AMD RyZen CPU Architecture for 2017

Looks like excellent multi-core usage in FH3.

What is the SMT 1-8? Is your SMT turned off?

No, I just separated Cores and SMT threads to better understand how each application scales. For example, it doesn't look like Destiny 2 uses SMT at all, only cores:

destiny2screenshot201zaamh.jpg
 
Tested UE4 lightmap building with a simulated 1600 (3+3 from the BIOS with 2 cores deactivated) and the 1700. Pretty good bump, every setting was the same except core count:

1vbzpx.jpg


I've also enabled this from the experimental settings:

2u3le9.png
 
Just got done building a new Ryzen system, 1700. Have the CPU running at 3.9, and memory at 3600.
I can get the CPU to 4.0, but then for some reason the memory downclocks.

EDIT: set the memory to 3466 MHz, CPU at 4.0, temps do not exceed 63 C in prime95, or Handbrake.
 
Last edited:
Ryzen Threadripper was secretly developed by engineers in their spare time without the company's knowledge and they were so impressed. Now I know why there is a difference between the AMD CPU side and AMD GPU, it's like there are 2 AMDs.

https://www.computing.co.uk/ctg/new...ped-threadripper-cpu-as-a-skunk-works-project
That sounds like Assassin's Creed which was supposed to be a Prince of Persia but was secretly turned into the game we know now. ^^
Management should trust some of their people more.
 
Ryzen Threadripper was secretly developed by engineers in their spare time without the company's knowledge and they were so impressed. Now I know why there is a difference between the AMD CPU side and AMD GPU, it's like there are 2 AMDs.

https://www.computing.co.uk/ctg/new...ped-threadripper-cpu-as-a-skunk-works-project
Actually no, it doesn't. We don't know how many possible "skunkworks" ideas there has been on the GPU front, not before one becomes huge success, they won't tell about them without that part
 
Article is paywalled, are they saying there had been no plan for 16 core version of Epyc? Or just that some engineers took that & messed with desktopising it.
 
Article is paywalled, are they saying there had been no plan for 16 core version of Epyc? Or just that some engineers took that & messed with desktopising it.
Maybe there had been no plan for dummy cores.
 
Article is paywalled, are they saying there had been no plan for 16 core version of Epyc? Or just that some engineers took that & messed with desktopising it.


The latter. There was a free version of the (same, entire) article I read it somewhere, but now cant find it anymore.
 
Ryzen Threadripper was secretly developed by engineers in their spare time without the company's knowledge and they were so impressed. Now I know why there is a difference between the AMD CPU side and AMD GPU, it's like there are 2 AMDs.

https://www.computing.co.uk/ctg/new...ped-threadripper-cpu-as-a-skunk-works-project

The biggest different is the marketing department, recently AMD radeon twitter account said there were 570 on stock and on "great price" of 280 dollars....Even I had to comment asking what was wrong with them.
 
I was pretty sure that was the case. That's why I didn't understand the connection between Epyc and 2-die Threadripper.

The architectural features that allow Ryzen to connect four dies allowed them to link two dies with comparatively little effort. While they were at it, they reused elements like the general socket format from EPYC, which likely explains why Threadripper inherited the mechanical requirements of having two placeholder dies.

It would have been harder to get preliminary work as far as it did if the overall architecture needed changes and real investment, made the business case harder once presented, and it would have taken longer to roll out. In corporate terms, it was more spontaneously put together than EPYC and Ryzen because Threadripper naturally falls out of the system's capabilities.
 
The architectural features that allow Ryzen to connect four dies allowed them to link two dies with comparatively little effort. While they were at it, they reused elements like the general socket format from EPYC, which likely explains why Threadripper inherited the mechanical requirements of having two placeholder dies.

It would have been harder to get preliminary work as far as it did if the overall architecture needed changes and real investment, made the business case harder once presented, and it would have taken longer to roll out. In corporate terms, it was more spontaneously put together than EPYC and Ryzen because Threadripper naturally falls out of the system's capabilities.
Yes also the timeframe given to show the project that were originally planned for 2018 and was moved a year to make the 2017 launch with in my opinion was the real game changer. Now that TR exists and its selling as good as it can be we can be sure of more refine design in the next generation, probably moving to a unique platform to make it easier and even cheaper to work and manufacture.

Its really rare to me that AMD didn't plan TR from the beginning since they were working on Epyc, TR just look like a natural path to me. Im exited to see how the next gen will look.
 
Back
Top