Within the present battle over GPU reminiscence speeds between AMD and Nvidia, Micron and Nvidia scored a decisive victory. Collectively the businesses made GDDR6X, which tops out round 21Gb/s. That reminiscence discovered its method to Nvidia’s high-end GPUs, named the RTX 3080 and 3090. AMD, in the meantime, used slower 16Gb/s modules on its RX 6900 XT, however then bumped that as much as 18Gb/s for the upgraded RX 6950 XT. As you’ll be able to see, there’s a reasonably clear delta there, though AMD’s choices had been no slouches because of Infinity Cache. Now Samsung has one-upped Micron, and has developed the world’s first 24Gb/s GDDR6. Be aware it’s not making use of an X to the RAM’s branding, so it’s 5 % much less cool-sounding.
Samsung made the announcement in a press launch. It says it’s designed particularly for “next-generation, high-end graphics.” Apparently, it didn’t title any clients, however it appears unlikely it might have unveiled it with out having a minimum of one large buyer already signed. Samsung says it was capable of hit these speeds because of the mix of a brand new circuit design and superior insulating materials. Not like its competitor, GDDR6X, GDDR6 is JEDEC-compliant. Samsung used its third technology 10nm course of together with excessive ultraviolet (EUV) expertise within the manufacturing course of.
Samsung is already sampling the brand new chips in 16Gb capability to unnamed corporations. It says its clients will being validating them within the close to future. This timing is intriguing since each AMD and Nvidia will launch next-gen GPUs quickly as properly. It looks as if a no brainer that AMD would undertake it, however it’s unknown if Micron and Nvidia have one thing like GDDR7 within the works. Both manner, if Samsung’s providing is quicker than no matter Nvidia is testing, it might possible change camps to acquire it. Earlier rumors indicated Nvidia had GDDR7 at 24Gb/s for its 40-series, however that’s nonetheless only a rumor.
The superior insulating materials Samung is utilizing is a Excessive-Okay Metallic Gate (HKMG), which prevents present leakage. This permits the IC to be extra environment friendly, and function at decrease temperatures. Samsung says the brand new modules help 1.1V of energy, versus the trade normal of 1.35/125V. It utilizing a brand new dynamic energy switching expertise to change backwards and forwards when essential. General, this permits for a 20 % discount in energy consumption in comparison with its earlier 18Gb/s modules. It’s additionally 30 % quicker, which permits for a most of 1.1TB/s of bandwidth.
This newest information from Samsung will definitely up the ante within the upcoming GPU wars. For this technology of playing cards (RDND2 vs. Ampere) AMD lagged behind Nvidia on the spec sheet, however largely made up the distinction with its Infinity Cache. Nvidia is rumored to be taking a web page from this playbook for its 40-series Ada Lovelace structure. It’s been speculated Nvidia will crank up L2 by 16x over Ampere. For its half, AMD is likely to be doubling down on Infinity Cache, utilizing 3D-stacking to actually double its measurement. That quantity of cache driving alongside (or on prime of) super-fast GDDR6 reminiscence might be a potent mixture certainly. For now, we should look ahead to the opposite shoe to drop. Will the primary buyer be AMD, Nvidia, or Intel?