vacancies advertise contact news tip The Vault
facebook rss twitter

Slides touting GDDR5X memory advantages leak online

by Mark Tyson on 26 October 2015, 13:01

Tags: Micron (NASDAQ:MU), AMD (NYSE:AMD), NVIDIA (NASDAQ:NVDA)

Quick Link: HEXUS.net/qacvpn

Add to My Vault: x

Please log in to view Printer Friendly Layout

A trio of slides detailing the advantages of employing GDDR5X in graphics cards have leaked online, reports VRZone. The slides are said to come from a major DRAM supplier's presentation deck. GDDR5X is promoted as a good choice for graphics cards makers going forward as it offers double the data rate per-pin and extends a well established standard, so will be easy to implement, adding money savings to its attractiveness.

GDDR5X may indeed help graphics card designers at AMD and Nvidia transition between GDDR5 and the energy efficient HBM. The newer, improved GDDR5X is a simple upgrade to graphics card makers offering the "same interface signal pin count". It requires only "limited effort" to upgrade the GDDR5 memory controller on existing designs, it offers reduced power consumption and, perhaps most importantly "takes performance to a new level".

Looking closer at the performance you can see from the small graph, in the slide embedded below, that GDDR5X offers double the data rate per pin of GDDR5. Initially the DRAM maker from which these slides were poached, says that the I/O rates targeted will be between 10 and 12 Gbps. In years to come there is potential of achieving up to 16Gbps. Current GDDR5 tech has plateaued, offering an 8Gbps maximum bandwidth.

Previous rumours going around have suggested that some Nvidia Pascal GPUs will be deployed on graphics cards equipped with GDDR5X memory. A more recent report, mentioned in passing by VRZone, says that AMD will look into using GDDR5X with some of its next-gen GPUs.

Which memory maker is said to be the source of the above slides? According to VRZone's source, 3DNews.ru, and as evidenced by lesser-cropped slides at Golem.de, these GDDR5X slides come from Micron.



HEXUS Forums :: 8 Comments

Login with Forum Account

Don't have an account? Register today!
So i'm guessing this will be used on lower end cards, I just hope it doesn't slow down the adoption of HBM.
DemonHighwayman
So i'm guessing this will be used on lower end cards, I just hope it doesn't slow down the adoption of HBM.

I think DDR4 will be used on the lowest end, this gives the mid range a bit more breathing room *if* it's adopted, which I have some doubts about. But HBM blows this out of the water and I don't think costs the earth, just time.
kalniel
… HBM blows this out of the water and I don't think costs the earth, just time.

I'll be interested to see if next year's mid range gets HBM1 while the top end get HBM2, or if HBM2 will just trickle straight down. After all, the potential capacity and bandwidth per stack for HBM2 is pretty phenomenal.

Can't really see the benefit of what is essentially just faster GDDR5. Board space will remain an issue, and I assume power consumption will too, since you'll still need a lot of power to drive those long PCB traces at high data rates. Plus didn't AMD save significant silicon by engineering Hawaii for a wider but slower memory bus? Presumably GDDR5X controllers will require those larger PHYs to drive the very high data rates? It seems like a retrograde concept tbh.
scaryjim
Can't really see the benefit of what is essentially just faster GDDR5. Board space will remain an issue, and I assume power consumption will too, since you'll still need a lot of power to drive those long PCB traces at high data rates. Plus didn't AMD save significant silicon by engineering Hawaii for a wider but slower memory bus? Presumably GDDR5X controllers will require those larger PHYs to drive the very high data rates? It seems like a retrograde concept tbh.
Yeah, I think it's basically for rebrands. If we get this/last years GPUs rebranded yet again with more memory, they might just plonk on some DDR5x chips and update the mem controller.
At this level of technological advancement of the current GDDR5 memory its speed is sufficient but the GPU suffers slowness that's why even if you add HBM to a GTX970 GPU don't expect frame rates similar to a 980 at 4k. I want GPU efficiency and more performance per dollar that's where Pascal and next gen Fury come to mind even if they will run at the current GDDR5 tech.