Pollard's kangaroo ECDLP solver - page 55.

WanderingPhilospher

full member

Activity: 1162

Merit: 237

Shooters Shoot...

Quote from: Robert_BIT on May 23, 2021, 11:22:36 AM

I am not too sure how to calculate this, if someone can help me.

We have at the Uni one NVIDIA DGX A100

It packs 5 petaFLOPS, eight A100 GPUs with 320GB of GPU memory.

Realistically, if we get permission to run it 24/7 (which I doubt lol) for the 120 key, how long would it take? Is it feasible? And more important, would the Kangaroo even work on it?

Just run them and see how long the program tells you til it solves. The estimated time is pretty close to being accurate. No way to know since I don't know the speed and grid size of the A100s.

WanderingPhilospher

full member

Activity: 1162

Merit: 237

Shooters Shoot...

Quote from: NotATether on May 23, 2021, 12:36:20 PM

Quote from: Robert_BIT on May 23, 2021, 11:22:36 AM

I am not too sure how to calculate this, if someone can help me.

We have at the Uni one NVIDIA DGX A100

It packs 5 petaFLOPS, eight A100 GPUs with 320GB of GPU memory.

Realistically, if we get permission to run it 24/7 (which I doubt lol) for the 120 key, how long would it take? Is it feasible? And more important, would the Kangaroo even work on it?

I remember WanderingPhilosipher saying that 256 V100's that were running simultaneously for several years only exhausted 15% of #120, and regardless of how faster an A100 is versus a V100, 8 of them definitely won't make the cut.

And you also have to worry about making Kangaroo dump the private key to a file (if it even supports that without stdout redirects), or you'll lose the key if your terminal session is disconnected.

Negative, that was for #64, using Bitcrack...not #120 using Kangaroo. And it's the #64 pool that has been set up for a few years and I know that the 256 V100s have ran for weeks at a time along with other cards. Last I checked it was around 15% of total keys checked for #64.

I do not know if Zielar is running his V100s at #120.

Robert_BIT

newbie

Activity: 33

Merit: 0

Several years? Wow.

On GitHub here https://github.com/JeanLucPons/Kangaroo it says 2 months with 256 V100's. Must be wrong then - someone ought to correct that.

I'll also ask our researchers and get back on this.

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: Robert_BIT on May 23, 2021, 11:22:36 AM

I am not too sure how to calculate this, if someone can help me.

We have at the Uni one NVIDIA DGX A100

It packs 5 petaFLOPS, eight A100 GPUs with 320GB of GPU memory.

Realistically, if we get permission to run it 24/7 (which I doubt lol) for the 120 key, how long would it take? Is it feasible? And more important, would the Kangaroo even work on it?

I remember WanderingPhilosipher saying that 256 V100's that were running simultaneously for several years only exhausted 15% of #120, and regardless of how faster an A100 is versus a V100, 8 of them definitely won't make the cut.

And you also have to worry about making Kangaroo dump the private key to a file (if it even supports that without stdout redirects), or you'll lose the key if your terminal session is disconnected.

Robert_BIT

newbie

Activity: 33

Merit: 0

I am not too sure how to calculate this, if someone can help me.

We have at the Uni one NVIDIA DGX A100

It packs 5 petaFLOPS, eight A100 GPUs with 320GB of GPU memory.

Realistically, if we get permission to run it 24/7 (which I doubt lol) for the 120 key, how long would it take? Is it feasible? And more important, would the Kangaroo even work on it?

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: bigvito19 on May 21, 2021, 07:21:32 PM

Anybody else see this:

~

Damn it, I didn't get the DP size formatting correct on Windows either. Sad

Yeah this is a known bug https://github.com/ZenulAbidin/Kangaroo-256/issues/3 which also appears to affect how speed is displayed too, I have to investigate this further when I have time.

Robert_BIT

newbie

Activity: 33

Merit: 0

The .exe from a.a works fine on a modest GPU, haven't tested yet on higher specs. Thanks again for compiling!

One question:

Any particular reason why the CPU usage jumps and stays to 100% as it runs further and further? This happens with Jean Luc's version too. The temp is normal btw. Maybe is just my machine but I was curious if it's something in the code that makes use of the CPU as well as GPU.

fxsniper

member

Activity: 406

Merit: 47

Quote from: bigvito19 on May 21, 2021, 07:21:32 PM

Anybody else see this:

if you not have GPU and small laptop

don't try with 265 bit range or 256bit key

Kangaroo-256 good for high performance PC with hi end space and GPU
many time I have problem with my small gpu with Kangaroo-256

Recommend to use Kangaroo.exe JeanLucPons is better for laptop and low space PC

Kangaroo still have limited use with puzzle only success possible, Can not use real key success it 256bit

if you want to try , test setting with keyspace low than 2^70 may be can work

see on image
range too high
dp is high

try test use DP = 10

bigvito19

full member

Activity: 706

Merit: 111

Quote from: a.a on May 21, 2021, 08:22:48 PM

Expected RAM is 135431276999271530496 MB

The Executable is running into OOM (out of memory) and silently fails?

So that's why the bits are shaved off that's causing the range to seem like its smaller? It doesn't fail, it does hash. It just the range is smaller.

bigvito19

full member

Activity: 706

Merit: 111

Quote from: fxsniper on May 21, 2021, 08:19:47 PM

Quote from: bigvito19 on May 21, 2021, 07:21:32 PM

Anybody else see this:

What is problem detail?

try test compare with binary windows .exe release and you compile version Did both have problem?

I can only use the windows.exe, I can't compile because I don't have a gpu. I used 2 different laptops, 1 had reduced range and the other had doubled the range. On top of that I couldn't use my work files with the kangaroo 256.

a.a

member

Activity: 126

Merit: 36

Expected RAM is 135431276999271530496 MB

The Executable is running into OOM (out of memory) and silently fails?

fxsniper

member

Activity: 406

Merit: 47

Quote from: bigvito19 on May 21, 2021, 07:21:32 PM

Anybody else see this:

What is problem detail?

try test compare with binary windows .exe release and you compile version Did both have problem?

bigvito19

full member

Activity: 706

Merit: 111

Anybody else see this:

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: jacky19790729 on May 21, 2021, 12:21:12 PM

if I remark this line , I can compile a EXE ~~

Code:

Int.cpp

void Int::Div(Int *a,Int *mod)

qhat = _udiv128 (nh , nm , _dh , &qrem);

1>..\SECPK1\Int.cpp(1097): error C3861: '_udiv128': identifier not found

Don't do that, if you comment that out you will break division and Kangaroo won't work properly.

Windows has a problem with 128-bits multiplication and division intrinsics, they have to be emulated as macros by Kangaroo because they don't natively exist in the compiler (at least not until 2019 apparently).

a.a

member

Activity: 126

Merit: 36

What are you suggesting? That I compiled something else?

Well I used Visual Studio 2019 (Professional) on Windows 10 with CUDA 10.2 and Windows 10 SDK using the Visual Studio 2019 compiler (v142)

And if you google for _udiv128 you will find https://docs.microsoft.com/en-us/cpp/intrinsics/udiv128?view=msvc-160

Quote

The _udiv128 intrinsic is available starting in Visual Studio 2019 RTM.

jacky19790729

jr. member

Activity: 82

Merit: 8

Quote from: a.a on May 20, 2021, 06:37:12 AM

Well like I posted before
https://github.com/ZenulAbidin/Kangaroo-256/issues/6

https://github.com/ZenulAbidin/Kangaroo-256

I can compile this project in Visual Studio 2015 + Windows 7 + CUDA 8 , only 1 errors ...
if I remark this line , I can compile a EXE ~~

Code:

Int.cpp

void Int::Div(Int *a,Int *mod)

qhat = _udiv128 (nh , nm , _dh , &qrem);

1>..\SECPK1\Int.cpp(1097): error C3861: '_udiv128': identifier not found

Robert_BIT

newbie

Activity: 33

Merit: 0

Quote from: a.a on May 20, 2021, 06:09:45 AM

There are various compile versions.

Cuda 10
CUDA 10.2
CUDA 8

and CUDA 10.2 has targets for x32, x64 and x64 for SM 30 (which is actually targetting SM61)

So I will compile once for x64 with CUDA 10.2

EDIT:

here

https://github.com/ZenulAbidin/Kangaroo-256/issues/6

Thank you! I will test it on a V100 and let you know of any issues.

a.a

member

Activity: 126

Merit: 36

Well like I posted before

https://github.com/ZenulAbidin/Kangaroo-256/issues/6

NotATether

legendary

Activity: 1568

Merit: 6660

bitcoincleanup.com / bitmixlist.org

Quote from: a.a on May 20, 2021, 04:07:33 AM

Which project do you actually need to be compiled? I could compile all variations, but would like to just compile the specific one you want...

I just need Release-x64, and whatever CUDA version that works for you.

Last time I checked, there was only one project inside the solution (though my memory might be failing me), which is the main Kangaroo program, though there might be a second project for kangaroo tests. I just need the project for main Kangaroo.

EDIT: Yeah I saw your issue, now it's ready https://github.com/ZenulAbidin/Kangaroo-256/releases/tag/v2.3

a.a

member

Activity: 126

Merit: 36

There are various compile versions.

Cuda 10
CUDA 10.2
CUDA 8

and CUDA 10.2 has targets for x32, x64 and x64 for SM 30 (which is actually targetting SM61)

So I will compile once for x64 with CUDA 10.2

EDIT:

here

https://github.com/ZenulAbidin/Kangaroo-256/issues/6

Topic: Pollard's kangaroo ECDLP solver - page 55. (Read 59389 times)