In context: As producers proceed to squeeze each final drop of efficiency out of present GDDR6 and GDDR6X reminiscence modules, Samsung has introduced a brand new and improved entry to the household — GDDR6W. Samsung claims that GDDR6W can compete with HBM2’s bandwidth and speeds.
In 2016, Samsung and different producers started manufacturing the successor to the quick (however flawed) high-bandwidth reminiscence (HBM) modules. High-bandwidth reminiscence 2 (HBM2) seemingly fastened all the problems with the earlier technology, rising capability, speeds, and bandwidth. Unfortunately, HBM2 was by no means considerably profitable within the desktop graphics card market.
Fury and Vega strains of playing cards utilized HBM and HBM2, respectively. Sadly, every flopped, and AMD reverted to GDDR6 reminiscence once more, beginning with the RX 5000 lineup. Some customers have been understandably disenchanted concerning the fast abandonment of HBM2.
On Tuesday, Samsung revealed its newest entry to the GDDR6 household, GDDR6W. The South Korean tech big wished to convey among the advantages of HBM2 to the already profitable GDDR6 platform, particularly the elevated bandwidth. Based on the small print and numbers Samsung offers, GDDR6W might be a sport changer in future GPUs.
Samsung put a heavy emphasis on digital actuality and “Metaverse” functions. However, there is no purpose that GDDR6W would not convey advantages to future discrete graphics playing cards usually.
Samsung began by taking its present GDDR6 platform and implementing what it calls “Fan-Out-Wafer-Level Packaging” (FOWLP). Rather than putting the reminiscence dies on a PCB, they’re mounted instantly onto a silicon wafer. Redistribution layers permit for “finer wiring patterns,” and since no PCB is concerned, the modules will probably be thinner total and have higher warmth dispersion.
The reminiscence modules are double-stacked, which Samsung notes can permit for twice the capability and double the reminiscence bandwidth. The reminiscence’s restrict is 24 GB per die for traditional GDDR6, so the brand new platform might theoretically help 48 GB on a single graphics card.
These adjustments to module placement and total die measurement lead to GDDR6W being 36% shorter than its GDDR6 counterpart. Due to the unchanged footprint, these modules might be “put into the identical manufacturing processes” utilized in present GDDR6 merchandise.
As seen above, GDDR6W’s bandwidth comes very near that discovered on HBM2E dies. The present GDDR6X’s bandwidth restrict sits at round 1 TB per second, and GDDR6W considerably bumps that by about 400 MB/s.
While we nearly actually won’t see GDDR6W on any Lovelace or RDNA3 graphics playing cards, there’s all the time an opportunity Nvidia or AMD will take a swing on the different and undertake GDDR6W in an upcoming technology. Regardless, Samsung’s new creation is an enormous step ahead for graphic processors, whether or not discrete or “for the Metaverse.”