Nvidia's upcoming player flagship Geforce RTX 3080 Ti based on the ampere architecture should deliver 21 Teraflops computing power and more. This and other detailed information can be found in a recent contribution from the YouTube channel Moore's Law Is Dead.
However, the alleged leak also comes up against criticism: According to 3DCenter, it is essentially known rumors that have only been evaluated. So what's with the latest information and what is really new about it?
Moore's Law Is Dead claims to be based on statements by an Nvidia engineer "Independent" Sources. This could be the well-known hardware leaker KittyKitties (formerly CorgiKitty) and kopite7kimi meant, who gave details of several ampere chips some time ago.
What does the alleged leak say about the RTX 3080 Ti?
- Shader units & power consumption: The full expansion of the GA102 chip should have 5,376 CUDA cores. An engineering sample of the RTX 3080 Ti also has 220 to 230 watts of power added.
- Clock rate & Teraflops: A boost clock rate of 2.2 GHz was achieved with the card. Thus, even the cropped variant of the GA102 chip should surpass 21 teraflops of computing power.
- RTX 3080 & RTX 3070: The GA103 & GA104 chips, which may be intended for smaller models, are even said to boost at 2.5 GHz. However, the engineer did not confirm this and was only expected. Moore's Law Is Dead sums up that 4K should have arrived as a standard in the middle class and that only the CPU is the bottleneck at 1080p & 1440p resolution.
- Video memory: GDDR6 with 18 Gbps (18 GHz effective memory clock) and a 384 bit wide interface should be used. This achieves a memory bandwidth of 864 GB / s. The RTX 2080 Ti would thus be exceeded by 40.3 percent.
- Comparison with RTX 2080 Ti: The RTX 3080 Ti is said to match the rasterization performance of the RTX 2080 Ti with 4K resolution "worst case" surpass by 40 percent. Usually the GA102 chip is 50 percent faster than the TU102 of the 2080 Ti, in some games even 70 percent.
- Raytracing: The number of so-called RT (ray tracing) cores should not necessarily be increased, but the performance of the individual cores will be significantly increased. Moore's Law Is Dead says that the intersections can be processed four times faster than with the Turing architecture (RTX-2000). Ampere should be able to calculate all common ray tracing effects such as global lighting, shadows and reflections at the same time and cost less performance than is the case with the RTX 2000.
Furthermore, some more Details about the equipment of the RTX 3080 Ti called:
- simplified design with fewer screws
- 3x DisplayPort 2.0
- 1x HDMI 2.1
- 1x USB-C
- PCI Express 4.0
How new and plausible is the leak?
With regard to the performance features, only the information on ray tracing performance and video memory is really new. So far there has been talk of an improvement in performance with activated real-time beam tracking and GDDR6 with 16 Gbps (16 GHz effective). Nevertheless, the information is relatively extensive and Moore's Law Is Dead explicitly refers to a source at Nvidia.
Overall, most of the specifications seem plausible – with the exception of the quite high-sounding boost clock rates of the GA103 and GA104 chips. However, they should still be enjoyed with due caution.
More and above all Tangible to amp architecture is expected as part of the subsequent keynote on the GPU Technology Conference (GTC) 2020 canceled due to the corona crisis, for which Nvidia CEO Jensen Huang has published an unusual teaser video:
You can see the keynote on Thursday (May 14th) from 3 p.m. on the Nvidia YouTube channel. While it will most likely be about professional cards or server products, certain conclusions about new player GPUs could be possible.
We will summarize all important information about the keynote afterwards for you on GameStar.de.