News for hackers. Stuff that matters.
Re:Testing Intel’s Arc A770 GPU for Deep Learning by cubefox on Wednesday August 09, 2023 @12:30PM (#37065115)
It's a mystery to me why graphics cards have hardwired RAM. For CPUs/motherboards RAM has been extensible for decades.
Re:Testing Intel’s Arc A770 GPU for Deep Learning by alargemoose on Wednesday August 09, 2023 @06:57PM (#37069507)
Yhe main reason is speed. For example DDR5 has a transfer rate of about 50GB/s. The GDDR6X used in GPUs has a transfer rate of about 1TB/s. Nvidia just announced H200 AI processors have HBM3e memory that they report can go as fast at 10TB/s.
Re:Testing Intel’s Arc A770 GPU for Deep Learning by ElectricalUnion on Thursday August 10, 2023 @10:26AM (#37076574)
Speed/signaling issues.

To me is a mystery that CPUs/motherboards still have extensible RAM, integrated RAM (if you don't skimp and go for cheap RAM) can be very fast - for example the Steam Deck has quad-channel 5500 MT/s LPDDR5, very console-like.

Re:Testing Intel’s Arc A770 GPU for Deep Learning by muttled on Friday August 11, 2023 @12:48PM (#37090817)
What is it that make extensible RAM inherently slower? Is it the connectors raised off the boards, adding physical distance to the traces that's the issue?
Re:Testing Intel’s Arc A770 GPU for Deep Learning by ElectricalUnion on Saturday August 12, 2023 @06:46PM (#37104912)
The mechanical connector layout forces sub-optimal signalling paths.

You're forced to make everything slower to make things consistent.

That being said, a more optimized connector, board layout and form factor could make large-capacity, lower-latency modules possible. Not sure for example if there is anything out there besides the Dell proposed CAMM form factor, and I don't remember that one being specifically promoted as "faster", mostly "smaller".