Topic: Bounty for X11 NVidia miner (Read 8516 times)

Hello all-

This thread is a little old, but I have both an AMD R7 265 mining rig and an nVidia 750ti mining rig.  Christian Buchner's "ccminer" has been developed and forked/modified by a crew of nVidia enthusiasts. Cudaminer still mines scrypt and scrypt variants, but the ccminer has been adapted to a number of algorithms.

When mining X11 algorithm with ccminer, my 5 card nVidia 750ti rig gets 2.600 MH/s per card, 13.000 MH/s per accepted share (yay!!!).  My 4 card AMD R7 265 on sph-sgminer by DJM34 gets an average of 1.340 MH/s per card, and 5.300 MH/s for the rig.  Granted, 5 cards versus 4 cards is not fair,  but the R7 265 gets only half the average hashing speed on a per-card comparison.

Both run quietly and with very little heat.

If anyone has any suggestions on how to tweak the R7 265 card for X11 algorithm, please post.  I have been told that xIntensity and Intensity have no affect on this algorithm.       -scryptr

I believe Christian from cudaminer is working on it currently. They were only a few algo's away from having it working.

Apparently they'll have quite the edge because of how efficient their miner is with groestl.

Is someone working on this?

Elbandi

Coming soon...

A cudaminer for Hirocoin would attract even more miners.

offtopic, but the "outstanding" 750Ti power savings vs the 280x or 7950, are a fallacy. Yes, in the end, the 750Ti still wins on hash/W, but it's misguided to tune the Tahiti card to max out hash speed, set whatever voltage you need to achieve it, and then complain it is not efficient!

If you want low power, you tune for low power! (and low heat and noise). You undervolt to whatever low voltage, e.g. 0.85V(?) to 0.95V. This allows ranges dependent on leakage, VRM and ASIC quality, but let's say 800~1030 Mhz core, 500~700 Kh/s at 110~150W. That's scrypt. If you mine others, you can cut up to 40W from the vram power consumption.

Not to speak of the lower density, the always present 100+W base platform overhead, the upfront cost of extra components per Mh/s, the higher cost of the 750Ti per hash, it's quirks when using risers, etc...

ontopic: of course it would be great for Nvidians to mine X11. I occasionally use my gaming/Linux GTX760 for mining, too.

User needs are subjective though. I have only had one Nvidia card in my main rig over the past 14 years, and that was the 8800GT, as it was great value for money at the time.

All my other cards have been AMD/ATI as they offered the best price/performance ratio in my price range. I currently have crossfired 7950s in my main rig.

So whilst AMD are winning the price/performance crown, you are forgetting that that crown isn't the be all and end all.

My cards do a great job mining, but are sitting in the mid-to-high 60s doing so whilst being fairly noisy. Heating my flat with them has been great over the winter, but it just won't be sustainable during the summer.

This is where the 750ti power efficiencies come in to play. Yeah, lower hash means it takes longer to recoup my initial cost, but after that, my earnings are actually better due to the significantly lower power usage. The fact that they are running in the mid-to-low 50s whilst being inaudible of my general livingroom noise levels is a MASSIVE plus too.

If someone is happy to endure a slower ROI for the benefit of lower power usage, noise and heat, then the 750tis are the best choice, and that is why some of us would like support for Nvidia to be extended.

280x is the ideal card for mining with 7970 being the close runner up (only because of loew resale value, otherwise hashrate is the same).

290x is not good for mining, the electrical use does not justify, nevermind the cost. 290 maybe ok but not as efficient as 280x for the hashrate.

I have used nVidia and AMD both for a very long time. I have had AMD CPUs and ATI graphic cards since early 90s.

I can tell you I would never use Intel + nVidia combo over AMD + AMD combo for any PC, gaming or mining.


You also mentioned putting two cards in a PC, that will make your cards very loud. Get a $20 box fan and your CARD will get much quiter. The poor thing is HOT. Your fan speeds should not get over 60%, if they do you need better air blowing at it or your ambient temperature is too high. If your room temp is over 85F or you are leaving it outside and outside temp is over 85F, you should simply SHUT OFF THE RIG, period.

http://www.amazon.com/gp/product/B00002ND67/ref=as_li_ss_tl?ie=UTF8&camp=1789&creative=390957&creativeASIN=B00002ND67&linkCode=as2&tag=subbicom-20

This year was the first time I ever bought Radeon cards. 2 R290x. I bought it because they are better for mineing.

However I somewhat regret it.

It was way more difficult to get them running right and the standard R290x is sort of a cheat package in my opinion because it doesn't runs as advertised ..they have a serious termal problem and their fan is too loud and too noisy and their GPU engine clock turtles down.

Before you say yeah it's the wrong card..should have taken a saphier etc ... true.. but I have an Nvidia690 which has a DUAL GPU and a cage fan too.
It looks very much the same from outside.

The Nvidia runs at max power without turteling down the GPU clock and it is SILENT and just performs 24/7 day in day out. No trouble at all.
And on top of that cbuchners cuda miner just run as it should. No strage and long time trying to figureing out of those cgminer settings.

It took me days to get the right cgminer config (yeah I am useing sgminer too) and it took me days to build a cooling system for those 2 cards in my PC that is working and is silent enough and I still can run the CPU at 100% too.

Ok once all is set up it runs. Just a question how long till the card will burn out.. as I read this is a common problem of those radeon cards.

BUT the real let down still followed: I figured out why I never wanted to buy a radeon card: Blenders cycle render! Yeah I forgot about it.

Simple put Blender is a high end free ware animation tool and their render is fast with Nvidia and cuda but does not work with OpenCl and Radeon

As I understand it it is not becuase they don't want but because of AMD's radeon drivers and OpenCL limitations!

The next generation of Nvidia cards comming out very soon and is a lot more energy efficient and of course increase performance.

It might be very significant for the second half of this year when it comes to mineing. We might see a surprise and a rethinking of Radeon vs Nvidia for that half year..who knows.

I sure will go back to Nvidia. Less problems less noise and can be used for more than just mineing.

nVidia has NEVER been faster or better at the same price point compared to AMD, NEVER, neither has Intel, again at the same PRICE POINT. The only thing AMD has continuously failed is the Catalyst drivers which they can never get right.

Radeon R7 265 2GB is the direct competitor to the 750ti You are comparing it to 7950 which is not a very good card for mining plus its 2 years old already. Compare to 280x at least thats 10 months old:

The R7 265 costs $150 gets 445 kh/s and uses 150W
nVidia 750ti also costs $150 gets 300 kh/s and uses 38-60W depending on which model you use

Because:

A) My 750tis mine at 50% hash of my 7950s for 25% of the power.
B) Summer is coming, I'd rather not have my 7950s heating up my apartment.
C) Decentralization, be it individual hash, pool hash, or hardware adoption (remember how AMD prices skyrocketed as a result of mining?) is better than a monopoly.

You should never mine with a laptop. You will make 40 cents in 6 months while spending $50 in electricity and at the end your laptop WILL FRY.

Here is a better solution and saves you your 0.8BTC.

Why dont you sell the POS nVidia cards on eBay and get a couple of monster R9 290s? Better speed for gaming, better scrypt mining and better resale value (AMD Graphics cards FLY off the shelves. Try putting on on Craigslist at the right price, you will get tons of messages).

It is true that MS does not support inline assembly on their x64 compilers, but it is still possible to externally link assembly code. I've done that a few months ago using masm and i was able to compile the same program using jwasm in linux.

That said, why not use OpenCL?

Can my nvidia card on my notebook mine some Hiro ?   

I don't think you can do it.

Proof me wrong

bump

stuff is haaaaard man

and if your a coder you know it goes in waves.. you get into it balls deep then need a break for a while
and i have been on a break for some time lol
the most appropriate person to do it would be the cudamniner dev he has the specific experience like no other.
like what i mean is it would be FAR harder for me to do what need to be done that it would be for him he has already learned a lot i have not Cuda wise.
i'd like to see it done though..

code wise a major issue has been for me at least before is multi languages used combined with multi platform support..
for example.. i looked into a making an nvidia GPU or CPU miner mod etc before for Quark but much of the hashing code used is often passed around online
in the form of optimized ASM and i always use 64bit builds for miners and the miners often have 64bit asm hashing code
so.. when i compile it on MingW x64 on windows that uses GCC it works fine but when i try and load that code up in Visual Studio i am stone walled
MS says that you can't do that with x64 anymore and you have to use what is called Intrinsics now and that would mean knowing the asm and the algo written in asm
enough to port it into Intrinsics format.. and yeah i know asm but not that well lol
My skill with ASM is loading up a program in OllyDBG or IDA and patching some jumps or calls lol
So just that one aspect winds up snow balling into a train of bullshit i have experienced first hand a LOT lol

it is complicated and VERY hard stuff to deal with requiring a wide range of skills.

needed..

x86 + x64 instruction set info
optimized cpu instruction set info for all platforms
c/c++ skill
asm skill
mingw / gcc compiler info
visual studio x86 + x64 compiler info
Microsoft x64 Intrinsics (an asm like optimization language for ms compilers)
knowledge of hashing algo's (not the easiest stuff to learn about)
Cuda programming

LOL i means this is a lot of advanced stuff to know about..

I briefly checked.

I think the best way is to convert those 11 algos 1:1 right out of the SPH-sgminer and integrate them into cbuchner's cudaminer.
They have to be correct in SPH-cgminer for darkcoin because it's running.

The SPH code seems very readable to me although I never pprogrammed anything in GPU. With the cuda code I have a bit more trouble.

I know I could do it if I dig into it but honestly the days I can bring up motivition for such a big work are a bit in my past.

But I hope someone will do it since I plan to buy Nvidia again as soon as the next generation is out!!

Hahah, how can we get you mega motivated ?