Have been benching the performance difference between "pure T1", T1 GDM and T2 in the Monero miner which is known to scale very well with both speed (bandwidth) and timings (latency).
The goal was to decide once and for all whether T1 GDM actually is slower than T2 due to half speed on mux .. But this benchmark can also be used to test whether you actually scale with higher flck speeds above 1900mhz (seems like hardly anyone do in real benchmarks otherthen Aida64)
Settings:
- 5950x @ 4700/4600 static OC, SMT enabled
- 4x8GB memory sticks
- Hwinfo open for all runs
- consecutive runs
- Flat CL 14-14-14-14 timings
T1 setup-time
19723.4 H/S over 15min run
upto 627 H/S per core
T1 GDM
19151 H/S over 15min run
upto 606 H/S per core
T2
19413 H/S over 15min run
upto 616 H/S per core
- 19413 / 19151 = T2 is ~1.3% faster than T1 GDM in this benchmark
- 19723 / 19413 = T1 setup-time is ~1.6% faster than T2 in this benchmark
- 19723 / 19151 = T1 setup-time is ~2.9% faster than T1 GDM in this benchmark
So I come to the conclusion that T2 is faster than T1 GDM .. But this is also a bit dependent on the fact that you actually have stable T2 settings.
T2 is more difficult to stabilize than T1 GDM as the latter "smooths over" / arranges and/or changes timing-mismatches, so if you want to retrieve these last 1.3% memory performance percentages then be prepared to spend many hours fine-tuning your settings .. For the regular user is it rather doubtful that this is worth it, as the time is probably better spent maximizing CPU overclocking in my eyes, as virtually everything scales with higher CPU speed..
I can also say that the degree of difficulty to stabilize "pure T1" is in a another league, especially with dual rank. (32GB or more)
Aida64 screenshot of my fastest T1 setup-time with my new 24/7 settings for CTR. (notice how straight the CPU mhz line is in 3dmark Timespy with this new version of CTR)
If anyone else wants to test / bench with the same mining program:
Instructions for installing monero can be found here