Testing GeForce GTX 750 Ti Maxwell for Alt Crypto Mining with CUDAminer

19 Feb
2014

nvidia-geforce-750-ti-reference-board

We have already ran some tests of the new GeForce GTX 750 Ti GPU based on the new Nvidia Maxwell architecture for mining scrypt-based crypto currencies such as LTC and DOGE with the help of CUDAminer (up to 265 KHS). The results it is giving, along with the low power usage of the card are making it an interesting option for people that are interested in mining crypto currencies with Nvidia GPUs. The GeForce GTX 750 Ti also shows good overclocking potential for even higher performance (up to about 300 KH/s) on a reference design board. After looking at some non-reference design boards such as ASUS GTX750TI-OC-2GD5 that has additional PCI-E power connector and much better cooling we considered that even higher performance could be achieved, however after checking some reviews of these cards it seems that the Power Target maximum on these cards is still 100%. This simply means that even though the card is clearly designed to handle serious overclocking, the power limiter will not allow you to go beyond the TDP rating of 60W, so as a result it seems that these cards are not overclocking much better than the standard reference boards.

yacoin-crypto-mining-maxell-cudaminer

We have also tested the performance that the new GeForce GTX 750 Ti Maxwell cards can offer in some other crypto currencies supported by CUDAminer that do not use Scrypt algorithm, but instead rely on different approaches – SHA-3 (Keccak), Scrypt-jane, Adaptive N factor Scrypt. The performance we got with them was also quite good, though overclocking the card did not help as much these as with Scrypt. Below you can find out what our results were with the different crypto coins we have tried that use the above mentioned alternative algorithms. Note that we are still trying out different configurations in order to get the maximum performance out of the GTX 750 Ti for these alternative crypto currencies, so any suggestions for settings to try with are welcome. As suggested we overclocked only the GPU for the latest results to the maximum of +135 MHz, without clocking the video memory as apparently for these algorithms it does not make sense (downclocking it down with -502 MHz does not seem to affect performance significantly). We are going to update the results if/when we get a better performance with the settings we find to work best for the specific algorithm and crypto coin combination. You are also welcome to request us to test with other coins that are supported by the CUDAminer software.


YACoin Scrypt-jane mining performance with:

cudaminer.exe -a scrypt-jane -o stratum+tcp://yac.coinmine.pl:9088 -u yourworker.1 -p password -L 4

Mining YACoin that uses Scrypt-jane has proven a bit tricky with CUDAminer as apparently it did not want to work properly with our custom settings for the kernel settings we used for Scrypt, the only way to make it work was to use the autotune functionality of the miner, otherwise we were getting errors. This means that we could not try our best performing kernel configuration T5x24 on Scrypt-jane here. After trying some suggestions for CUDAminer and YACoin and adding lookup gap with -L 4 (the best performing lookup gap setting for GTX 750 Ti) we got around 1.7 KHS out of the card. Overclocking the GTX 750 Ti to +150 MHz GPU gets us up to about 1.78 KHS of hashrate for mining YACoin (T74x1 automatically selected, so far has the best performance, as setting manually kernel does not appear to work properly).


VTC Adaptive N factor Scrypt mining performance with:

cudaminer.exe -a scrypt:2048 -o stratum+tcp://stratum.vertco.in:8080 -u yourworker.1 -p password -i 0 -l T5x24 -C 1 -H 2

For VTC mining that uses Adaptive N factor Scrypt algorithm with the T5x24 configuration and manual settings got us about 125-129 KHS hashrate with the GTX 750 Ti running at stock frequencies. After overclocking the Maxwell to +135 MHz GPU the performance increase has boosted our hashrate to about 135-137 KHS.


MaxCoin Keccak mining performance with:

cudaminer.exe -a keccak -o stratum+tcp://maxpool.1gh.com:17333 -u mGeS2LBCTJvz1Mb4Hfzt5fU1C9C1tYrxuY -p password -i 0 -l K1000x24 -C 1 -H 0

For MaxCoin mining that uses Keccak (SHA-3) algorithm we could also use the K1000x24 kernel configuration and manual settings and that got us about 62000 KHS hashrate with the GTX 750 Ti running at default clocks. Overclocking the GTX 750 Ti to +135 MHz GPU has increased a bit the hashrate to about 72600 KHS.

Meanwhile the author of CUDAminer has released an updated version of the miner software, though it is not yet specially optimized for the new Maxwell architecture you might want to update to it as it has some fixes and improvements. Furthermore there is a mention about upcoming BlakeCoin (BLC) support, so it will be interesting to see what kind of performance the GTX 750 Ti will be able to offer for BLC mining (Blake-256 algorithm) where the typical hashrate is about 3x the one you get when Scrypt mining.



Other Similar Publications:

8 Responses to Testing GeForce GTX 750 Ti Maxwell for Alt Crypto Mining with CUDAminer

Christian Buchner

February 19th, 2014 at 17:34

unfortunately this was not done well.

Yacoin needs lookup gap -L 2, 3, or 4 with mid range cards for good performance. High end cards need 6 or 7 even. Autotune for each lookup gap value.

Keccak needs huge launch configurations (like -l K1024x32) and the K kernel (T sucks for Keccak).

A

February 19th, 2014 at 19:34

Your VTC hash rate for 1000khs seems awfully off. Even the 290x can’t get past 475khs when mining vertcoin.
can you please redo that test and make sure you don’t get any rejected work units and that the pool actually shows the same 1000khs hash rate you reported?

Marc Hayes

February 19th, 2014 at 19:45

Please seriously consider NOT using +630 Mem clock. Very very rarely does Mem clock actually affect Scrypt, Scrypt-Jane or Keccak speeds and end up eatting into the total TDP of the card for no reason. There is zero Memory bottleneck and the only benefit of increased Mem clock is a reduction in memory latency, but the gains here are counteracted more heavily by the TDP limiting the core clock back down (the is GPU Boost 2.0 behavior 1.0)

Consider going negative on the Mem clock instead, this allows more voltage to the Core clock allowing higher speeds overall. And if no extra clock is achieved the total TDP of the card can go down a good 2-5% on Kepler’s at the -502 mem clock.

That said Maxwell may be different in this behavior… Mem might be needed… But please re-test this with negative Mem clocks.

zach

February 20th, 2014 at 00:54

Is that a misprint on the VTC numbers? 950-980 KHS is huge??

admin

February 20th, 2014 at 01:19

Actual VTC hashrate is 125-129 KHS without overclock, and about 135-137 KHS with OC.

sourlain

March 4th, 2014 at 09:43

I can safely say you get your max boost not from clock but from raising mem clock … I have been doing this for like 3 days now…

d4wn

March 12th, 2014 at 09:49

Your keccak numbers are way off. I was able to get 150000 kh/s with -L 128 -l T1024x16 And that is just based off my old fermi numbers for my gtx580 which got 90000 kh/s. These cards are a huge improvement for keccak.

admin

March 12th, 2014 at 10:53

d4wn, these results were achieved with an older version of cudaminer, the latest one cudaminer-2014-02-28.zip introduces speed-up for YAC (compute 3.0 or later) and keccak (compute 3.5 or later), so it is normal to get better results by using it…

Leave a Reply

Your email address will not be published. Required fields are marked *

top