- Mar 3, 2017
- 1,659
- 6,101
- 136
ah, again, you got like 25% off strix or something there. good nuff for proper perf modeling.I was talking about Geekbench, as easily gameable as it is
LOL! Not really - lots of hard work to get it right.of course. child's play.
No, some workloads respond pretty well to core bloat. others less so.LOL! Not really - lots of hard work to get it right.
This means Zen 5 officially launches before August 25th.Looks like Zen 5 will be discussed at this year's Hot Chips by Brad Cohen and, of course, Mike Clark.
https://hotchips.org/advance-program/
earlier than many-many people even expect kek.This means Zen 5 officially launches before August 25th.
Of course you are right. I'm sure after a month the simulations at max frequency were showing zero errors.No, some workloads respond pretty well to core bloat. others less so.
Gaming loves fat ROBs and all.
earlier than many-many people even expect kek.
Wait a tiny little bit and learn Mandarin.Oh yeah? Which month?
They don't have to do anything since they reserved Computex opening keynote.Another in long series of (pre) announcements today. In comparison, nada from AMD on Zen 5, Strix Point.
Wait a tiny little bit and learn Mandarin.
They don't have to do anything since they reserved Computex opening keynote.
They didn't? I don't remember them doing anything particularly sus perf-wise. Power OTOH.I noticed someone on Twitter point out why Qualcomm is hiding performance details, playing shell games...
+20-25% multithreaded improvement would be disappointing imho. Many people including me would be happy not paying the zen5 tax and simply buy a discounted 7950X(3D), or an equivalent Intel.see? this one is close.
They didn't? I don't remember them doing anything particularly sus perf-wise. Power OTOH.
It doesn't matter.+20-25% multithreaded improvement would be disappointing imho
That's the only metric that really matters in client CPUs.Not everyone spends hundreds of $ to brag about single core scores
A whole lot higher.a bit higher than those of an ipad
earlier than many-many people even expect kek.
ST score is what allows your PC not to choke on the latest Slack update.+20-25% multithreaded improvement would be disappointing imho. Many people including me would be happy not paying the zen5 tax and simply buy a discounted 7950X(3D), or an equivalent Intel.
So hopefully the MT improvement is very close to the ST one, or it's at least reachable with a bit of undervolting and overclocking. Not everyone spends hundreds of $ to brag about single core scores a bit higher than those of an ipad (or to game at uselessly high fps).
It matters to me and every single person I know irl that uses a desktop or workstation.It doesn't matter.
You keep saying that as if it's some great religious truth, but it's just your opinion. It's not an unreasonable opinion, but still just an opinion.That's the only metric that really matters in client CPUs.
+20-25% multithreaded improvement would be disappointing imho. Many people including me would be happy not paying the zen5 tax and simply buy a discounted 7950X(3D), or an equivalent Intel.
So hopefully the MT improvement is very close to the ST one, or it's at least reachable with a bit of undervolting and overclocking. Not everyone spends hundreds of $ to brag about single core scores a bit higher than those of an ipad (or to game at uselessly high fps).
Memory bandwidth and power limits (depending on SKU).Why would it have a more limited improvement compared to ST?
Why would it have a more limited improvement compared to ST?
Desktop or workstation is a tiny-tiny part of the PC TAM.It matters to me and every single person I know irl that uses a desktop or workstation.
Because it is one.You keep saying that as if it's some great religious truth
Because the core is maxxed around peak 1t perf.Why would it have a more limited improvement compared to ST?
you forgot the 5) abject death of SMT scaling due to core being 1t-maxxed.1) Power limits are hit, this is usually true especially when the new gen is on a similar node as the last gen. This happened in the Zen 2 -> Zen 3 transition.
2) Cache. Depending on the cache structure, when all cores are fired up, it can put more pressure on any shared cache.
3) Linked to 2, memory bandwidth saturation. With higher performance cores, many MT threads require increased memory bandwidth. If there isn't a sufficient increase in memory BW, then the MT increase will be limited.
4) Linked to 1, temperature. Obviously this one has several variables involved and is usually not really a consideration given that the end customer can use better cooling to mitigate it.
you forgot the 5) abject death of SMT scaling due to core being 1t-maxxed.
+20-25% multithreaded improvement would be disappointing imho. Many people including me would be happy not paying the zen5 tax and simply buy a discounted 7950X(3D), or an equivalent Intel.