1T was also unattainable for stable running – in the end we could only squeeze out 3-3-3-10-18-2T from our memory which is nearly the fastest we can run it at, although not quite.
The Everest results reflect this with a marginally slower latency and slightly lower raw memory performance. The SiSoft results are a little different, showing an equally high raw memory performance but very low memory latency.
Prev
