4 ms·
The article seems to be a little confused. Everything I've read tells me that the current NVidia GPUs are, in fact, big arrays of scalar cores. They aren't ve
by ajross 17y ago
The article seems to be a little confused. Everything I've read tells me that the current NVidia GPUs are, in fact, big arrays of scalar cores. They aren't vector processors as traditionally defined at all.
- manvsmachine 17y agoI don't know very much about traditional vector processors, but I could see how it could be said that CUDA uses GPU's as vector processors. Whenever you create a CUDA kernel, you need to create a one, two, or three dimensional thread block of some specified size to operate on your data. Each core then isn't a vector processor, but I guess you could say that you're treating each block as one.
- btilly 17y agoYes, the chip has 16 not quite independent cores. However every instruction executed by those cores is executed on 32 parallel threads. That "32 parallel threads" bit looks like a vector processor to me. See http://techreport.com/articles.x/17670/2 http://techreport.com/articles.x/17670/2 for a more detailed explanation of this.
- wmf 17y agoThat's because NVidia uses nonstandard terminology. A real Fermi core looks quite a bit like a vector processor with 32 FPUs. http://www.realworldtech.com/page.cfm?ArticleID=RWT093009110932&p=5 http://www.realworldtech.com/page.cfm?ArticleID=RWT093009110... I also found Andy Glew's perspective interesting: http://events.berkeley.edu/?event_ID=21533&date=2009-08-27&tab=all_events http://events.berkeley.edu/?event_ID=21533&date=2009-08-...