I can tell you right now that I/O and memory bandwidth is where you'll get 8x gains, not CPU speeds. ECC, ok, yes, spend money on that. But otherwise splurge for an NVMe SSD, long before spending $1 extra on the CPU. -Adam
On November 16, 2017 11:53:22 PM CST, Trevor Cordes trevor@tecnopolis.ca wrote:
On 2017-11-16 Kevin McGregor wrote:
The question is, which (types of) applications have sucky performance specifically because of this? It may be a perfectly reasonable design choice for some/many use cases.
Possibly. Adam's article actually c&p'ed the ryzen paragraphs from other (game-centric) ryzen articles (or vice-versa?).
However, none of the articles I read tell me what the upside is... they just list the downside (latency). I assume AMD chose bandwidth over latency, so one would hope the bw would be much higher on ryzen to compensate for latency. Then your app will dictate whether ryzen or xeon is better.
As Adam said, one would hope AMD made this design decision on purpose, rather than as a mistake. As a mistake, think lack of barrel shifter on P4's. No one said chip designers are perfect... so I wouldn't be too surprised either way.
I stopped caring about that "last tiny bit of performance" about 10 years ago. If my next workstation is around 8X faster than my current one for my workload, then I don't really care if it is actually 7X or 9X, as long as the price is semi-affordable. And I get ECC. And I get a x16/x16 slot for discrete graphics. _______________________________________________ Roundtable mailing list Roundtable@muug.ca https://muug.ca/mailman/listinfo/roundtable