Update 5/3/2020 -> Another new NVIDIA graphics card has just been discovered in the GeekBench benchmark database, this time with 124 computing units (CUs). Which in turn should mean the presence of 7936 CUDA cores, if the processor count remains the same for each computing unit. (It has changed a few times in the past).
In addition, it appears that the graphics card has 32GB of HBMe2 memory, and the base frequency of 1.1GHz. The result was 222,377 points.
Once again we mention that these units are probably samples of engineering, so the performance could be significantly higher in the final product.
Original News -> This is the classic NVIDIA … When a guy is on a weekend, and is really ready to not tip … Here are a series of ‘leaks’ and rumors about the new 7nm NVIDIA Ampere graphics cards! Lifting the veil over the base specifications, the expected performance, among other interesting things.
New NVIDIA Ampere cards: 8192 cores and 24 / 48GB of memory !?
So, as you may know, the next NVIDIA architecture is called Ampere, and right after that we must have Hopper! But calm down, let's focus on what's about to hit the market.
Well, the new NVIDIA Ampere are based on the 7nm process. And as such, it looks like they will bring about a serious performance increase, as well as a new level of energy efficiency.
As we said yesterday, two Geekbench results were found for two new NVIDIA graphics cards. One with 7552 CUDA processing cores, the frequency of 1.11GHz and 24GB of memory (GDDR6 or HMB2e).
Then, we also had a glimpse of the performance of another card with 6912 CUDA cores, the frequency of 1.01GHz, and an incredible 48GB of memory. However, these cards that already impress, are not even the top of the range!
NVIDIA Ampere (known GPUs):
- NVIDIA Ampere GPU 1: 6912 colors
- Ampere GPU 2: 7552 colors
- Ampere GPU 3: 8192 colors (Full Version)
As we said in the past, these benchmarks are clearly made on Tesla boards. In other words, we are not even talking about the Quadro range, let alone the GeForce range. That said, with this level of performance, the Ampere architecture puts the current Tesla V100 in your pocket. Now see:
Geekbench 5 results
- GPU Ampere 1: 184,096
- GPU Ampere 2: 141,654
- Tesla V100: 154,606
- NVIDIA GV100: 142,837
- NVIDIA TITAN RTX: 132,804
Taking these results into account, TITAN RTX currently costs 2500 $, being the most expensive Turing card from NVIDIA. That is, it costs 1000 $ more than the super popular RTX 2080 Ti.
However, even the weakest Ampere board discovered in this 'leak' overtakes the TITAN RTX with some ease. With the version of 7552 cores reaching an astonishing 184,096 points. (Attention, there is still a better version than this one!).
Still, if you're already super excited about the arrival of the new RTX 3000… Be calm! None of these GPUs will bring a new GeForce card to life. However, these figures give us an idea of the performance increase that we will see in the coming months. And my friends, it won't be low! AMD Radeon take care, that I don't see ‘Big Navi’ rivaling this.
Furthermore, what do you think about all this? Share your opinion with us in the comments below.