Previous HN discussion:
https://news.ycombinator.com/item?id=3004446
The url linked there redirects to the new site, but the old site is still available through the Wayback Machine:
Modern GPU
11–20 of 30 posts
Re: Modern GPU
#12So how do I use this with the new and modern Radeon HD7990 card? Hmm looks like there is no CUDA library for it... (flagging post for misleading title)
To the best of my knowledge, CUDA is Nvidia only. You would need to look at OpenCL or something similar. I guess the techniques presented in the article could be ported to OpenCL.
I think that's the point... The title conveys a more general article.
Re: Modern GPU
#13So how do I use this with the new and modern Radeon HD7990 card? Hmm looks like there is no CUDA library for it... (flagging post for misleading title)
To the best of my knowledge, CUDA is Nvidia only. You would need to look at OpenCL or something similar. I guess the techniques presented in the article could be ported to OpenCL.
Right, CUDA is Nvidia only (8-series and above, so basically anything after 2006). The stuff in the article can totally be replicated in OpenCL. Especially since OpenCL 1.1 where buffers can be shared between GL and CL contexts, which IIRC was the only real technical capability CUDA had that OpenCL did not.
Re: Modern GPU
#14The title here, although it matches that of the linked page, conveys little information to anyone who doesn't already know what it's about. ("Modern GPU is code and commentary intended to promote new and productive ways of thinking about GPU computing. This project is a library, an algorithms book, a tutorial, and a best-practices guide.") Perhaps something like "Modern GPU: NVIDIA's guide to making effective use of…
I absolutely agree, but I stopped editorializing titles because they were either reverted to the original title, o re-editorialized by the moderators.
I can't change the title anymore, hope some moderator will.
Re: Modern GPU
#15For those interested in articles like these, there's a really great course on iTunes U (and YouTube, among other places) from UC Davis titled "Graphics Architecture", taught by John Owens. [1]
I clicked on it being tangentially interested in GPUs and ended up watching all the lectures within the next few days with rapt attention. The topic ended up being far more interesting than I thought and was delivered in a very approachable way.
[1] https://itunes.apple.com/us/itunes-u/graphics-architecture-w...
Re: Modern GPU
#16The title here, although it matches that of the linked page, conveys little information to anyone who doesn't already know what it's about. ("Modern GPU is code and commentary intended to promote new and productive ways of thinking about GPU computing. This project is a library, an algorithms book, a tutorial, and a best-practices guide.") Perhaps something like "Modern GPU: NVIDIA's guide to making effective use of…
Totally agree. However every single time I've tried to editorialize a title like this one to make it better the mods have changed it back to the page's title. I imagine a lot of people have given up on even trying to make informative titles. It's the worst part about this forum in my opinion.
And the suggestion to make a blog post about it, then submit that, is really just encouraging blog spam, which typically gets flagged into oblivion. I know that I will do so - a one or two sentence blog post should not IMO be submitted in lieu of the article itself. I personally thought the ability to editorialize with only flamebait titles being changed by the mods was way better. Yes, it led to arbitrary decisions by the mods, but I see two problems with the current system:
- Information is often missing from the original, as it is here
- The mods are not always so quick to rename, and it's frustrating to see an article that I previously read changed to another title, resulting in confusion when I see it.
Further, The current system actually incentivizes over-editorializing the title to get a post boosted onto the front page, knowing it will be causally re-titled by the mods in due course.
Re: Modern GPU
#17So how do I use this with the new and modern Radeon HD7990 card? Hmm looks like there is no CUDA library for it... (flagging post for misleading title)
What do you suggest, he writes a new title it will get flagged because it does not have the original title (or it gets reverted).
Suggest you undo your flag or write a better article with a better title.
Re: Modern GPU
#18So how do I use this with the new and modern Radeon HD7990 card? Hmm looks like there is no CUDA library for it... (flagging post for misleading title)
What do you suggest, he writes a new title it will get flagged because it does not have the original title (or it gets reverted). Suggest you undo your flag or write a better article with a better title.
OK undone my flag.
However, it was flagged originally because it promotes bad way forward for the community who want to learn GPU computing. It is pushes material for one company NVIDIA and its products and SDK, while there is an alternative open standard.
Instead of embracing it they continue splitting the community and do it an deceiving way "Modern GPU"? I gave an example of a modern GPU that this doesn't apply.
> a better article with a better title.
That is a bit dis-ingenious. Every time someone is wrong on the internet I will not go and spend hours re-writing their stuff but better. I thought a vote of flag was supposed to achieve some of that. Yes, it is a lot better to contribute and improve but I pick my battles, and filtering out bad stuff is another way of contributing.
Re: Modern GPU
#19For those interested in articles like these, there's a really great course on iTunes U (and YouTube, among other places) from UC Davis titled "Graphics Architecture", taught by John Owens. [1] I clicked on it being tangentially interested in GPUs and ended up watching all the lectures within the next few days with rapt attention. The topic ended up being far more interesting than I thought and was delivered in a very…
There's also a GPU programming course being taught at Udacity:
Re: Modern GPU
#20Guide looks very well written, and I'm actually studying parallel programming right now...bookmarked. Side Note: Maybe this is Nvidia's way to promote sales, since AMD cards sold like crazy cause of the mining craze.
It's not reactionary: nvidia invested heavily in getting to market first and advertising to the HPC crowd and they have been reaping the rewards of a monopoly for a few years now. Their latest high-end cards are dramatically more expensive (vs AMD at the same raw computational power) while offering completely crippled GPGPU capabilities. I recall incredulity on IRC when people found that their new 6xx cards were a step backwards from the corresponding 5xx cards for their GPGPU apps. Pro media & HPC users fork over the cash for Tesla because they can't jump to AMD due to their legacy CUDA code. Gamers don't care since they don't use double precision arithmetic.
The bitcoin community was the only one nimble enough to switch to AMD. For those of us in academia, we'll be paying the piper for the foreseeable future :/