Encryption Projects as SU group

[moonbase]

Compiled using Vis Studio 22 - latest version.
I did not include the ability for pre RTX3xxx cards to run - saves on file size, and setup & tweaks would probably not run on those anyway.
I increased the number of threads, and limited the number of registers used. Nothing really crazy, just trying to squeeze a little more out of the cards.

3.8bkps 1 instance on 3070

Thank you for the info.
I will test the new versions on an RTX 4090.

 

[sdrfgs]

date tested 13/02/24

new version 30xx up

I7 7700 3.6GHZ overclock -> 4.1GHZ
Win 10 10.0.019045
16 GB
3060 TI 8GB

input file

000000000000
010000000000
47420097FBD6514CF8A45710DEDC2BA8
47420090ACA55EC7AD6798DA9939A615
4742009C34C7B45CBFF0718AEDBE387B
1

connected remotely via vnc (please note this gives lower speed due to some gpu usage in streaming the desktop display)

scrolling version used

1 copy 3029.112320 keys/s (previous 64 version 2718.562816 keys/s)
cpu usage 13.8% - 14.8% taskmanager process

2 copys
copy 1 2034.757504 keys/s previous copy1 1831.798912
copy 2 2034.998400 keys/s previous copy2 1839.892480
cpu usage 14.5% - 15.% each taskmanager process
4069.755904 combined

3 copys
copy 1 1359579392 keys/s previous 64 version copy 1 1268.473728
copy 2 1363957888 keys/s previous 64 version copy 2 1261.937536
copy 3 1348746752 keys/s previous 64 version copy 3 1259.185920
cpu usage 13.5% - 15.% each taskmanager process, 95.9MB per copy
4072.284032 combined

overall about a %10 speed increase over the previous "64" version

 

[sdrfgs]

@Lak7
Does this make any use or advantage of the Re-Size BAR function?

 

[Lak7]

@Lak7
Does this make any use or advantage of the Re-Size BAR function?

I would be curious if someone did enable it and see if there is any difference.
I don't have a platform that supports the Resize BAR function. Looks like intel 10th gen and newer AND motherboard that supports it.

 

[moonbase]

I would be curious if someone did enable it and see if there is any difference.
I don't have a platform that supports the Resize BAR function. Looks like intel 10th gen and newer AND motherboard that supports it.

I have tested the previous versions with Resizeable Bar turned on and off and it made very little difference to speed.

 

[sdrfgs]

Yes but the newer recompiled release may do something?

I dont have resizeable bar support on my machine. to test it

The other thing we don't know, is if AMD cpus / chipset work better.

Speaking of that , Zluda got released, this gives cuda compatibility on non nvidia cards!

GitHub - vosen/ZLUDA: CUDA on AMD GPUs

CUDA on AMD GPUs. Contribute to vosen/ZLUDA development by creating an account on GitHub.

github.com

 

[moonbase]

Latest scrolling version v12_3_VS22 runs slower for me on an RTX4090 compared to v12_3_64.

 

[moonbase]

I have tested both the scrolling and non scrolling versions, each with Resizeable Bar on and off.
All four tests with v12_3_VS22 run slower than the same tests with v12_3_64.

Tests so far from other testers with 3nnn series cards with v12_3_VS22 have yielded speed increases compared to v12_3_64.
The drop in speed I am getting with the RTX 4090 might possibly be the way in which CubasBISS v12_3_VS22 addresses the architecture of an RTX 4090 GPU?

It would be useful if some of the other downloaders off the app could post their test results to see if there is a pattern ?

 

[Lak7]

The drop in speed I am getting with the RTX 4090 might possibly be the way in which CubasBISS v12_3_VS22 addresses the architecture of an RTX 4090 GPU?

I would agree with that.
Need a real coder to coalesce the memory calls.

 

[moonbase]

I would agree with that.
Need a real coder to coalesce the memory calls.

Is there a fix that you can implement for use with 4090 cards?

 

[Lak7]

Is there a fix that you can implement for use with 4090 cards?

If the previous version is faster, then that's the one to use. It was just slower on the 3xxx cards, so I tweaked it for their chip structure.

 

[sdrfgs]

another fast test on the above 3060TI I7 machine

14/02/24

1 copy NOT connected via vnc, slightly faster, all i did was srtart it thenn connect back later for the result


input file

000000000000
FFFF00000000
47420097FBD6514CF8A45710DEDC2BA8
47420090ACA55EC7AD6798DA9939A615
4742009C34C7B45CBFF0718AEDBE387B
1
1

3138686464 keys/s

 

[sdrfgs]

Is there a fix that you can implement for use with 4090 cards?

Which Nvidia driver did you use? there is chance that some nvidia settings, upset things
On the 3060TI machine im using

driver 526.98 DCH
the 3060 ti is the LHR model, I think they disable the LHR limits in release 522.25

 

[moonbase]

Which Nvidia driver did you use? there is chance that some nvidia settings, upset things
On the 3060TI machine im using

driver 526.98 DCH
the 3060 ti is the LHR model, I think they disable the LHR limits in release 522.25

The 3090 and 4090 series of cards never had LHR implemented as a matter of policy, they were the top of the range.

 

[sdrfgs]

I know that I was just adding it for reference,

Which driver are you testing with?

 

[moonbase]

Which driver are you testing with?

NVCleanstall_NVIDIA_546.33_x64_dch_Desktop_Setup

 

[sdrfgs]

How much slower is it than the previous?

It definetly works much better on the 3060ti compared to the previous release

What about the settings here?

https://imgur.com/a/JK1gESV

is optimize compute performance on or off? and prefer maximum performance?

 

[moonbase]

How much slower is it than the previous?

is optimize compute performance on or off? and prefer maximum performance?

NVIDIA settings and entire PC configuration and settings same as was previously, no change.
Only variable is the version of CudaBISS, v12_3_VS22 runs slower than v12_3_64 on a 4090 GPU.

No test results yet from the other downloaders since the newer RTX versions were made available.
Seems like it is only the two of us publishing test results?

 

[Lak7]

NVIDIA settings and entire PC configuration and settings same as was previously, no change.
Only variable is the version of CudaBISS, v12_3_VS22 runs slower than v12_3_64 on a 4090 GPU.

No test results yet from the other downloaders since the newer RTX versions were made available.
Seems like it is only the two of us publishing test results?

After you guys reported, there were about 15 downloads.
Was hoping someone was interested in trying to coalesce the memory calls and bring the code into the "modern" age - it's 15 years old now LOL

 

[Lak7]

NVCleanstall_NVIDIA_546.33_x64_dch_Desktop_Setup

It probably does not matter which driver you use.
I'm running 551.23