| Intrinsic said: I don't get why people are throwing around HBM like its the holy grail of memory. And in the process underselling GDDR6. The fastest GDDR6 ram (aka as highest capacity) will have 24GB and a bandwidth of 864GB/s. In comparison the next best thing on the HBM front will have a bandwidth of around 901GB/s. We are talking about a bandwidth difference of less than 50GB/s. And a price difference of more than 50% and thats not factoring in the significantly lower yeilds. I think we need to stop looking at the tech and more at the performance when it comes to this.... anything higher than 500GB/s bandwidth will be more than sufficient for 4k gaming. There is also that possibility that console makers can even go with a custom "extra wide bus" which would result in even more bandwidth from their chosen ram. |
In terms of memory bandwidth, it's absolutely top notch and I don't wish for 4K, I wish for physically based dynamic global illumination ... (BW might become a severe bottleneck when doing ray traversal in ray tracing so I want this mitigated as soon as possible for next generation)
Also the difference is much larger than 50 GB/s. The fastest HBM 2 memory module can let us achieve rates as high as 1.25 TB/s on a 4096-bit bus width while the maximum a GDDR6 standard memory module can achieve on a 512-bit bus width will net 1 TB/s. GDDR6 has a massive 20% BW deficit compared to the fastest HBM 2 memory module ...
