الاثنين، 19 سبتمبر 2022

Nvidia’s Latest 40-Series Teaser Hints at Jaw-Dropping Numbers

We’re one day away from the launch of Nvidia’s next-generation GPUs, and a new teaser has successfully raised an eyebrow or two. The company is trying to create a hype train leading up to the launch of its 40-series GPUs by dropping hints on Twitter. Previously it had only released a name — Project Beyond — but its latest teaser includes some numbers. Nobody knows for sure what those numbers mean, but there are some educated guesses. If those guesses turn out to be true, we could be in for a very powerful GPU indeed.

The official Nvidia GeForce Twitter account began dropping clues about the impending launch a few days ago. The first clue was a simple teaser saying its name and two more hints. The first is titled, “The future has a past,” and it shows documents laid out on a table. Upon close inspection, it’s a chart titled, “Diagram for the computation by the Engine of the Numbers of Bernoulli,” according to Videocardz. The site posted a higher-res version, which is known to be the first documented computer “program.” It was written by none other than Ada Lovelace in 1843. This confirms Nvidia is launching its “Ada” architecture on Sept. 20. We all figured that was what was happening, but now it’s confirmed.

The next teaser is even more cryptic, as it shows the monitor above the previously disclosed chart. It includes a post-it note with a phone number on it: 208-629-7538. It consists of the tagline, “this number will self-destruct.” As far as what it means, one guess is that it hints at the chip’s specs and performance. The “208” could mean a 208 percent uplift from RTX 3090. That’s been rumored previously, but it’s hard to believe for obvious reasons. Nvidia hasn’t offered that level of performance from one generation to the next in quite some time, if ever. That was definitely not the case going from Turing to Ampere.

The following number — 629 — could be the size in millimeters of the AD102 die, which is the full chip. Interestingly, AD102 from Ampere was the same size, except it was built on Samsung’s 8nm process. That means Nvidia used the smaller TSMC 4nm process for Ada but kept the die the same size. That would allow for a lot more transistors. Ampere’s GA102 packed 28.3 billion transistors into that die, so what does that mean for Ada?

That leads us to the final number: 7538. That could be the transistor count in billions of the full chip, 75.38 billion. If that’s true that means it’s tripled from Ampere to Ada, which would be an unprecedented leap. For reference, the RDNA2 die from AMD was 520mm with 26.8 billion transistors, according to TechpowerUp’s database. A chip with 75.38 billion transistors would be, frankly, insane.

Dialing the number reveals a computerized voice asking, “Tell us, how fast would you like to go?” We’ll go with 4K, 144Hz, please.

All will be revealed tomorrow at GTC, as Jensen is expected to take the stage on Tuesday at 8 a.m. PST.

Now Read:



sourse ExtremeTechExtremeTech https://ift.tt/8ex6OUq

ليست هناك تعليقات:

إرسال تعليق