• Guest, The rules for the P & N subforum have been updated to prohibit "ad hominem" or personal attacks against other posters. See the full details in the post "Politics and News Rules & Guidelines."

Info Ryzen 4000 Mobile Chips Unveiled at CES

Page 9 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

naukkis

Senior member
Jun 5, 2002
298
137
116
Still I'd hoped that the effect would be minimal due to the integrated memory controller.
It can be if memory latency is lower and bandwidth higher than with Matisse. In those test you linked both of those are far inferior to that Matisse with 3600Mhz ram.
 
  • Like
Reactions: uzzi38

uzzi38

Senior member
Oct 16, 2019
532
506
96
It can be if memory latency is lower and bandwidth higher than with Matisse. In those test you linked both of those are far inferior to that Matisse with 3600Mhz ram.
Which is my point. If the latency is worse against the Matisse sample, then yes of course Renoir will be worse than Matisse for SC performance. It clocks lower and has less L3 cache.

Single channel + 2667mhz DDR4 is a horrible sample to be making comparisons with.
 

tamz_msc

Platinum Member
Jan 5, 2017
2,443
2,112
106
Which is my point. If the latency is worse against the Matisse sample, then yes of course Renoir will be worse than Matisse for SC performance. It clocks lower and has less L3 cache.

Single channel + 2667mhz DDR4 is a horrible sample to be making comparisons with.
It's the only sample we have at the moment.
 

teejee

Senior member
Jul 4, 2013
283
78
101
Looks like Geekbench 4 results are out:
It boosts to ~4.2GHz in this run. Here is a 3700X at ~4.4GHz for comparison:
As you can see integer score is up roughly 10% from a 5% increase in clock speed. To me it seems like Renoir is losing IPC compared to Matisse.

:confused:
You compare a 15w laptop chip with 3700x...

Compare with 3700u instead, they have GB4 scores between 3500-4500 (I checked about a dozen ones). Most of them just below 4000. So GB4 score of 4900 for a 4700u with very low memory speed is a very good improvement.
 
  • Like
Reactions: killster1

lobz

Golden Member
Feb 10, 2017
1,011
813
106
Looks like Geekbench 4 results are out:
It boosts to ~4.2GHz in this run. Here is a 3700X at ~4.4GHz for comparison:
As you can see integer score is up roughly 10% from a 5% increase in clock speed. To me it seems like Renoir is losing IPC compared to Matisse.

:confused:
I never understood how some of you here thought that a decrease of memory latency (no matter how significant) could offset tens of Megabytes of missing L3 cache - especially in an APU.
 

tamz_msc

Platinum Member
Jan 5, 2017
2,443
2,112
106
You compare a 15w laptop chip with 3700x...

Compare with 3700u instead, they have GB4 scores between 3500-4500 (I checked about a dozen ones). Most of them just below 4000. So GB4 score of 4900 for a 4700u with very low memory speed is a very good improvement.
TDP is irrelevant if I'm interested in single-core.
 

uzzi38

Senior member
Oct 16, 2019
532
506
96
It's the only sample we have at the moment.

You're welcome :p

If I was more bothered I'd have played around a little more to replicate the bandwidth and latency, especially considering mine is still better than that laptop, but this'll do.

One thing to keep in mind - this is an all-core OC of 4.2GHz
We know how Ryzen boosts though, so while mine here is a stable 4.2GHz, the same might not be said of the sample.

In any case, it's a little closer to the Renoir sample now. A difference of 200 in terms of single core integer score.
 

naukkis

Senior member
Jun 5, 2002
298
137
116
I never understood how some of you here thought that a decrease of memory latency (no matter how significant) could offset tens of Megabytes of missing L3 cache - especially in an APU.
It's about average memory latency. That smaller L3 of Renoir has probably lower latency than that bigger L3 in Matisse, so when data fits in Renoir's L3 Renoir should be faster. When it won't fit it's about which one gives better average memory latency(=bandwidth) smaller but faster L3 with faster memory access or slower and bigger L3 with slower memory access. With Matisse AMD did not have any choice but make L3 big as memory controller is off-die and slow with random accesses.
 

tamz_msc

Platinum Member
Jan 5, 2017
2,443
2,112
106

You're welcome :p

If I was more bothered I'd have played around a little more to replicate the bandwidth and latency, especially considering mine is still better than that laptop, but this'll do.

One thing to keep in mind - this is an all-core OC of 4.2GHz
We know how Ryzen boosts though, so while mine here is a stable 4.2GHz, the same might not be said of the sample.

In any case, it's a little closer to the Renoir sample now. A difference of 200 in terms of single core integer score.
The .gb4 reports ST clocks right? In that case the sample also reports 4.2GHz.
 

uzzi38

Senior member
Oct 16, 2019
532
506
96
The .gb4 reports ST clocks right? In that case the sample also reports 4.2GHz.
I don't trust anything any benchmarking tool reports for clocks. Take a look at the .gb4 for my run, and I locked it in to a static frequency.

Somehow or another there's always some error of margin for it's reporting.

EDIT: Take for example my 8300H laptop here. For general use I locked it out of the 4GHz SC Turbo, just left it at 3.9GHz all-core and even still you can see here at times it reported up to 3950mhz: https://browser.geekbench.com/v4/cpu/14011037.gb4

Just assume it's correct +-100mhz honestly.
 
Last edited:

DrMrLordX

Lifer
Apr 27, 2000
14,607
3,585
136
Still I'd hoped that the effect would be minimal due to the integrated memory controller.
Main memory is (at best) going to be in the 66ns latency ballpark, and that's if it's tuned very carefully. L3 on Matisse has at least twice as much bandwidth as main memory and less than a third the latency. For any working set that fits well into Matisse's larger L3, reducing L3 when moving to Renoir is going to have a negative impact on performance with or without the IMC.

The only edge I see Renoir potentially having is the reduced physical distance between cores and IMC from monolithic die design.

You compare a 15w laptop chip with 3700x...

Compare with 3700u instead
3700u is Picasso (Summit Ridge derived). 4700u is Renoir (Matisse derived). So comparing 3700x vs 4700u ST is somewhat appropriate, since they share the same fundamental core design.
 
  • Like
Reactions: scannall

amd6502

Senior member
Apr 21, 2017
752
209
86
You're brave. That or you're hoping for a placard to ship with it that gets you good parking.
My current system is also lenovo invalid; $300 laptop that's decent even if longevity isn't a strongpoint. Sign me up for an upgrade to a 12 thread Zen2 APU.

So checking the reddit chatter today there's a nice Lisa Su interview where she mentions about 20 products (projects?) in the works (possibly all or most on 7nm)

I can only think of a handful possible 7nm projects besides Renoir (**=guess/speculation).

1. Zen3 8c chiplet (for server and AM4)
2.** quadcore ULP mobile oriented Zen3, small APU
3. big navi
4.** medium navi gpu/igpu chiplet
5. ** vega or navi based AM4 IO hub (also doubling as small dgpu for die salvage)
6. and 7. console projects
 
Last edited:
  • Like
Reactions: lightmanek

RetroZombie

Senior member
Nov 5, 2019
292
132
76
1. Zen3 8c chiplet (for server and AM4)
2.** quadcore ULP mobile oriented Zen3, small APU
3. big navi
4.** medium navi gpu/igpu chiplet
5. ** vega or navi based AM4 IO hub (also doubling as small dgpu for die salvage)
6. and 7. console projects
7nm/7nm+:
1. Zen2 chiplet
2. Vega20
3.4. Navi10, Navi14
5.6. XboxX, XboxXX
7.8. PS5, PS5pro
9.10. Renoir, Small Renoir
11. BigNavi
12. Zen3 chiplet

Curious about the other 8 products, unless amd count them separately like zen 2 chiplet is counted three times for Ryzen, TR and Epyc.
 
  • Like
Reactions: amd6502

amd6502

Senior member
Apr 21, 2017
752
209
86
7nm/7nm+:
1. Zen2 chiplet
2. Vega20
3.4. Navi10, Navi14
5.6. XboxX, XboxXX
7.8. PS5, PS5pro
9.10. Renoir, Small Renoir
11. BigNavi
12. Zen3 chiplet

Curious about the other 8 products, unless amd count them separately like zen 2 chiplet is counted three times for Ryzen, TR and Epyc.
Yes it would make sense the Zen2 and Zen3 chiplets would get counted as 3x because these are 2 to 3 quite different applications (needing two different IO hubs). I consider it more 2 different applications because I think TR is basically a 1P server tuned to hotter and higher frequencies.

Zen2 and Zen3 look like they will reuse IO hubs. But it would make sense these hubs at some point get upgraded to 7nm. I'd think with a modest igpu for the AM4 hub, and with maybe even a L4 for the server hubs.
 
Mar 11, 2004
19,529
1,900
126
Anyone know if they talked about the video processing block any? That should be a key feature of these and so I'm interested to see what its gonna be like. I believe their APUs generally had their leading (at the time) video processing block, but not sure if that's entirely true, but its generally been different (and not necessarily in a good way, I don't know if Handbrake supports the APU's processing block but does support the older VCE in Polaris and I believe Vega now).
 

ASK THE COMMUNITY