Hynix Announces World's Fastest, First 1 Gb GDDR5 Memory
Hynix Semiconductor has announced the world’s first 1 Gigabit GDDR5 graphics DRAM. The chips are also the fastest yet, which should help to enable a new generation of high performance graphics cards.
The new graphics memory is built on a 54 nm process technology and is capable of operating at a record setting speed of 7 Gb/s. When used along with a 32-bit memory bus, the memory can process up to 28 Gigabytes of data per second and when used along with a 512-bit memory bus, a bandwidth of up to 448 GB/s should be possible.
For comparison, Qimonda had announced volume production of 512 Mb GDDR5 memory rated at up to 4.5 Gb/s back in May 2008 and Samsung had announced 512 Mb GDDR5 memory rated at 6 Gb/s late last year. The ATI Radeon HD 4870 graphics card series of products were the first to ship using GDDR5 memory, which used Qimonda’s 512 Mb GDDR5 memory running at 3.6 Gb/s.
The new GDDR5 memory from Hynix is also is designed to be more energy-efficient than that of the previous generation of GDDR5 memory, requiring only a 1.35 V power supply instead a 1.5 V power supply. Compared to the currently popular GDDR3 memory, GDDR5 memory running at 1.5 V already offered an approximate 20-percent improvement in power consumption. A reduced power consumption could mean less generated heat and increased battery life for notebooks that use the memory.
The new Hynix graphics memory conforms to the JEDEC standard, with Hynix planning to start volume production in the first half of 2009. According to Samsung, it is expected that “GDDR5 memory chips will become the de facto standard in the top performing segment of the market by capturing more than 50 percent of the high-end PC graphics market by 2010.” Intel’s upcoming Larrabee GPU is also expected to use GDDR5 memory when it is released in late 2009 or 2010.
Get Tom's Hardware's best news and in-depth reviews, straight to your inbox.
-
LAN_deRf_HA This basically means we'll have cards with about 211.2 GB/s by 3rd quarter next year. They never run the chips at their max rated speed so they'll be at 6,600 effective and they won't be slapping 512 bit buses back onto cards until there's a need for more bandwidth, which there really isn't past 110 GBs (notice how ati's 4870 can trade blows with a 280 GTX when heavy filters {8xaa} are used despite having less bandwidth). So overall expect only slight memory performance increases over the next year. The focus will be on simplifying PCBs and reducing power consumption. I think the biggest thing we might see is standardization of 1 GB of memory for midrange cards.Reply -
the last resort but, ATI could create a card with a 512 bit bus and 1+ GB ram, because if you can totally destroy the competition, then why not?Reply -
jaragon13 LAN_deRf_HAThis basically means I somewhat agree,but unfortunately I think the real improvement needs to be in common memory.we'll have cards with about 211.2 GB/s by 3rd quarter next year. They never run the chips at their max rated speed so they'll be at 6,600 effective and they won't be slapping 512 bit buses back onto cards until there's a need for more bandwidth, which there really isn't past 110 GBs (notice how ati's 4870 can trade blows with a 280 GTX when heavy filters {8xaa} are used despite having less bandwidth). So overall expect only slight memory performance increases over the next year. The focus will be on simplifying PCBs and reducing power consumption. I think the biggest thing we might see is standardization of 1 GB of memory for midrange cards.Reply -
neiroatopelcc "from Hynix is also is designed"Reply
Remove one of the words please!
Can't help but to wonder how much faster a 280 would be with 2gb of these modules - the 4870 is only really fast because of the memory ... -
hannibal Well, not only for that. The 280 and 4870 have guite near in memory bandwide. The difference is that the ATI solution is cheaper to produce...Reply