csa not detected
- Thread starter victor
- Start date
CUDA is an NVIDIA technology.
Must be an nvidia card with a high number of cuda cores ,not any nvidia card that supports cuda.
Depends on how much money you want to expend.
There are cards with 5120 cuda cores for $19.000.00.
But I am sure you won't be going there.
But then you can get couple 3000 cuda cores for $2000.
In the $400 for 1600 cuda cores.
- Messages
- 24,798
MY NVIDIA GTX 1080
GPU Engine Specs
CUDA Cores
2560
I know i would'nt ever spend $2,000 let alone $19k just for a few hundred extra cores , i had previously GTX 660 With 960 cores i can not see that much difference other than the big fact my GTX 660 would struggle much more with watching 4:2:2 Feeds ..
searching keys is much the same ,but i'd say go for NVIDIA GTX 1080 @ £480.00 ... mine i paid £360 on Black Friday deal
which was decent value for money i thought : )
GPU Engine Specs
CUDA Cores
2560
I know i would'nt ever spend $2,000 let alone $19k just for a few hundred extra cores , i had previously GTX 660 With 960 cores i can not see that much difference other than the big fact my GTX 660 would struggle much more with watching 4:2:2 Feeds ..
searching keys is much the same ,but i'd say go for NVIDIA GTX 1080 @ £480.00 ... mine i paid £360 on Black Friday deal
which was decent value for money i thought : )
Thinking that cuda processes increment exponentially,it makes it worth to have few extra thousand cores.
I started with gtx 260 with less tan 200 cores back in the times we were testing cudabiss,then got a gtx460,465,I mean,each time made me think how much important is to have some extra cuda cores.
I assume some people has the means to afford them.
campag5242
Feed Hunter
- Messages
- 2,585
Cuda processing power increases more or less linearly with core_count*core_freq.
If using Colibri's original tools, you can throw thousands at GPU hardware, but never get lower than the disk search time: the three tasks involved in rbt key lookup are done sequentially in those tools: 1. calc all end6 values; 2. search disk for end6/start6; 3. rebuild chains.
Massive GPU power will shorten the first and last of those three steps, but the disk search time remains fixed by speed of disk hardware & search algo. Conclusion: there comes a point where huge GPU power is wasted if using Colibri's tools.
Rewrite, doing all three steps concurrently (focusing mostly on disk search optimisation), and a £100 GTX 1060 can outpace any hardware using Colibri's tools, and by a long way.
If using Colibri's original tools, you can throw thousands at GPU hardware, but never get lower than the disk search time: the three tasks involved in rbt key lookup are done sequentially in those tools: 1. calc all end6 values; 2. search disk for end6/start6; 3. rebuild chains.
Massive GPU power will shorten the first and last of those three steps, but the disk search time remains fixed by speed of disk hardware & search algo. Conclusion: there comes a point where huge GPU power is wasted if using Colibri's tools.
Rewrite, doing all three steps concurrently (focusing mostly on disk search optimisation), and a £100 GTX 1060 can outpace any hardware using Colibri's tools, and by a long way.
I tried to say cuda cores increment exponentially with time : now you can get ten fold the cuda cores for the same money you paid for GPU 10 years ago.
And of course search time,HDD speed access makes a big impact.
Not contending the facts.