 2 x 8 when both occupied and see almost no difference.

[ZePompom]

Usually PCIE x 4 is good enough for folding, I use a mobo with a 16 and an 8 lane which is 2 x 8 when both occupied and see almost no difference.

It's enough with/from wich gen?

Since 8x in PCIe 1.1 (my case)have similar bandwidth than a 4x in PCIe 2.0 and only 2x in PCIe 3.0.


I know my GTX 1660 uses around 25% of its 16x 3.0 for example (between 21 and 27% depending on the WU).





So theorically it would use 50% of a 16x 2.0 and 100% of a 16x 1.0 or 1.1

But for this card (MSI GTX 1660) a 8x 1.1 would provide only half the bandwidth it need to fold at full power.


I would like to do the same math with a P106-90, but I would need someone who actually have a P106-90 to give me that information.

SInce the P106-90 do less than 50% of 1660's PPD I suppose it also needs less than 50% its PCIe bandwidth though. But I would like to be sure

[PantherX]

I believe that kiore might be talking about Gen 3. Do note that OS matters too. IIRC, on Windows Gen 3 x8 is the lowest you can go without any impact but for Linux, it is Gen 3 x2. You can read more about it here:
Brief recent discussion: viewtopic.php?f=38&t=34932
Lengthy previous discussion: viewtopic.php?f=38&t=28847

[ZePompom]

I believe that kiore might be talking about Gen 3. Do note that OS matters too. IIRC, on Windows Gen 3 x8 is the lowest you can go without any impact but for Linux, it is Gen 3 x2. You can read more about it here:
Brief recent discussion: viewtopic.php?f=38&t=34932
Lengthy previous discussion: viewtopic.php?f=38&t=28847

Thanks.

Just try to figure out is 8x PCIe 1.1 is enough, so "4x PCIe 3.0 is enough" doesn't help me, sadly

I do plan to use Lubuntu on my folding machine. I already installed it and made some tests. For SSH, remote desktop and FAH control remote connection.