Remix.run Logo
jsheard 3 days ago

Tell that to Nvidia, Blackwell is already up to 752 cores (each with 32-lane SIMD).

phkahler 2 days ago | parent | next [-]

640K cores should be enough for everyone.

fooblaster 3 days ago | parent | prev [-]

b200 is 148 sms, so no

jsheard 3 days ago | parent [-]

Each SM cluster contains 4 independent 32-wide compute units, and GB202 has 192 SMs, although only 188 of them are enabled on the largest shipping SKU. IMO that makes for 752 "cores", but depending on where you draw the line it could be 188, 752, or 24064.

fooblaster 3 days ago | parent | next [-]

sms is the Nvidia definition of processor, and cuda device properties returns it, not anything else. If you want a marketing number, use cuda cores, it doesn't consistently match to anything in the hardware design.

markhahn 3 days ago | parent | prev [-]

no, you really can't.

NVidia's use of "cores" is simply wrong. unless you think a core is a simple scalar ALU. but cores haven't been like that for decades.

or would you like to count cores in a current AMD or Intel CPU? each "core" has half a dozen ALUs/FP pipes, and don't forget to multiply by SIMD width.