AMD: ZEN3 architecture finished – expects 15% faster IPC
Click here to post a comment for AMD: ZEN3 architecture finished – expects 15% faster IPC on our message forum
wavetrex
Great !
Keep it going.
The world was stuck for too long in CPU near-stagnation... it's time to see massive leaps for CPUs like we did for GPUs in the past 10-12 years.
fantaskarsef
Please AMD, give me this in H1/2020 (won't happen, I know), and I'll gladly ditch Intel as soon as I can get my hands on this stuff. Even if there's a fan on the chipset for no reason (I don't need PCIe4)
Stormyandcold
If they want to take the market, then, such performance improvements will definitely raise more eyebrows.
Astyanax
5Ghz will not save intel now, Zen 3 will be faster in games at lower clocks.
wavetrex
slyphnier
so Zen3 still DDR4 not DDR5 right? thus they said am4 support zen3? cmiiw
usually the more performance increase the more likely it will move to new platform
Undying
For the first time in 10 years im not even looking at the Intel cpus i dont even know the names of their 10nm chips thats how uninteresting the've become. Despite everything Intel still stays on top but thats about to change it seems.
airbud7
the price of success is hard work and dedication....them folks at AMD are not playing around!
magiQx
cryohellinc
Great news!
Richard Nutman
Borys
ZEN3 will be the Intel "end of line"
Zen2 is already faster with the same clock. Zen3 will get higher clocks, better memory latency and will be more stable (talking about voltage)
In 2 years the CPU marketshare will be 50-50%... now Intel just remains to cry man...
Denial
I don't get where they say they expect a 15% increase?
This line suggests the 15% for Zen 2 was an anomaly - that zen 3 would be what you would normally expect. I normally expect about 5% tbh.
anticupidon
Well, good.
But IF only software will be coded so well that it takes fully advantage of there improvements.
Don't throw rocks at me yet!
Let's not forget the Threadripper and Windows scheduler issues!
Hardware is nothing with proper software, and you know that.
Let's hope that new AI and new software compilers will be able to use all the hardware instructions.
NewTRUMP Order
The original estimate was
The original estimate was 8% improvement. Now throw in beefier L2 - L3 cache sizes, some "secret" architecture changes, TSMC 7+ node and 15% doesn't sound too far fetched.
HybOj
Denial
HybOj
Kaotik
Denial