Obsessed with technology?
Subscribe to the latest tech news as well as exciting promotions from us and our partners!
By subscribing, you indicate that you have read & understood the SPH's Privacy Policy and PDPA Statement.
News
News Categories

Your next graphics card may feature new GDDR5X VRAM chips

By Wong Chung Wee - on 2 Nov 2015, 11:51am

Your next graphics card may feature new GDDR5X VRAM chips

(Image source: Golem.de)

In the near future, your next-gen graphics cards may feature new GDDR5X video memory modules.  According to slides leaked from a major DRAM manufacturer, GDDR5X can be considered an extension of the existing GDDR5 standard. However, there are some major changes; for a start, the bus width of GDDR5X will be increased to 64-bit, in comparison, the bus width of GDDR5 is limited to 32-bit. This is because GDDR5X prefetch buffer size is 16n, i.e., 16 datawords per memory access; while GDDR5’s prefetch buffer size is 8n. With this increment, the new GDDR5X standard is looking at memory I/O rates from 10- to 12Gbps, with a possibility of reaching 16Gbps. Currently, high-density GDDR5 memory chips have an I/O rate of 8Gbps per pin. Therefore, GDDR5X has the potential to double the current memory I/O rate per pin. In addition, the new graphics memory standard is touted to have the same pin layout as GDDR5. As a result, GPU manufacturers may not have to incur substantial financial costs on implementing the GDDR5X standard.

(Image source: Golem.de)

In comparison to the AMD HBM technology, GDDR5X already has a cost-saving advantage. However, in terms of performance, it may have some ways to go in order catch up with HBM. According to the AMD, the memory bandwidth of the typical high-speed GDDR5 maxes out at 28GB/s. With reference to the specifications of the NVIDIA GTX 980 Ti card, its GDDR5 memory chip is roughly reaching this performance threshold, with the given calculation of memory bandwidth per chip as [Effective memory clock speed = 2 x memory clock} * 2 * [Bus width/8] = 2 * 1753 * 2 * [ 32 / 8 ] = 28.048GB/s.

Based on WCCFTech’s report on GDDR5X technology, they have looked at Micron’s presentation on GDDR5X and determined the memory clock speed of GDDR5X will be the same as GDDR5. As a result, we are able to estimate GDDR5X’s memory bandwidth per chip, which should be doubled that of GDDR5, at 56.096GB/s. You can also plug the numbers in the formula, i.e., 2 * 1753 * 2 * [ 64 / 8 ] = 56.096GB/s. In comparison with HBM’s bandwidth of > 100GB/s per stack, GDDR5X still needs to double its efforts to catch up.

Therefore, WCCFTech remarked that HBM is VRAM standard of the future, in terms of its implementation and performance; however, GDDR5X is a clever extension of the GDDR5 ecosystem by effectively doubling the bandwidth of the incumbent standard, and with the possibility of not costing too much for GPU manufacturers to feature in their coming graphics offerings. As a result of their difference in costs, HBM technology will most likely be exclusive to high-end graphics solutions, while GDDR5X and GDDR5 will be featured on graphics offerings for "mainstream" and "performance" solutions.

(Source: Golem.de via techPowerup, WCCFTech, Wikipedia)

Loading...