5 ms·
I think it would be amazing if there were also able to stick a GB or two of HBM onto the same chip as kind of a L4 cache / VRAM for the GPU. Performance would b
by lykr0n 9y ago
I think it would be amazing if there were also able to stick a GB or two of HBM onto the same chip as kind of a L4 cache / VRAM for the GPU. Performance would be amazing, especially considering that Zen's performance gets better with RAM speed and GPUs are almost entirely RAM starved.
- wastedhours 9y agoI've never considered that as an idea before, but now am left wondering why it hasn't been done. Guessing the manufacturers make more money from full-unit upgrades than they'd expect to make from incremental memory unit purchases (even if they made the RAM slot proprietary).
- bryanlarsen 9y agoIntel has a new chip coming out with AMD graphics and HBM memory: https://www.anandtech.com/show/12220/how-to-make-8th-gen-more-complex-intel-core-with-radeon-rx-vega-m-graphics-launched https://www.anandtech.com/show/12220/how-to-make-8th-gen-mor...
- pkaye 9y agoHBM is fairly new so it is probably a cost issue for the target market.
- undersuit 9y agoThere was a time where you could put an extra DDR2/3 DIMM on some AMD motherboards to complement their chipset based graphics. http://www.tomshardware.com/reviews/790gx-graphics-sideport,2088-2.html http://www.tomshardware.com/reviews/790gx-graphics-sideport,...
- dragontamer 9y agoRumor is that an interposer alone is $20, and HBM is also rather expensive. Basically, you'd easily double the cost of these $99 chips if you put HBM2 onto them... maybe triple the costs if you consider that AMD should charge a margin and actually make a profit from these designs. After that, the GPU would be too-strong on memory and too weak on compute. The minimum HBM speed 1024-bits / 250GB/s from one stack. Yeah, the same memory-bandwidth as a RX 580 with 2304 shaders, while the APU 2400G only has 704 shaders. The Intel + AMD chip manages to solve the problem: the CPU, GPU, and HBM2 are all on the interposer. So the expensive interposer has a shared design. Second, the GPU is a fully dedicated Vega design that's roughly the size of an Rx 580. That way, the dedicated-GPU can "keep up" with the super-fast HBM2 RAM.
- amaranth 9y agoIt's also not using the AMD-style interposer but Intel's EMIB which they claim is much cheaper.