 |
|||||||||||
|
|
|
|||||
Late last month on June 24th AMD finally released its Fury X graphics card, possibly the most hyped product of 2015. The week prior to that we received some refreshed AMD GPU’s in the form of the Radeon 300 series. Most notably the R9 390X and R9 390, which were architecturally identical to the previous R9 290X/290 in every conceivable way. The only upgrades made included 8GB of faster memory and an ever so slightly higher core clock... The faster memory offered a reasonable performance boost and certainly helped strengthen the performance of AMD’s high-end GPU’s. The only issue being the price, as the R9 390X currently retails for over $100 more than the R9 290X, and if you are to overclock the 290X you end up with the same performance even in games that use more than 8GB of VRAM.  So not the most exciting release, but if AMD had of been more aggressive with their pricing we would have been pleased nonetheless. Looking forward all eyes were on the upcoming Radeon R9 Fury X, codenamed ‘Fiji XT’. Built using the same GCN 1.2 architecture as the R9 380, the Fury X features more than twice the SPUs and 45% more than the R9 390X. The core specifications of the Fury X are impressive, but a truly impressive statistic is the memory bandwidth, which at 512GB/s is 33% faster than that of the R9 390X and 60% higher than the R9 290X. That is a huge leap on what was already a very fast memory subsystem using high speed GDDR5 6Gbps memory. Enabling this massive memory bandwidth is High Bandwidth Memory (HBM) memory, a new high-performance memory standard that uses vertically stacked DRAM dies and fast microscopic interconnects called through-silicon vias (TSVs). The three-dimensional die stacking and tiny interconnects allows for superior bandwidth at lesser power consumption than both GDDR5 and DDR4. By most every memory metric, HBM is a superior memory technology to GDDR5. HBM offers a GPU more bandwidth with less power usage than GDDR5, and does so in 19x less surface area than GDDR5 requires. One limitation of HBM is that for now it only works at up to 500MHz (1.0Gbps), but with a 4096-bit wide memory interface this still allows for 512GB/s of memory bandwidth. Another limitation being the memory capacity, which limits the Fury X to just 4GB’s of VRAM. That said, with such a screaming high throughput it seems unlikely the Fury X will be memory bound. HBM certainly is an exciting new memory technology that will enable AMD to develop some truly innovative GPUs, and the Fury X looks like the best spot to start... |
|||||
|