Zen 2 vs Zen 3 improvement for DC?
- Thread starter mose1999
- Start date
Neither. They both wait for non existent work units at the same general power consumption.
What work do you care to perform? Different projects limit on different aspects of hardware. Some tasks are able to fully saturate all execution units and not be cache or memory bandwidth bound, and rely simply on core count, IPC, and cooling. Others tend cache or memory bandwidth bound and one gains no performance by adding additional cores, and in some cases even loses performance.
To answer your question more directly, Zen 3 is likely to be more efficient in terms of work out per unit power in, yet, it may or may not be worth the cost delta depending on your system costs, power costs, etc.
What work do you care to perform? Different projects limit on different aspects of hardware. Some tasks are able to fully saturate all execution units and not be cache or memory bandwidth bound, and rely simply on core count, IPC, and cooling. Others tend cache or memory bandwidth bound and one gains no performance by adding additional cores, and in some cases even loses performance.
To answer your question more directly, Zen 3 is likely to be more efficient in terms of work out per unit power in, yet, it may or may not be worth the cost delta depending on your system costs, power costs, etc.
Zen 3 averages 15% more instruction per clock over Zen 2. I have not seen benchmarks for distributed computing.