View previous topic :: View next topic |
Author |
Message |
Csimbi I post too much Reputation: 94
Joined: 14 Jul 2007 Posts: 3110
|
Posted: Wed Feb 14, 2024 12:41 pm Post subject: Happy Valentine's day! |
|
|
And year ago today we got a CE release.
Keeping fingers crossed for a new release
In the meantime, happy V's y'all!
|
|
Back to top |
|
|
LeFiXER Grandmaster Cheater Supreme Reputation: 20
Joined: 02 Sep 2011 Posts: 1055 Location: 0x90
|
Posted: Wed Feb 14, 2024 1:34 pm Post subject: |
|
|
Happy Valentine's Day <3
|
|
Back to top |
|
|
Dark Byte Site Admin Reputation: 458
Joined: 09 May 2003 Posts: 25298 Location: The netherlands
|
Posted: Thu Feb 15, 2024 3:02 am Post subject: Re: Happy Valentine's day! |
|
|
Csimbi wrote: |
Keeping fingers crossed for a new release
|
soon I hope.
I've spend a few weeks on a pointerscan for CUDA again but I had to give that up for now (Just can't get the performance to even match that of a CPU)
But it did help the normal pointerscan to get a 30% performance boost (and eats a little more RAM now)
_________________
Do not ask me about online cheats. I don't know any and wont help finding them.
Like my help? Join me on Patreon so i can keep helping |
|
Back to top |
|
|
TsTg Master Cheater Reputation: 5
Joined: 12 Dec 2012 Posts: 334 Location: Somewhere....
|
Posted: Thu Feb 15, 2024 6:15 pm Post subject: |
|
|
little late, but Happy V's day
looking forward to a new CE release, thanks for your hard work DB .
|
|
Back to top |
|
|
Csimbi I post too much Reputation: 94
Joined: 14 Jul 2007 Posts: 3110
|
Posted: Fri Feb 16, 2024 11:08 am Post subject: Re: Happy Valentine's day! |
|
|
Dark Byte wrote: | I've spend a few weeks on a pointerscan for CUDA again but I had to give that up for now (Just can't get the performance to even match that of a CPU)
But it did help the normal pointerscan to get a 30% performance boost (and eats a little more RAM now) |
From what I gathered, pointer scans are more memory-bound than CPU-bound.
CUDA helps in massively parallelization.
I think the problem is, the massive volume of requests from a 1000 cores to the memory will quickly saturate and overwhelm the memory controller.
Hence, beyond a certain point, more and more cores means less and less overall speed.
30% is an impressive improvement though for non-CUDA scans!
I am looking forward to it.
|
|
Back to top |
|
|
|