AMD Raven Ridge 'Zen APU' Thread

Page 48 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

LightningZ71

Golden Member
Mar 10, 2017
1,661
1,945
136
I still haven't seen a single thing from anyone that indicates to me that AMD has bothered to use a different die floor plan for Raven Ridge from the Horned Owl SoC. The Horned Owl SoC, from AMD's own official slide, has listed an 8MB L3 cache. Given how resource constrained AMD is, it would be financially irresponsible to make a completely different die for Horned Owl and Raven Ridge where they will only really differ in packaging and perhaps some micro-code revisioning. Where they might, and I mean MIGHT decide to use a different die is in the smaller, more cut down Soc family under Horned Owl. It's got 2 cores, half the Gcn cores, and half the L3 listed. With all of that disabled, is it more economical to use recovered Raven Ridge dies (assumes high failure rate) or is it more economical to make a half sized SoC die (could generate over twice as many per wafer, making running cost basis much lower while still meeting the needs of the SoC world well enough). I think that a full dress desktop Raven Ridge part will be made, it will have 8MB of L3, and probably clock as high as AMD can manage. In a tiny, well designed case, it would be a great home theater/STB part that gives enough performance for lots of casual gamers. Priced reasonably, I'd jump on one in a heartbeat for my living room.
 
Reactions: raghu78

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
I still haven't seen a single thing from anyone that indicates to me that AMD has bothered to use a different die floor plan for Raven Ridge from the Horned Owl SoC. The Horned Owl SoC, from AMD's own official slide, has listed an 8MB L3 cache. Given how resource constrained AMD is, it would be financially irresponsible to make a completely different die for Horned Owl and Raven Ridge where they will only really differ in packaging and perhaps some micro-code revisioning. Where they might, and I mean MIGHT decide to use a different die is in the smaller, more cut down Soc family under Horned Owl. It's got 2 cores, half the Gcn cores, and half the L3 listed. With all of that disabled, is it more economical to use recovered Raven Ridge dies (assumes high failure rate) or is it more economical to make a half sized SoC die (could generate over twice as many per wafer, making running cost basis much lower while still meeting the needs of the SoC world well enough). I think that a full dress desktop Raven Ridge part will be made, it will have 8MB of L3, and probably clock as high as AMD can manage. In a tiny, well designed case, it would be a great home theater/STB part that gives enough performance for lots of casual gamers. Priced reasonably, I'd jump on one in a heartbeat for my living room.

I agree with you that Raven Ridge is likely to sport 8MB L3 cache for standard notebook and desktop SKUs. I would like to correct you that Banded Kestrel is the 2C/4T CPU with 4MB L3 and 3 Vega NCUs. Great Horned Owl is 4C/8T with 8MB L3 and 11 Vega NCUs. I think Raven Ridge will shine in 35-45w notebook SKUs. I think Vega's sweet spot for clock / voltage seems to be around 1200 Mhz where its highly efficient. Moreover I think Raven Ridge will run into significant bandwidth bottlenecks at high clocks (which would also hurt power efficiency) and therefore AMD is unlikely to aim for clocks beyond 1400 Mhz even for desktop SKUs.
 

LightningZ71

Golden Member
Mar 10, 2017
1,661
1,945
136
Thank you for the correction. I knew that there was a product in that 2C/4T space, but couldn't recall the name. I wouldn't be surprised if they do the R3-APU off of that core either, though, instead of R3, it may even be an R1 and R3 will be Raven Ridge with no SMT. Banded Kestrel based R1-APUs could fit into the space where the cat cores currently live.

As for the desktop APU parts, as long as they allow them to be overclocked (unlike BR it seems) both for RAM speeds and core clocks, the amount of starvation for the VEGA units might not be so extreme. Assuming the same memory controller as the existing Ryzen chips, that supports overclocked ram speeds in the 3400-3800 range. While that's still not "a lot", it's considerably more than Bristol Ridge at 2400 with more efficient memory management and better GCN cores. Also, as for core overclocking, it should also be a different beast from BR in that regard. BR, lacking an L3, would quickly starve the CPU cores due to bandwidth starvation and heat generation combined with power consumption would limit the GPU's performance. RR, with the L3, should see benefits from core overclocking as well as GPU overclocking as core clocks would increase L3 performance while GPU overclocking may improve more compute heavy games. BR has been shown to give playable frame-rates in many modern games at 720p with modest quality settings and 1080P with minimal settings. I expect RR to give very playable 720p at high quality settings (a boon to the low end gaming notebook market) and playable 1080p with modest quality settings, which will be just fine for most home theater/STB SFF platforms. With a well thought out overclock, it could provide very good 1080p gaming in most titles (and, no, I'm not saying that you can run every game at 1080p/ultra quality). So, while I suspect that RR will still suffer from bandwidth starvation, it will not be as big of a problem as it was in BR.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,655
136
http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=index.php?threads/2519335

IIRC, previously we knew that RR would be built on improved 14nm process (which is now called 12LP). Does that mean RR will be delayed? Or will it stay using the same process with the Ryzen?
I don't think we "knew" that. It was conjecture like a lot of weird things. RR was always a 2H 2017 product. We knew 14LP was going to up for general manufacturing this year. I think a lot of people assumed that it's later release was for 14LP. But no it's about AMDs limited bandwidth for releases and RR being mobile first, you have to tack on 6 months or so for OEMs to design stuff or you basically announce a product with no hardware almost half a year early.

I see no reason why RR is a 14LP CPU when 14LPP fits it's use case better than anything and therefore I don't expect the slight time shift for 12LP to affect it at all.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
Say hello to 14XM
Or that's at least what I think the 12LP is in reality.

EDIT: Or surely GlobalFoundries simply didn't rename the tuned version of 14nm LPP to 12nm LP?
They wouldn't dare, right?
 
Last edited:
Reactions: Drazick and CatMerc

ao_ika_red

Golden Member
Aug 11, 2016
1,679
715
136
Say hello to 14XM
Or that's at least what I think the 12LP is in reality.

EDIT: Or surely GlobalFoundries simply didn't rename the tuned version of 14nm LPP to 12nm LP?
They wouldn't dare, right?

14XM targeted fin pitch is already served by their existing 14LPP. This 12LP should be comparable with Samsung's 14LPU process. But it's nothing more than custom optimised 14LPP process by GloFo.
 

dnavas

Senior member
Feb 25, 2017
355
190
116
EDIT: Or surely GlobalFoundries simply didn't rename the tuned version of 14nm LPP to 12nm LP?
They wouldn't dare, right?

I thought 14nm LPP was <<10% smaller than competitive 16nm [blather]>>, while this seems to be touted as "up to" 15% smaller? Doesn't seem like a huge change to me. Relax a few rules, harden the weak spots so you can certify for autos, let marketing "spin" whatever story they want to. Ryzen's next spin, which used to be aimed at 14+ is now aimed at 12. Completely new node? Of course not, such a thing would be crazy to embark on now, just consider the lead times necessary. This is a rename -- a way to package the improvements that can be buffed for public consumption.

Either that or 7nm is in serious trouble, and AMD is being forced to yank a backup to the fore. I suppose if you believed that Pinnacle was ever aimed at 7nm, you might buy that line, but I always thought it was fairly well established that it wasn't :shrug:
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
I thought 14nm LPP was <<10% smaller than competitive 16nm [blather]>>, while this seems to be touted as "up to" 15% smaller? Doesn't seem like a huge change to me. Relax a few rules, harden the weak spots so you can certify for autos, let marketing "spin" whatever story they want to. Ryzen's next spin, which used to be aimed at 14+ is now aimed at 12. Completely new node? Of course not, such a thing would be crazy to embark on now, just consider the lead times necessary. This is a rename -- a way to package the improvements that can be buffed for public consumption.

Either that or 7nm is in serious trouble, and AMD is being forced to yank a backup to the fore. I suppose if you believed that Pinnacle was ever aimed at 7nm, you might buy that line, but I always thought it was fairly well established that it wasn't :shrug:

https://www.globalfoundries.com/new...-technology-for-high-performance-applications
https://www.anandtech.com/show/11854/globalfoundries-adds-12lp-process-tech-amd-first-customer

GF 12LP brings transistor level enhancements and design library improvements to the 14LPP process (7.5T for 12LP vs 9T for 14LPP) . The CPP and MMP are identical to 14LPP. So the entire area shrink comes from a smaller track library. The performance improvement is derived from transistor level enhancements to improve performance by 10% at same power and complexity. GF 12LP (based on 14LPP) can be compared to TSMC 12FFC (based on 16FFC) as an optimization of a mature node.
 
Reactions: dnavas

Asterox

Golden Member
May 15, 2012
1,028
1,786
136
With 8MB L3 taking up just 16 sq mm I don't think AMD cut L3 cache by half for Raven Ridge. If they did so its not for area benefit but probably for power benefit. AMD might have weighed the perf loss and power reduction and deemed the tradeoff worthwhile.

Raven Ridge seems to be around 210 - 220 sq mm. I doubt its as big as 240 sq mm.

http://www.bitsandchips.it/52-english-news/7622-rumor-two-versions-of-raven-ridge-under-development
https://techreport.com/forums/viewtopic.php?t=119682

For much better comparison, Raven Ridge CUs vs a few cheaper AMD Polaris GPU-s and two "old-new Bristol Ridge APU-s".

A8-9600, 6 CU / 384 GPU cores

A12-9800, 8 CU / 512 GPU cores

RX 550, 8 CU / 512 GPU cores

RX 460, 14 CU / 896 GPU cores

RX 560, 16 CU / 1024 GPU cores

 

Engineer

Elite Member
Oct 9, 1999
39,234
701
126
Some pretty good prices on Ryzen chips using PayPal with eBay. See VirtualLarry's thread about the Ryzen 1200 in Hot Deals.

So darn tempting to get a 65W - 8 Core Ryzen 1700 - $254.99 right now and others. Need restraint, lol!
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
https://www.globalfoundries.com/new...-technology-for-high-performance-applications
https://www.anandtech.com/show/11854/globalfoundries-adds-12lp-process-tech-amd-first-customer

GF 12LP brings transistor level enhancements and design library improvements to the 14LPP process (7.5T for 12LP vs 9T for 14LPP) . The CPP and MMP are identical to 14LPP. So the entire area shrink comes from a smaller track library. The performance improvement is derived from transistor level enhancements to improve performance by 10% at same power and complexity. GF 12LP (based on 14LPP) can be compared to TSMC 12FFC (based on 16FFC) as an optimization of a mature node.

The critical question is: Will the 12LP allow higher Fmax or just higher frequencies at ISO power...
Because 14nm LPP cannot exceed certain limits, no matter how much the power budget is stretched.

AMD needs >= 10% Fmax improvement for Ryzen as soon as possible, otherwise they will fall too far behind.
CFL 8700K will already have < 15% higher ST frequency and > 16% higher MT frequency (in addition to the higher IPC), than AMDs fastest consumer models (R5 1600X / R7 1800X).
 

ao_ika_red

Golden Member
Aug 11, 2016
1,679
715
136
The critical question is: Will the 12LP allow higher Fmax or just higher frequencies at ISO power...
Because 14nm LPP cannot exceed certain limits, no matter how much the power budget is stretched.
(

Yeah, Vega 64 really stretches its arch beyond its limit. I remember reading an article that Vega64 consumed more than 500W when overclocked but only have minimal gain.
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,867
3,418
136
Vega is good to about 1600mhz, 1600 to 1700 Is were it hits the same kinda wall Zen does.

Hopefully zen can get a two fold boost, 1 from process and 2 from further optimisations of critical paths.
 
Reactions: lightmanek

SpaceBeer

Senior member
Apr 2, 2016
307
100
116
I suppose it will be similar as Kaveri -> Carrizo/BR (28nm) optimization. So I would expect little bit higher base (up to 3.8 GHz) and boost (up to 4.2-.4.3 GHz) clocks, but not much more than that. Which is fine in my opinion
 

CatMerc

Golden Member
Jul 16, 2016
1,114
1,153
136
TN/RL = 240.41mm² (32nm)
KV/GV = 244.29mm² (28nm)
CZ/BR = 246.39mm² (28nm)

I already said it's TN/KV/CZ ish in size
The size is rather odd for me tbh.
Sounds a bit large for 4c Zen + 11 CU's. Does the GPU include an L3 cache of its own to alleviate bandwidth issues? Would cost about 60mm^2-70mm^2 for 32MB of L3. Would be managed by the HBCC. If you've got it might as well use it
 
Last edited:

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
The critical question is: Will the 12LP allow higher Fmax or just higher frequencies at ISO power...
Because 14nm LPP cannot exceed certain limits, no matter how much the power budget is stretched.

AMD needs >= 10% Fmax improvement for Ryzen as soon as possible, otherwise they will fall too far behind.
CFL 8700K will already have < 15% higher ST frequency and > 16% higher MT frequency (in addition to the higher IPC), than AMDs fastest consumer models (R5 1600X / R7 1800X).

I am hoping the 12LP process allows AMD to hit 10-12% higher max frequencies. Zen is a brand new architecture on a relatively new process. My guess is GF 12LP is primarily designed for AMD and significant amount of learning from the first gen Zen and 14LPP ramp should allow AMD and GF to make both design and process level improvements to hit higher frequencies.

https://www.globalfoundries.com/new...-technology-for-high-performance-applications

"The new 12LP technology provides as much as a 15 percent improvement in circuit density and more than a 10 percent improvement in performance over 16/14nm FinFET solutions on the market today.

In addition to transistor-level enhancements, the 12LP platform will include new market-focused features specifically designed for automotive electronics and RF/analog applications—two of the fastest-growing segments in the industry
."

We have seen how a well optimized design can extract the most out of a process (Nvidia Pascal on 16FF+). With 12LP having transistor level enhancements and AMD further optimizing Zen for higher frequencies I think the possibility is there for >10% fmax frequency increase. Another question I have for you is AMD used 9T libraries for Zen at 14LPP. 12LP brings 15% area reduction and >10% perf increase at iso power. The area reduction comes from the move to a 7.5T library. Would AMD be able to to design at 12LP with 9T libraries and would that bring them additional performance ? Given that AMD is GF's primary customer I think there is a high possibility for a custom 9T high performance library for AMD's needs.

Vega is good to about 1600mhz, 1600 to 1700 Is were it hits the same kinda wall Zen does.

Hopefully zen can get a two fold boost, 1 from process and 2 from further optimisations of critical paths.

I think the factor 2 from further optimization of critical paths could be a significant contributor as Zen is a brand new architecture. There would have been a lot of learning in bringing up Zen from tapeout to high volume production and the volume ramp process. 10% from process and 5% from design is not entirely unrealistic for such a brand new architecture.
 
Last edited:

nathanddrews

Graphics Cards, CPU Moderator
Aug 9, 2016
965
534
136
www.youtube.com
And how performance is? 15w/35w/45w parts catch up with Intels 15w/35w/45w parts on CPU performance?
Here's an easy summary citing no actual data for unreleased products:

Intel > AMD single-threaded perf
AMD ≥ Intel multi-threaded perf
AMD > Intel graphics perf
Intel > AMD perf/watt
AMD > Intel perf/dollar
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
Here's an easy summary citing no actual data for unreleased products:

Intel > AMD single-threaded perf
AMD ≥ Intel multi-threaded perf
AMD > Intel graphics perf
Intel > AMD perf/watt
AMD > Intel perf/dollar

As you said there is no actual data to make conclusions. Still even assuming Intel has superior CPU perf the summary should be

Intel CPU perf/watt > AMD CPU perf/watt
Intel GPU perf/watt << AMD GPU perf/watt

AMD's GPU perf lead will be far greater than Intel's CPU perf lead especially for power constrained notebooks.
 

nathanddrews

Graphics Cards, CPU Moderator
Aug 9, 2016
965
534
136
www.youtube.com
As you said there is no actual data to make conclusions. Still even assuming Intel has superior CPU perf the summary should be

Intel CPU perf/watt > AMD CPU perf/watt
Intel GPU perf/watt << AMD GPU perf/watt

AMD's GPU perf lead will be far greater than Intel's CPU perf lead especially for power constrained notebooks.
You think AMD will be that efficient to beat Intel at perf/W? I'm sure that in purely multithreaded applications, that will be true, but on the whole?
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
You think AMD will be that efficient to beat Intel at perf/W? I'm sure that in purely multithreaded applications, that will be true, but on the whole?

Raven Ridge will be a fantastic performer in low power notebooks. Higher DDR4 memory speeds (compared to Bristol Ridge), improved memory bandwidth compression and hopefully better drivers for Draw Stream binning rasterizer which reduces bandwidth needs should allow Raven Ridge to really shine especially at lower GPU clocks in power constrained notebooks. At higher clocks in desktops the bandwidth bottlenecks will be more evident. I expect Raven Ridge 15w to be atleast 50% faster than the 15w Kabylake-R parts at 15w. The iris Pro EDRAM parts might be able to reduce the gap but still AMD should be ahead by a larger margin than Intel's CPU perf lead over AMD.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |