Renoir shows increased IF performance

Simbelmynë · July 1, 2020, 04:59

So it seems that Renoir can reach 2100 MHz on the IF. This permits 4200 MT/s memory in 1:1 mode and much lower latencies. For EPYC users this is not too exiting, but for budget users (threadripper and below), or people limited by licensing, it may give a substantial performance boost in the coming releases.

Mostly rumors at this stage, but fun nonetheless.

flotus1 · July 1, 2020, 05:21

Isn't Renoir AMDs codename for Zen2 APUs? IIRC, they feature significantly less L3 cache than regular Zen2 CPUs without integrated graphics. 4MB for 4 core models, and 8MB for 6+8 cores. Compared to 32MB on a Ryzen 5 3600.
L3 cache size is one of the reasons that makes Zen2 so interesting. Compared to Intel, memory latency is still on the slow side. The brute-force way to hide that to a certain extent is larger caches.

Simbelmynë · July 1, 2020, 05:29

Yeah. The cache size is just one quarter of standard Zen 2. Some sources point to that being the reason for the increased IF speed. However, there are some discussions on whether this can also be due to the refined process (although likely it is the smaller cache as you point out). Anyways, I think it will be interesting to see how the APU performs in our benchmark. Some rumors talk of a 4200 MT/s system with CL14 (!).

Edit: May also be due to the monolithic design of the 4000 series APUs?

July 1, 2020, 04:59	Renoir shows increased IF performance	#1
Simbelmynë Senior Member Join Date: May 2012 Posts: 546 Rep Power: 15	So it seems that Renoir can reach 2100 MHz on the IF. This permits 4200 MT/s memory in 1:1 mode and much lower latencies. For EPYC users this is not too exiting, but for budget users (threadripper and below), or people limited by licensing, it may give a substantial performance boost in the coming releases. Mostly rumors at this stage, but fun nonetheless.

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Suspicious Harware Performance	CaptainCombo	ANSYS	0	September 4, 2018 08:34
Abysmal performance of 64 cores opteron based workstation for CFD	Fauster	Hardware	8	June 4, 2018 10:51
Scaling from 1 to 2 nodes shows 122% performance increase	evcelica	Hardware	5	October 31, 2012 10:58
parallel performance on BX900	uzawa	OpenFOAM Installation	3	September 5, 2011 15:52
Linux kernel performance enhancement	ConnClark	OpenFOAM	0	January 17, 2011 18:18

July 1, 2020, 05:21		#2
flotus1 Super Moderator Alex Join Date: Jun 2012 Location: Germany Posts: 3,399 Rep Power: 46	Isn't Renoir AMDs codename for Zen2 APUs? IIRC, they feature significantly less L3 cache than regular Zen2 CPUs without integrated graphics. 4MB for 4 core models, and 8MB for 6+8 cores. Compared to 32MB on a Ryzen 5 3600. L3 cache size is one of the reasons that makes Zen2 so interesting. Compared to Intel, memory latency is still on the slow side. The brute-force way to hide that to a certain extent is larger caches.

July 1, 2020, 05:29		#3
Simbelmynë Senior Member Join Date: May 2012 Posts: 546 Rep Power: 15	Yeah. The cache size is just one quarter of standard Zen 2. Some sources point to that being the reason for the increased IF speed. However, there are some discussions on whether this can also be due to the refined process (although likely it is the smaller cache as you point out). Anyways, I think it will be interesting to see how the APU performs in our benchmark. Some rumors talk of a 4200 MT/s system with CL14 (!). Edit: May also be due to the monolithic design of the 4000 series APUs?