AMD Raven Ridge 'Zen APU' Thread

Page 58 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
CatMerc

CatMerc

Golden Member
Jul 16, 2016
1,114
1,153
136
Abwx said:
Surely, but if say 25W were necessary for 4C/8T at 2GHz then a chip like the R7 1700 would be at least 125W.

Think about it, 8C/16T@3GHz fit within 65W, and half this core amount would require 25W@2GHz.?..
Click to expand...
One has a GPU to feed, the other does not.
 
  • Like
Reactions: beginner99
Jan Olšan

Jan Olšan

Senior member
Jan 12, 2017
572
1,129
136
PeterScott said:
G4560s are not low power laptop parts. I thought the Argument for BK (Banded Kestrel) was Ultra Low power tablet part?
Click to expand...
I dunno, maybe? But the more important target are likely cheap desktops and notebooks, though. You don't want to just leave that market to the competition, as I said before. That's the primary reason for the small APU, possible usage as low-power tablet part or something is going to be secondary.
You don't usually punt cheap lowend designs for "ultramobile" usually, since those are premium devices. If this chip was meant for this segment primarily, it would have more powerful iGPU and dual-channel memory, not 3 CUs + 1channel like Stoney Ridge. Actually, everything about it screams "cats"/Stoney Ridge replacement, so it should be marketed as Stoney Ridge Replacement? It would be good if it had AM4 version though, or at least sufficient supply of BGA-based desktop motehrboards if not.

PeterScott said:
Using one tapeout with disabled cores, fits with AMD strategy for over a decade. It will continue for Raven Ridge.
Click to expand...
Their strategy since 2011 till now has been to have a small/cheap separate chip for lowend - both lowend notebooks and desktop/SFF.
Zacate, Kabini, Beema, Stoney Ridge now. Dualcore Zen on as small die as possible makes sense for CPUs/APUs intended to sell for 40-60 $ in retail. Not saying they didn't cancel it for whatever reason but it would be better if they still planned it.
 
Last edited:
DrMrLordX

DrMrLordX

Lifer
Apr 27, 2000
22,901
12,967
136
CatMerc said:
Until we see both A11 running on the same OS and software stack as other SoC's, we can't really tell performance in apples to apples comparison. The OS makes a huge difference in tests.
A large part of Apple's SoC's strength is that they're running a top to bottom show, meaning their software stack is incredibly optimized for their hardware, and their hardware is incredibly optimized for their software stack.
Click to expand...

Anyone run SPEC on the A11x yet? That would be a good place to start.
 
FIVR

FIVR

Diamond Member
Jun 1, 2016
3,753
911
106
Saw this on reddit, not sure it's been posted here yet. Looks like we can expect Ryzen mobile sometime in November or December.
srv6b6fmgdsz.png
 
  • Like
Reactions: wilds, DeeJayBump, Madpacket and 2 others
DrMrLordX

DrMrLordX

Lifer
Apr 27, 2000
22,901
12,967
136
Bondrewd said:
Running SPEC on iOS is about as valid as using ICC for benchmarking.
Click to expand...

Using ICC for benchmarking is valid, since it tends to produce code that is faster on all x86 CPUs vs other "neutral" compilers. Though honestly I haven't seen a comparison of ICC vs GCC vs LLVM on Ryzen yet.

PeterScott said:
And Apple won't be selling them to 3rd parties, so they won't be a competitor anyway.
Click to expand...

Heh. Okay.
 
V

.vodka

Golden Member
Dec 5, 2014
1,203
1,538
136
AGESA 1.0.0.7 is around the corner.

https://www.reddit.com/r/Amd/comments/76jzl4/elmorrog_rep_on_agesa_1007/

elmor @ C6H thread on OCN said:
AGESA 1007 comes with support for Raven Ridge APUs. AMD has also changed the entire BIOS base structure so we have to do a lot of work to port everything to the new version, which may result in further bugs. The advantage is that it makes it easier to support future CPUs (Raven Ridge, Pinnacle Ridge).
Click to expand...

looncraz said:
I was told last month that AGESA 1.0.0.7 was a complete rewrite of the memory management and a major revamp of the microcode that was going to bring Ryzen's performance up quite a bit if motherboard companies implement it. I didn't believe it until this news as AMD usually does things through evolutionary steps and the version increment is so small, but my friend just relays what he hears at work... so some is right on, but most is somewhat distorted.

No idea what the performance jump is worth, but it didn't sound like anyone expects it to be inconsequential - but expectations mean nothing.
Click to expand...

looncraz said:
That's why I didn't bother saying anything until another party confirmed the base rewrite
Click to expand...

looncraz said:
They rewrote / redesigned the memory related code for sure and supposedly the updated microcode improves on some performance issues. That's all I know that I will say in public.

5% performance uplift would be quite big in this context, I don't think anyone should expect more than that - and certainly not as a general improvement.

The memory changes are certainly compatibility centric, but some latency reduction and performance improvement would be par for the course.

AGESA 1.0.0.4 and 1.0.0.6 both gained performance and improved memory handling. This is just a bigger jump than that.
Click to expand...


While they're at it, AM4 should get some more polishing in a month or two.


Yeah, that's our very own @looncraz, I think :D Thanks for the heads up!
 
  • Like
Reactions: wilds, DeeJayBump, Madpacket and 2 others
moinmoin

moinmoin

Diamond Member
Jun 1, 2017
5,242
8,456
136
DrMrLordX said:
Interesting.

1.0.0.6 got me from DDR4-3200 with ehhh subtimings to DDR4-3466 with tight subtimings. 14-14-14-28 in both cases. So I really liked that update.

Something bigger than that would be um. Great?
Click to expand...
Something bigger would be better auto-tuning of memory settings. While the latest AGESA versions are great for allowing tight subtimings at higher frequencies with manual tweaking I'm pretty sure the amount of users getting to that point is minuscule.
 
DrMrLordX

DrMrLordX

Lifer
Apr 27, 2000
22,901
12,967
136
auto-tuning is pretty limited, though. In fact I don't think that's ever been a "thing". The problem is that, for whatever reason, Ryzen's IMC likes wildly different timings/subtimings at any given frequency based on even miniscule changes in the types of IC in use.

So Samsung B, Samsung E, Hynix, and Micron ICs (not to even go in the subtypes of Hynix and Micron DDR4 ICs) could all need different sets of subtimings at say DDR4-3200, and that's assuming that anyone's even selling Micron DIMMs at that speed right now. Then take binning into account, and you may have subtiming variance +/- x% at a given standard vDIMM.

If the DIMM manufacturers took the time to test and tune their sticks before selling them, then AMP profiles could solve all this for us. As it is we have FlareX and that's about it? XMP subtimings are a joke on Ryzen.
 
moinmoin

moinmoin

Diamond Member
Jun 1, 2017
5,242
8,456
136
Auto-tuning is very limited atm indeed (or were you talking about 1.0.0.7?) and that plus as you said the ridiculously bad coverage so far outside FlareX is exactly why it needs to be improved. Respectively an improvement there should have a positive performance effect for all users that didn't successfully manually tweak memory settings already. Maybe @looncraz knows 1.0.0.7 contains such improvements?
 
DrMrLordX

DrMrLordX

Lifer
Apr 27, 2000
22,901
12,967
136
It would be nice if 1.0.0.7 has the proper settings to end-run XMP subtimings and substitute timings suitable for AMD's IMC given the IC type. I just don't know that AMD is going to put that kind of focus into the product. I could be wrong.
 
J

jpiniero

Lifer
Oct 1, 2010
16,800
7,249
136
ianmills said:
That's great news if RR gets full android support. Since Intel stopped supporting its a great opportunity for amd
Click to expand...

As mentioned it's probably Chromebooks/ChromeOS. I don't know the extent of ChromeOS's Android support but obviously it's good enough to run Geekbench.
 
NewFatMike

NewFatMike

Junior Member
Jun 24, 2017
22
1
11
jpiniero said:
As mentioned it's probably Chromebooks/ChromeOS. I don't know the extent of ChromeOS's Android support but obviously it's good enough to run Geekbench.
Click to expand...
Starting in the next few weeks, it should be all Chromebooks will have full Google Play Store app support. It's been in beta for a long time, it'll be really nice once it's fully up and running.

Sent from my Pixel using Tapatalk
 
You must log in or register to reply here.

TRENDING THREADS

COMPANY
RESOURCES
FOLLOW
Top Bottom