GDDR7 memory technical specifications approved, when will the first graphics cards be available?

The wait won't be too long. The next generations of GeForce and Radeon will definitely take advantage of this new VRAM.

GDDR7 has been in the news for several months now, mostly through Samsung. But JEDEC (Joint Council on Electronics Engineering) has just published the official specifications of this standard. It is likely that this memory will be used by AMD and NVIDIA for their next generation GPUs: RDNA 4 (Radeon RX 8000) and Blackwell (GeForce RTX 50 Series). at least for the most high-end models.

GDDR6 at its best

Today, all graphics cards use GDDR6 memory. This version, a logical successor to GDDR5, debuted in 2018 during the marketing of the GeForce RTX 20 (Turing) series. Speeds were initially 14 Gbit/s. have continued to increase over the years. NVIDIA now uses GDDR6X from Micron. It delivers 23 Gbit/s in the case of the GeForce RTX 4080 SUPER. Radeon RX 7900 XTX / XT's GDDR6 peaks at 20 Gbit/s.

GDDR7 will drastically increase speeds. so by extension, the memory bandwidth of graphics cards. Manufacturers have repeatedly reported 32 Gbit/s for the first generation of chips. However, they estimate that 36 Gbit/s will be a fairly easily achievable value.

The JEDEC publication contains most of the data known so far. However, the organization goes further in terms of bandwidth: it says a maximum of 192 GB/s per device or 48 Gbit/s per pin. To make a head-to-head comparison with GDDR6, note that chips clocked at 14 Gbit/s offer 56 GB/s. to 24 Gbit/s – the theoretical maximum for GDDR6 – from 96 GB/s.

To give a concrete example, with a 256-bit bus, GDDR7 at 32 Git/s results in a memory bandwidth of 1024 GB/s. With GDDR6X at 24 Gbit/s, this is 768 GB/s (the GeForce RTX 4080 SUPER, which uses such an interface, has a bandwidth of 736.3 GB/s).

PAM-3; Is that what ?

The magic potion of such flow rates is due in part to this formula: PAM-3 (pulse width modulation, or just pulse width modulation in French). This means that GDDR7 uses three signaling levels (-1, 0, +1) to transmit three bits of data in two cycles.

GDDR6 uses NRZ (non-return to zero) signaling and GDDR6X uses PAM-4 signaling. The first involves the transmission of one bit per cycle, the second two bits per cycle. Therefore, GDDR7 transmits three bits in two cycles, or 50% more data than GDDR6 operating at the same frequency. If you're wondering why PAM-4 isn't more widely used, it's because it's more expensive and complex to implement.

In addition, the DRAM reference voltage is increased to 1.2 V instead of 1.35 V. The maximum density, limited to 32 Gb for GDDR6/GDDR6X, is doubled to 64 Gb. the same for the number of independent channels (from 2 to 4).

As written at the beginning of the article, it is very likely that GDDR7 will be released from the flagship GeForce RTX 50 Series and RX 8000 within a few months.

πŸ”΄ To not miss any news from 01net, follow us on Google News and WhatsApp.



Leave a Reply

Your email address will not be published. Required fields are marked *