TR Forums

Something I wondered about is if all 4 cores on a CCX package ever become L3 limited. I mean we are talking 8MB of L3 per CCX, which is very generous. But does the 3+3 setup for the 6c/12t parts like the 1600x afford those 3 cores any benefit from only having 3 cores instead of 4 to share that 8MB with per CCX? Could we see benefits in for instance in games that take real advantage of L3 cache if it isnt saturated. I guess the argument could become,

3 Cores (8MB L3) < 4 Cores (8MB L3)

Is there ever a point where having 1 less core and more L3 would be a benefit? Perhaps if the coders were lazy and left certain game assets to be cached that aren't called on in a predictive way but randomly need to be called on very quickly?

Any insight on programs that are L3 heavy?

1) Most games aren't going to take advantage of more than 6 threads anyway. More commonly that limit is 4 threads. IPC is still king. We can see that in the Ryzen 1800X reviews. Despite having twice the cores as the i7-7700K it still loses. Even the quad core i5s hold their own against the lineup with their only significant downfall being their slightly lower clocks. Regardless, I'd bet AMD is distributing threads out evenly over L3 pools to maximize cache availability.

2) Check out this review of the illusive i7-5775C and it's 128MB L4 cache. More cache can make a difference.

Welch wrote:

Something I wondered about is if all 4 cores on a CCX package ever become L3 limited. I mean we are talking 8MB of L3 per CCX, which is very generous. But does the 3+3 setup for the 6c/12t parts like the 1600x afford those 3 cores any benefit from only having 3 cores instead of 4 to share that 8MB with per CCX? Could we see benefits in for instance in games that take real advantage of L3 cache if it isnt saturated. I guess the argument could become,

3 Cores (8MB L3) < 4 Cores (8MB L3)

Is there ever a point where having 1 less core and more L3 would be a benefit? Perhaps if the coders were lazy and left certain game assets to be cached that aren't called on in a predictive way but randomly need to be called on very quickly?

Any insight on programs that are L3 heavy?

More likely, in the case of Ryzens arch it uses the extra cache the same either way assuming the other cores aren't being used.