AMD '#Rekt son' reply to Intel's "four glued together" statement

[XenosTech]

4 minutes ago, leadeater said:

Well I can never pass up the opportunity to make a terrible joke or pun, I really am a terrible person lol.

 

But yea a graph of inter-core latency is utterly meaningless in the context of workload performance if you don't even know what the workload is and how it interacts with CPU resources. If there is never any data passing between cores or caches, other than from L2 to L1 in the same core, the latency could be 10 seconds and it wouldn't matter (stupid extreme used on purpose).

 

That's also ignoring thins like OS processor queue length like @System Error Message was talking about at a slightly lower level, hardware wise.

Yeah pretty much, so EPYC would perform just fine in data centers where most of the data is cached somewhere waiting to be accessed by a cpu or gpu. Core latency may somewhat matter maybe in the consumer space where we don't have elaborate cache arrays setup to keep current/frequently used data sets that are being worked on.

[System Error Message]

8 minutes ago, XenosTech said:

Yeah pretty much, so EPYC would perform just fine in data centers where most of the data is cached somewhere waiting to be accessed by a cpu or gpu. Core latency may somewhat matter maybe in the consumer space where we don't have elaborate cache arrays setup to keep current/frequently used data sets that are being worked on.

core latency doesnt matter in consumer space either, remember the core2s, pentiums? the pentiums had horrible latency, the core2quads was 2 core2duos put together.

core to core latency doesnt really matter, what really matters is compiler optimisation, ensuring that the pipelines are fed in the most effective instruction ques where there are no conflicts and with every pipeline stage filled with useful work and not a noop. Because regardless of the workload, its gonna use registers, cache, RAM.

 

Whats important is that the CPU is fast, and this AMD CPU is fast.

[XenosTech]

Just now, System Error Message said:

core latency doesnt matter in consumer space either, remember the core2s, pentiums? the pentiums had horrible latency, the core2quads was 2 core2duos put together.

core to core latency doesnt really matter, what really matters is compiler optimisation, ensuring that the pipelines are fed in the most effective instruction ques where there are no conflicts and with every pipeline stage filled with useful work and not a noop. Because regardless of the workload, its gonna use registers, cache, RAM.

I've never had any of the core series processors. I pretty much had a pentium extreme processor up til recently for a daily driver pc for anything non gaming lol. People here make it seem as if only AMD needs software optimisations for their products. I guess Nvidia releases drivers ever so often for their health and software vendors like microsoft release updates for their software daily/weekly/monthly for their health also.

[Sunshine1868]

On 7/19/2017 at 7:34 PM, Jito463 said:

*shuttered  

_{(sorry, I couldn't resist)}

You don't waste good chewing gum!  Use the old stuff.  I'm sure I kept some under the couch.  Just need to soften it up a little.....

fixed

[Drak3]

On 7/19/2017 at 10:55 PM, leadeater said:

Well I can never pass up the opportunity to make a terrible joke or pun

Better than some forum members. One forum member.

 

At least your jokes and puns aren't the worst on the forum.

On 7/19/2017 at 10:55 PM, leadeater said:

I really am a terrible person

No, a horse is a terrible person.

 

On 7/19/2017 at 10:55 PM, leadeater said:

lol

*Place holder until I can post a picture that demonstrates my feelings towards that acronym*

[Jito463]

1 hour ago, Drak3 said:

*Place holder until I can post a picture that demonstrates my feelings towards that acronym*