Will be great to see. My watercooled gtx 780 wont get past 90% usage if I let it overclock as it has no thernal limits only power. But at 250w I can give it a lot more power compared to the 300w of a 7950.

And with 3gb of memory, hardly any of it is used. Say 300mb and I can not figure out how to use the benchmark as it can give me hash rates of 900khash/s or even more. But I dont know how to set the options or where it gets such high values.

My cudaminer in auto tune has been freaking out lately. Giving rates as low as 6khash/s when normally it hits 500 khash/s on an auto. The 900khash/s rate is using the basic benchmark command on 64 bit.

32 bit gives me better hash rated though

I didnt think you could run both atall. Or is that just that you can not run amd and nvidia dedicated gpus together?

@bigjme I knew that my motherboard (M4A88T-M) doesn't support connecting monitors to the onboard and the dedicated graphics card at the same time, but since there are pcie raisers and converters and all that stuff, virtually all you need is to power your graphics card to mine with it.
I read somewhere that it's possible to mine using both NVIDIA and ATI/AMD cards on the same board.



Yep, in time when I'll have coins I'll thank this.

4GB GT 630 for $30 : wow! good price. I have yet to enable the Fermi kernel for scrypt-jane though.

Tesla kernels don't need to explicitly enable a texture for cached reading, as they automatically pull their data through this cache (look up what the __ldg intrinsics do in the latest CUDA programming guide)

I might try to figure out a way to chop the scrypt-jane kernels into a series of smaller kernel launches, which may make make it less taxing on the display and also allowing the use of interactive mode again.

Titan kernels are now scrypt-jane enabled! I get 3.2 kHash/s on GTX 780Ti using -l T7x3 now. And power use is cut in half compared to LTC mining. What a pity the 780Ti doesn't have 6 Gigs of RAM, or I could use -l T14x3, doubling the speed. Someone should try this launch config with the 6 GB Geforce Titan models though. Could yield some 6 kHash/s.

I also have a crazy idea that would basically remove the memory limitations for scrypt-jane mining. It requires joining the A and B kernels into a single kernel again and re-using the scratchpad memory on the GPU. So instead of giving each thread a unique 4 MB scratchpad, we may be able to reuse the same scratchpad memory for all non-concurrently executed thread blocks. I think this is a similar concept that the "intensity" parameter on the ATI cards is controlling when running cgminer. Unfortunately this idea might be incompatible with the texture cache, as this cache does not guarantee read/write coherency within a single kernel invocation. But hey, it could get my 780Ti's to 6 kHash/s...maybe.
EDIT: okay, I made a mistake in my thoughts here. with so few thread blocks running on the GPU, ALL of them would be executing concurrently. And hence the memory reuse concept falls flat.

Christian

3.2 kHash/s?  Is that a typo?  Or perhaps that's the increase you're seeing now?  Sorry, I'm confused.

So im best to ignore the high hash rate benchmarks?
For now I have a random value set because of auto tune messing up so badly.

I tried T11x13 on my 780. This is said to give 880. Gives around 440, yet I am able to get 503 from another config but I can not figure out which to use as the benchmark values are so off

I guess you have overclocks. Autotune doesn't seem to like that.
Anyway, cudaminer by default uses scrypt, that's what you're using, scrypt-jane is something else.


Some scrypt-jane hashrates for comparison:
i5 3570k - 0.41 kH/s
i7-3930K - 1.20 kH/s
FX-8350 - 0.57 kH/s

5770 - 1.0 kH/s
7950 - 1.3-1.5 kH/s
R7 250 - 1.44 kH/s

I was just thinking, for the scrypt-jane. Would there be a way to force some system memory to be used? Yes it is much slower then gpu memory, but if you can give it 16gb, it can do over 5 times that of 3gb gpu at around a 3rd of the speed. So in theory doing more for the time

Compilation from git is relatively painless on Linux. There should be some guides around on the Internet, because some people have been doing this on rented amazon EC2 instances previously.

I just had a quick look and gddr5 is twice as fast as system ddr3.

But ddr3 is better for smaller files. 64mb or smaller. With jane being 8mb ddr3 system memory would be ideal. Then a memory war would start. If 3gb of gpu memory can do 3.2kh, 12gb of system should do double.

3gb gpu is rougly 6gb system.
Those systems with 32gb should be able to pass 26gb of system memory over. That should be the same speed as 13gb of gpu memory. Might be worth a look if it can be done. Im no software developer so sorry if this is all waffle