- No mention of cuda core count.
cuda core seems in the range of double the one of a titanX, around 5k
i'm expecting as well a 2x boost from this new tech, no more
Topic: Nvidia Pascal speculations (Read 1949 times)
cuda core seems in the range of double the one of a titanX, around 5k
i'm expecting as well a 2x boost from this new tech, no more
- A brief mention of a 2x performance per watt improvement, without any precision on the type of application. Pure marketing that doesn't mean anything.
AMD seems to be saying the same with their new Fury cards. Does it have anything to do with that they are both using stacked memory?
Never heard about pascal before I clicked this thread. Did a quick google search and found this article that gave a good overview, if anyone else needs an update. http://wccftech.com/nvidia-pascal-gpu-gtc-2015/
Careful with wccftech, they have a history of reporting rumors and loose fact checking.
Better go straight to the source: http://blogs.nvidia.com/blog/2015/03/17/pascal/ (the blog post embeds the Nvidia keynote).
- A brief mention of a 2x performance per watt improvement, without any precision on the type of application. Pure marketing that doesn't mean anything.
AMD seems to be saying the same with their new Fury cards. Does it have anything to do with that they are both using stacked memory?
Never heard about pascal before I clicked this thread. Did a quick google search and found this article that gave a good overview, if anyone else needs an update. http://wccftech.com/nvidia-pascal-gpu-gtc-2015/
- A brief mention of a 2x performance per watt improvement, without any precision on the type of application. Pure marketing that doesn't mean anything.
AMD seems to be saying the same with their new Fury cards. Does it have anything to do with that they are both using stacked memory?
Never heard about pascal before I clicked this thread. Did a quick google search and found this article that gave a good overview, if anyone else needs an update. http://wccftech.com/nvidia-pascal-gpu-gtc-2015/
I don't know much about Pascal but I don't think the performance gain will be anywhere close to 10x for mining.
The die size is the most important factor I think. The increase in efficiency will put even Maxwell cards at a disadvantage but to what degree, we'll see.
This and memory bandwidth and access... The die size is the most important factor I think. The increase in efficiency will put even Maxwell cards at a disadvantage but to what degree, we'll see.
Hope they will go for HBM memory
They will. At least that's what they announced in the keynote. They mentioned HBM2. I don't know if is an improvement on HBM or just a one-up aimed at AMD...
Does memory bandwidth have a big impact on performance? I guess it varies with each algorithm...
I don't know much about Pascal but I don't think the performance gain will be anywhere close to 10x for mining.
The die size is the most important factor I think. The increase in efficiency will put even Maxwell cards at a disadvantage but to what degree, we'll see.
This and memory bandwidth and access... The die size is the most important factor I think. The increase in efficiency will put even Maxwell cards at a disadvantage but to what degree, we'll see.
Hope they will go for HBM memory
They will. At least that's what they announced in the keynote. They mentioned HBM2. I don't know if is an improvement on HBM or just a one-up aimed at AMD...
Does memory bandwidth have a big impact on performance? I guess it varies with each algorithm...
I don't know much about Pascal but I don't think the performance gain will be anywhere close to 10x for mining.
The die size is the most important factor I think. The increase in efficiency will put even Maxwell cards at a disadvantage but to what degree, we'll see.
This and memory bandwidth and access... The die size is the most important factor I think. The increase in efficiency will put even Maxwell cards at a disadvantage but to what degree, we'll see.
Hope they will go for HBM memory
I don't know much about Pascal but I don't think the performance gain will be anywhere close to 10x for mining.
The die size is the most important factor I think. The increase in efficiency will put even Maxwell cards at a disadvantage but to what degree, we'll see.
The die size is the most important factor I think. The increase in efficiency will put even Maxwell cards at a disadvantage but to what degree, we'll see.
I keep reading that Nvidia Pascal architecture is going to be a game-changer, especially for the mining world, everybody quoting the 10x performance improvement from Nvidia's blog title.
I have serious doubts about that. The 10x performance improvement is mentioned for deep learning applications, which have very little in common with mining.
So, let's analyse the spec sheet step by step:
- Up to 32 GB of RAM, no use for us
- Higher memory bandwidth, no use for us
- Faster inter-GPU communication, no use for us
- Mixed-precision computing. Not sure floating point operations are used for mining, are they?
- No mention of cuda core count.
- A brief mention of a 2x performance per watt improvement, without any precision on the type of application. Pure marketing that doesn't mean anything.
- Smaller die (is it 14 or 16nm?), that one definitely means an efficiency improvement.
I'm not saying these features won't be killer in some applications (like gaming and AI), and there will be an improvement for mining. Just not 10x, from what I've read so far.
Does anyone have concrete information on Pascal that I missed?
I have serious doubts about that. The 10x performance improvement is mentioned for deep learning applications, which have very little in common with mining.
So, let's analyse the spec sheet step by step:
- Up to 32 GB of RAM, no use for us
- Higher memory bandwidth, no use for us
- Faster inter-GPU communication, no use for us
- Mixed-precision computing. Not sure floating point operations are used for mining, are they?
- No mention of cuda core count.
- A brief mention of a 2x performance per watt improvement, without any precision on the type of application. Pure marketing that doesn't mean anything.
- Smaller die (is it 14 or 16nm?), that one definitely means an efficiency improvement.
I'm not saying these features won't be killer in some applications (like gaming and AI), and there will be an improvement for mining. Just not 10x, from what I've read so far.
Does anyone have concrete information on Pascal that I missed?
Jump to: