Encryption Projects as SU group

cayoenrique

Senior Member
Messages
489
@ViaHussun & C0der

From what you say it seems there is no clinfo.exe by default. Listen I need a few hours to finish the tutorials. Then we will create a project to build clinfo.

I got trouble my self doing
Code:
https://github.com/Oblomov/clinfo
This one compile and run easier because it show just a few things
Code:
https://github.com/hnw/clinfo

Now be aware that this apps do not do opencl, they do not have a opencl kernel to send to gpu. They only query info from gpu.

Wait until I finish tutorials
 

sdrfgs

Registered
Messages
116
I see some posts here about comparing cudabiss results. It needs to be standardized with some basic comparitive things such as
CPU model MHZ -normal or overclocked speed
ram size and type e.g ddr3/ddr4/ddr5
ram speed eg 2133mhz 3200 mhz etc
Gpu make and model (some may have the same model number such as 3060 TI etc but not all are clocked at the same MHZ some have higher boost clocks etc

GPU Driver version, and has it been deebloated installed with nvclean?

But most importantly
Nobody said which version of Cudabiss they are using! or their input.txt parameters

From the cudabiss release Cudabiss51264.exe is the one all should be using as it is fastest ,

dated in properties details
size 1.23MB Date Modified 14/03/2011 10:16p.m

I will have more to add in future postings
 

moonbase

VIP
Donating Member
Messages
611
I see some posts here about comparing cudabiss results. It needs to be standardized with some basic comparitive things such as
CPU model MHZ -normal or overclocked speed
ram size and type e.g ddr3/ddr4/ddr5
ram speed eg 2133mhz 3200 mhz etc
Gpu make and model (some may have the same model number such as 3060 TI etc but not all are clocked at the same MHZ some have higher boost clocks etc

GPU Driver version, and has it been deebloated installed with nvclean?

But most importantly
Nobody said which version of Cudabiss they are using! or their input.txt parameters

From the cudabiss release Cudabiss51264.exe is the one all should be using as it is fastest ,

dated in properties details
size 1.23MB Date Modified 14/03/2011 10:16p.m

I will have more to add in future postings


Good points.
My cudabiss results posted earlier in this topic were obtained from various setups with the following specs:

Latest version at the time of the GPU NVIDIA driver installed using NVclean Install after removal of all previous NVIDIA drivers.
PC's debloated from a lot of Windows 10 default junk using a Windows 10 debloat tool.
Latest version of cudabiss.
Mode 1, full mode, no filtering.
CPU's are i7-5930k or i7-5960x always at stock speeds.
RAM size always 16gb of DDR4, generally at 3000MHz.
GPU 2080 Ti (low end card, only used if higher spec cards are allocated to other jobs).
GPU 3080 TI's (Founders Edition or MSI Suprim X which is a high boost clock GPU).
GPU 3090's (Gigabyte Aorus Extreme, high boost clock GPU).
GPU 4090 awaiting purchase and testing.
All GPU's run at default speeds, no overclocking with apps like Afterburner.
 
Last edited:

sdrfgs

Registered
Messages
116
Good points.
My cudabiss results posted earlier in this topic were obtained from various setups with the following specs:

Latest version at the time of the GPU NVIDIA driver installed using NVclean Install after removal of all previous NVIDIA drivers.
PC's debloated from a lot of Windows 10 default junk using a Windows 10 debloat tool.
Latest version of cudabiss.
Mode 1, full mode, no filtering.
CPU's are i7-5930k or i7-5960x always at stock speeds.
RAM size always 16gb of DDR4, generally at 3000MHz.
GPU 2080 Ti (low end card, only used if higher spec cards are allocated to other jobs).
GPU 3080 TI's (Founders Edition or MSI Suprim X which is a high boost clock GPU).
GPU 3090's (Gigabyte Aorus Extreme, high boost clock GPU).
GPU 4090 awaiting purchase and testing.
All GPU's run at default speeds, no overclocking with apps like Afterburner.
Yes I know you

And you know me also

Im aware of your setups
 

cayoenrique

Senior Member
Messages
489
@sdrfgs

I have not ignore your post. But I have problem with sites using cloudflare. A soon as I can download I will comment. I hope to see sources that we can use in the discussion of Opencl.
 

cayoenrique

Senior Member
Messages
489
@sdrfgs
I seen that. Now it will be used as a different kernel for encryption on Block Cypher. Normally we used the code Colibry public with his source code for RBT brute-force.
I was hoping you will produce a full decrypt with Block Cypher and Stream Cypher. This is what we need to mimic CUDABISS.

But in any case thanks for the offer. Very appreciated. Regards speed, I already said that a little or not faster is unimportant to me. What surprise me ist that is done in 1 day with best GPU.
But you are correct little things can speed up 10% and using GPU on dumb servers without any other programs can ensure that all resources are dedicated to brute-force.
 

sdrfgs

Registered
Messages
116
Well i'm not a programmer, just a user of various software with many years of computer experience. Ive done quite a lot of testing of cudabiss and tested how to speed it up.

The main issues are that its in efficient made for old GTX 560 and below card. You can run nvprofiler on it and see all the weakness such as how poorly it uses the gpu memory etc.
Thats why on any modern gpu with cudabiss you need to maximise performance by running multiple instances at the same time.



Everyone here should be running version Cudabiss51264 as listed in my other post

from the original instructions

"Cudabiss22464 and Cudabiss44864 are intended for GTX460/465/470
Cudabiss24064 and Cudabiss48064 are intended for GTX480/570
Cudabiss25664 and Cudabiss51264 are intended for the GTX560"

just for reference The GTX 560 had just 336 cuda cores and 1 GB ram

A few tips

I tested Cudabiss in SLI mode using 2 gtx 980 at stock, no improvement whatsoever as expected, not supported even if gpu is forced to sli mode. Yes it will run but at no increased speeds. Much better to run 2 cards independently with multiple instances. (So don't waste time experiemnting with SLI there is no advantages to be gained)

Cudabiss itself uses very little CPU, around %7 cpu usage on my I7 5820k running 1 copy. It doesnt take advantage of multi threading. I have played around with pinning individual copys to specific cpu cores with higher priority. However theres little advantage to be gained unless your pc is running a lot of other tasks at the same time.

Memory bandwidth is important, to max out gpu performance, e.g going from 2133 MHZ ram to 3200 mhz ram will gain you performance. So enable XMP if you have it, also check out uncore mode in your bios (but thats another topic)

Gpu overclocking can gain you around %10 speed boost, Use msi afterburner, a basic guide below

don't touch any voltage options!

set power limits slider to maximum
set core clock anywhere between +150-250 mhz (this will entirely depend on your gpu and needs extensive testing)
leave memory clock alone (though this can add a small boost)
set gpu fan speed to %100 manual

For best performance of your GPU it should be installed using the NVCLEAN tool to remove all the nvidia bloatware

Also make sure you are runing your gpu in MSI mode
https://www.reddit.com/r/OptimizedGaming/comments/107blhi
 

cayoenrique

Senior Member
Messages
489
@sdrfgs

I have no knowledge or experience of NVIDIA. I only have AMD. But what you have posted I am pretty sure is correct. Seems similar to AMD.
And I am on the conquest to build OCLBISS to mimic what CUDABISS do.
 

sdrfgs

Registered
Messages
116
I don't have any results of users using AMD cpus with Nvidia cards for Cudabiss as to whether they perform better or worse or the same. Most likely come down to ddifferences in memory bandwidth. As Ryzen is different in that reguardl, there might be situations wheere older gpus are bottlenecked
 

cayoenrique

Senior Member
Messages
489
NO no CPU .

We mean AMD GPUs. Old old stuff works with less core and core speed do works. Clearly slower. But is the learning process what it count.
 

sdrfgs

Registered
Messages
116
To start with Cudabiss, you can try to decompile using ghidra since we dont have a source code


ai programming would be another option if someone can get some code running.

I think cudabiss was made in visualbasic if you open the original exe a lot of strings can be read
 

cayoenrique

Senior Member
Messages
489
@sdrfgs

You are my men then. You have prove to be of great value.
Listen. I do not need to get sources for the main app visualbasic or what ever.

What we need is the CUDA kernels.
Lets take for example
Code:
http://colibri.bplaced.net/CSA-Rainbow-Table-Tool_V1.03.zip

after unzip twice (sources) I see:
CSA-Rainbow-Table-Tool - V1.03/CSA-Rainbow-Table-Tool/cuda/cuda.cu

That is the kernel used by colibri.

I love to have the CUDA kernel of CUDABISS.
 

dvlajkovic

Well Known Member
Messages
502
It looks to me that cudabiss 51264 based on cudart64_31_9.dll has been compiled with
c:\ProgramData\NVIDIA Corporation\NVIDIA GPU Computing SDK\C\src\simpleAtomicIntrinsics\template_kernel.cu

Then some dudes, obviously with access to cudabiss source files, managed to go even further and added support to newer versions of cudart64_65_14.dll and cudart64_75_18.dll .
The new ones have been compiled with
C:/ProgramData/NVIDIA Corporation/CUDA Samples/v7.0/0_Simple/TemplateCB/template.cu

I have said it takes almost 24 hrs to cover full range on RTX4090 because I was using public cudabiss 51264 based on cudart64_31_9.dll .
Updated cudabiss versions run full range in 11-12 hrs based on cudabiss internal estimation using the same graphics card.
Haven't tried them myself as yet 'cause these days we have temp around 38°C which melts down everything, but will try as soon as the temp drops down.

Anyway, here are cuda kernels that come with the latest drivers for win10 64-bit RTX4090 (version 15, 16, 17) - download
Path to reach them is: \NVIDIA GPU Computing Toolkit\v12.2\extras\visual_studio_integration\CudaProjectVsWizards\15\ProjectTemplates\kernel.cu
Hope it helps to develop our OpenCL project.
 

moonbase

VIP
Donating Member
Messages
611
Then some dudes, obviously with access to cudabiss source files, managed to go even further and added support to newer versions of cudart64_65_14.dll and cudart64_75_18.dll .
The new ones have been compiled with
C:/ProgramData/NVIDIA Corporation/CUDA Samples/v7.0/0_Simple/TemplateCB/template.cu

I have said it takes almost 24 hrs to cover full range on RTX4090 because I was using public cudabiss 51264 based on cudart64_31_9.dll .
Updated cudabiss versions run full range in 11-12 hrs based on cudabiss internal estimation using the same graphics card.
Haven't tried them myself as yet 'cause these days we have temp around 38°C which melts down everything, but will try as soon as the temp drops down.


Does this mean that CudaBiss 51264 can have its speed increased by simply updating the cudart dll file version in the app folder?
Or does it require an internal code change to CudaBiss 51264 to get it to reference a more recent cudart dll file?

If the speed to run a full range of CudaBiss can be reduced from 24 hours to 11 or 12 hours it is a doubling of its speed.
 
Top