We haven’t seen much information from NVIDIA regarding their Lovelace GPUs in a while. However, kopite recently shed some light on NVIDIA’s RTX 4080.
Let's make some updates.
16GB 23Gbps GDDR6X
total card power 340W
— kopite7kimi (@kopite7kimi) August 23, 2022
This seems nearly similar to our previous leaks where the TDPs for the 4070 and 4080 were reduced. Initially, the 4080 featured 10240 Cuda cores which was cut down to 9728 cores in another leak a while back. The main changes include a slightly higher total card power (340W vs the previously reduced 320W) and faster 23Gbps memory.
The RTX 4080 was rumoured to make use of 21Gbps memory, however, team green is bumping this number up to 23Gbps for 736GB/s of effective memory bandwidth.
|RTX 4080 New||RTX 4080 Old|
|23Gbps G6X Memory||21Gbps G6X Memory|
|340W total card power||320W total card power|
|Effective memory banwdith of 736GB/s||Effective memory banwdith of 672GB/s|
Some Performance Metrics
Kopite didn’t exactly provide us with a TSE score showing this may just be a rumour or a speculation. However, NVIDIA does change its specs every now and then. For the efficiency stats, we compiled a chart showing performance per watt ratios across various GPUs.
While this does put our initial expectations a bit lower. We should know that higher effective memory may allow for faster TSE scores. In our case, we took the power consumption as 340W and the TSE score as ~15000 points.
No official announcement has been made by NVIDIA regarding Lovelace. Two years back, at this very time we already knew about the announcement date for NVIDIA’s Ampere. However, fast forward to today, we are still waiting for NVIDIA’s signal.