Equal clock comparison R5 3400G vs R5 2400G
We compared the 3400G vs the 2400G on the same clocks (CPU, GPU and RAM) to see if there is some „secret sauce“ inside the new silicon.
The System we are using is:
System Info:
R5 3400G Vega @ 1240 mhz
R5 2400G Vega @ 1240 mhz
Corsair RAM 16 GB @ 2933-17-19-19-19-39
Asrock AB350 Pro4
Crucial P1 500GB NVME
Win 10 / 1903
Driver: AMD Radeon 19.10.22.03
The following results have been appeared:
| Programm/App | 2400G | 3400G | % change+ |
|---|---|---|---|
| Unigine Superposition | 5743 | 6201 | 7,9 |
| Unigine Valley | 977 | 1047 | 7,1 |
| Firestrike 1.1 | 3218 | 3245 | 0,8 |
| Timespy | 1192 | 1218 | 2,1 |
| Cinebench R15 | 769 | 788 | 2,4 |
| CPU-z | 415 / 2233 | 420 / 2262 | 1,2 |
| Shadow Of Mordor | 45,51 | 46,85 | 2,9 |
| Basemark 1.1-DX12 | 874 | 910 | 4,1 |
| Aida64 GPGPU | See picture below | See picture below | See picture below |
We think the double memory-read in the Aid64 GPGPU test is a bug in the program (but it is still not recognizing the „same“ GPU-part of it)
Final words:
At the moment we think there have been made some slightly changes to the 3400G GPU-part. If all results would have been inside the error margin of (0-2%) then we would not talk about a silicon change. But since we have sometimes over 5% performance in Unigine benchmarks, for example, there need to be smoething going on at the new APUs.
Additionaly there are some small hints it can´t be 100% the same silicon:
- Why does the Vega 11 does not work with the old 2400G driver?
- Why does Aida64 does not clearly recognize the GPU part?
- Why does it have slightly faster performance?
We asked an AMD employee and hope to get an answer soon 🙂
Dont forget to check out our video to this blogpost:
https://www.youtube.com/watch?v=GM8_J5tZeLc
