Showing posts with label HBM. Show all posts
Showing posts with label HBM. Show all posts

Friday, 9 October 2015

AMD to release Dual 'Fiji' GPU Graphics Card sooner than expected

Dual 'Fiji' GPU R9 Fury X2 Graphics Card
A few days back shipping information listing 'Fiji Gemini' being shipped from AMD's headquarters in Canada. 'Gemini' is the code name that has been used in AMD's previous Dual-GPU Radeon R9 295X graphics card. Although only speculation, but the new graphics card is likely to be called the Radeon R9 Fury X2.
It is certain that AMD will include 2 x 'Fiji' GPU's that are currently used in the R9 Fury, R9 Fury X and R9 Fury Nano graphics card. The shipping information also states that the dual 'Fiji' card will include a Cooler Master heat sink, and most certainly come with an all-in-one water cooling solution to keep temperatures down.

Shipping information for the Radeon R9 Fury X2

The dual 'Fiji' graphics card will feature a whopping 8192 Stream Processors with 128 GCN Compute units, 128 Render Output Units, 512 Texture Mapping Units, and 8GB of HBM (High-Bandwidth-Memory) with 4GB per GPU chip. The memory bandwidth and frequency are expected to remain the same speed as the R9 Fury X.

The sudden move by AMD to launch the R9 Fury X2 sooner could be a response to Nvidia's plans to release it's dual 'Maxwell' graphics card.

Source: WCCFtech

Tuesday, 6 October 2015

Nvidia's 'Pascal' set to feature HBM 2.0

Leaked slides from Nvidia's GTC Taiwan Presentation show HBM Memory in their Next-Gen GPU Range
 Nvidia's upcoming next-generation 'Pascal' is expected to feature HBM 2.0 (High-Bandwidth-Memory) in it's product line due to be released sometime in the first quarter of 2016. A leaked slide from Nvidia's GTC Taiwan 2015 presentation shows clearly an illustration of it's next generation GPU with HBM memory on the GPU die. According to the leaked slides, Nvidia will name it's HBM stacked memory as '3D Memory'.

The move to HBM by Nvidia makes sense as their main competitor AMD has recently released it's 'Fiji' range with HBM. If Nvidia were to include HBM 2.0 in it's upcoming GPU lineup, it will have a memory bandwidth of around 1 TB/s, which is double the speed of AMD's current generation 'Fiji' based GPUs.

A leaked slide showing Pascal's key features including '3D Memory'
Current rumors also indicate that Nvidia will include up to 16 GB of HBM 2.0 memory, however looking at the illustration on the slide, it's unlikely that this will be true as the picture shows the GPU with 4 stacked memory modules, AMD have the same setup with 1 GB of HBM on each stacked module, this would mean that each module would have to be 4 GB each. As HBM technology is at it's early stages, it's unlikely that 16 GB can be achieved.

What we do know is that Nvidia will most likely manufacture it's 'Pascal' GPU in 16nm FinFET process by TSMC.