AMD Bristol/Stoney Ridge Thread

[cbn]

http://www.notebookcheck.net/Bristo...0-9600P-Against-the-Competition.168477.0.html

Ok, time to revive the thread.

I need to say something first.
The way AMD is trying to match their performance is like this:
Core i7 -- FX
Core i5 -- A12
Core i3 -- A10
Pentium U -- A9
Celeron U/Pentium N -- A6
Celeron N -- E2


The only real match from AMD is the E2. In my opinion, the way on how they are need to compare are like this:

Core i7 / Unmatched (need to say more?)
Core i5 / AMD FX
Core i3 / AMD A12
Pentium U / AMD A10 (due ST performance)
Celeron U / AMD A9 (due threads available)
Pentium N / AMD A6 (due cost)
Celeron N / AMD E2

Using the comparison from AMD, is not a good product, it falls behind even Broadwell. However using the realistic way (comparing to Pentium U), it has a good chance to be a decent competitor.

We found the dedicated Radeon R7 M440 GPU of our test laptop relatively disappointing. It rarely performed faster than the integrated Radeon R5 and, when it did, the IGP was suffering performance drops due to its limiting TDP. Instead, HP should have installed the higher-performance A10-9630P (35-watt). This would have boosted CPU and GPU performance. This statement is supported by the fact that the CrossFire setup did not produce higher frame rates in none of the games we tested.

Performance Consistency - Sustaining High CPU and GPU Loads

In games, the TDP limit is temporarily exceeded. However, after a maximum of two minutes (depending on the previous load and temperatures of the laptop), the 15-watt limit strikes back, dropping clock speeds. For example, in “Diablo III”, the CPU and GPU clock speeds start at 1800 and 550 MHz respectively. As the game runs, the speeds drop to 1100 to 1200 (CPU) and 380 to 420 MHz (GPU). The frame rates drop in parallel in our static gaming scene from 72 to 49 fps.

We want to praise the AMD IGP. Once again, the manufacturer steals the performance crown in the 15-watt segment. The only Intel IGP capable of surpassing the AMD competitor is the Iris Graphics 540, which is very expensive. We hope that many manufacturers will install Bristol Ridge without a dedicated graphics card and thus make good use of this strong advantage. Dual Graphics can only create a large lead over the competition in synthetic benchmarks. When running demanding applications, such as games, the setup does not offer much better performance.

It is just amazing to me 15W APU + dGPU with 64 bit DDR3 continues to happen when 35W APU is so much more efficient and elegant. (Particularly as HP decided to use two DDR4 SO-DIMMs with this laptop).

P.S. A side benefit of transitioning to 35W APUs (over 15W APU plus Oland dGPU) would be that the Oland dGPU silicon could be freed up for use as desktop dGPU. And with so many surplus SFF desktops out there in need of a better GPU (eg, Intel Core i5-2400) it would be a great to see this happen.

 

[dooon]

Bristol Ridge "Socket AM4"
PRO A12-9800(?) AD980BAUM44AB photo

hp OEM motherboard

 

[AtenRa]

Bristol Ridge "Socket AM4"
PRO A12-9800(?) AD980BAUM44AB photo

hp OEM motherboard

Nice thanks,

The socket looks small or what ??

 

[VirtualLarry]

Pins are closer together, so it stands to reason that possibly the socket is smaller.

 

[18pd]

Has anyone seen a 35w laptop part in the wild yet? Really fancied one after Skull Canyon ended up being stupid loud/overpriced, if they only end up in a few obscure Chinese Mini PCs it'll be a tragedy tbh.

 

[NostaSeronx]

:whiste:
Bristol Ridge and Stoney Ridge successors to be on GlobalFoundries 22FDX Platform. To be called Raven Ridge-L and Raven Ridge-E. No, they will not be using the same die orientation(4C/8CU//2C/3CU).

Raven Ridge-L; 15h xxh APU
6 Digger Cores - 128-bit DDR4-3600 @ 1.2v (AMP/XMP-4266 @ 1.35v)
9 GCN4_LP Clusters - 576 ALUs
6 -- [1+2+2+2+2]/72 TMUs/8 ROPs
MaxTDP: 65 watts
~162 mm²

Raven Ridge-E; 15h yyh APU
2 Digger Cores - 64-bit DDR4L-4266 @ 1.05v
5 GCN4_LP Clusters - 320 ALUs
2 -- [1+2+2]/40 TMUs/4 ROPs
MaxTDP: 15 Watts
~79 mm²

0.4v to 1.2v RR-L/RR-E vs 0.6v to 1.4v of Bristol/Stoney.

For Raven Ridge [Zen APU] the GPU is GCN4X_LP which means it is not using Polaris CUs.

Digger is the accumulation of all Bulldozer architectural changes. That did not happen to get into Excavator. Which are;
Bulldozer on 28UMC (28HPM - SuVolta)
Bulldozer on 28STM@GF (STM28FDSOI)
Bulldozer on 22SHP (IBM 22PDSOI)
Bulldozer on 20LPM (IBM 20nm Bulk)
Bulldozer on 14XM (IBM Bulk FinFET)
Bulldozer on 14HP (IBM SOI FinFET) -- All of these never got to tapeout, some got to RTL/GDSII though or what ever the step before sign off of the tapeout.

 

[sweetusernames]

:whiste:
Bristol Ridge and Stoney Ridge successors to be on GlobalFoundries 22FDX Platform. To be called Raven Ridge-L and Raven Ridge-E. No, they will not be using the same die orientation(4C/8CU//2C/3CU).

Raven Ridge-L; 15h xxh APU
6 Digger Cores - 128-bit DDR4-3600 @ 1.2v (AMP/XMP-4266 @ 1.35v)
9 GCN4_LP Clusters - 576 ALUs
6 -- [1+2+2+2+2]/72 TMUs/8 ROPs
MaxTDP: 65 watts
~162 mm²

Raven Ridge-E; 15h yyh APU
2 Digger Cores - 64-bit DDR4L-4266 @ 1.05v
5 GCN4_LP Clusters - 320 ALUs
2 -- [1+2+2]/40 TMUs/4 ROPs
MaxTDP: 15 Watts
~79 mm²

0.4v to 1.2v RR-L/RR-E vs 0.6v to 1.4v of Bristol/Stoney.

For Raven Ridge [Zen APU] the GPU is GCN4X_LP which means it is not using Polaris CUs.

Digger is the accumulation of all Bulldozer architectural changes. That did not happen to get into Excavator. Which are;
Bulldozer on 28UMC (28HPM - SuVolta)
Bulldozer on 28STM@GF (STM28FDSOI)
Bulldozer on 22SHP (IBM 22PDSOI)
Bulldozer on 20LPM (IBM 20nm Bulk)
Bulldozer on 14XM (IBM Bulk FinFET)
Bulldozer on 14HP (IBM SOI FinFET) -- All of these never got to tapeout, some got to RTL/GDSII though or what ever the step before sign off of the tapeout.

Digger.

Fitting name for a core/module that's gonna end up as Bulldozer's grave

edit: or will it? I kinda feel Bulldozer still has a future. We'll see.

 

[Azuma Hazuki]

:whiste:
Bristol Ridge and Stoney Ridge successors to be on GlobalFoundries 22FDX Platform. To be called Raven Ridge-L and Raven Ridge-E. No, they will not be using the same die orientation(4C/8CU//2C/3CU).

Raven Ridge-L; 15h xxh APU
6 Digger Cores - 128-bit DDR4-3600 @ 1.2v (AMP/XMP-4266 @ 1.35v)
9 GCN4_LP Clusters - 576 ALUs
6 -- [1+2+2+2+2]/72 TMUs/8 ROPs
MaxTDP: 65 watts
~162 mm²

Raven Ridge-E; 15h yyh APU
2 Digger Cores - 64-bit DDR4L-4266 @ 1.05v
5 GCN4_LP Clusters - 320 ALUs
2 -- [1+2+2]/40 TMUs/4 ROPs
MaxTDP: 15 Watts
~79 mm²

0.4v to 1.2v RR-L/RR-E vs 0.6v to 1.4v of Bristol/Stoney.

For Raven Ridge [Zen APU] the GPU is GCN4X_LP which means it is not using Polaris CUs.

Digger is the accumulation of all Bulldozer architectural changes. That did not happen to get into Excavator. Which are;
Bulldozer on 28UMC (28HPM - SuVolta)
Bulldozer on 28STM@GF (STM28FDSOI)
Bulldozer on 22SHP (IBM 22PDSOI)
Bulldozer on 20LPM (IBM 20nm Bulk)
Bulldozer on 14XM (IBM Bulk FinFET)
Bulldozer on 14HP (IBM SOI FinFET) -- All of these never got to tapeout, some got to RTL/GDSII though or what ever the step before sign off of the tapeout.

Nosta, I love you, but where in Cthulhu's unholy name do you get this stuff? And, what happened to those Harvester and Crane cores you said were coming on 22nm FD-SoI?

 

[NostaSeronx]

Nosta, I love you, but where in Cthulhu's unholy name do you get this stuff?

Entropic data streams. Hail, Cthulhu's Might.

And, what happened to those Harvester and Crane cores you said were coming on 22nm FD-SoI?

Harvester is now Tunnelborer so it is 14FDX or 10nm SOI FinFET. The name change comes with some 15h optimization breaking architecture enhancements{xxh -> xyh}. Crane is the 14XM/14LPP/14HP successor to the 20LPM Excavator-derived architecture. All names post-Excavator have been switched Excavation related machinery. Harvester/Crane might reappear around 10nm/7nm/5nm though. Crane would more or less be a shrink for Tunnelborer and Harvester a re-architecture{xxh -> xyh} of Tunnelborer.

Largely none of the above is as finished as the Digger micro-architecture. The great final shrink/mastery of 15h architecture. That includes all accepted proposed changes from 28nm Depleted Bulk, 28nm FDSOI, 22nm PDSOI, 20nm Bulk, 14nm FinFET, and 14nm SOI FinFET low-balled/indefinitely delayed/cancelled projects.

 

[Azuma Hazuki]

What is an entropic data stream in this context, pray tell?

As to the rest, I have no idea how we'd go about even beginning to verify any of this. A hell of a lot of the products you predicted earlier, say the Basilisk APU for example, just vanished into the aether; where did they go?

 

[NostaSeronx]

What is an entropic data stream in this context, pray tell?

A bunch of angry; laid off employees to employees about to get laid off. Linkedin, resume searches, very subtle leaks in pdf documentation, website cache, Czech{Dresden is right there next to the border} dark trade of ditched/engineering dies, etc. There are also interviews in Korean, Hindi, Simplified Chinese, and German that have leaked info. Like the whole Puma Phone SoC leak that was quickly shutdown and canned.

As to the rest, I have no idea how we'd go about even beginning to verify any of this. A hell of a lot of the products you predicted earlier, say the Basilisk APU for example, just vanished into the aether; where did they go?

Basilisk APU = Bristol Ridge & Stoney Ridge. The codename was for a CPU though; Viperfish(10-core)-DDR3/PD(2xh), Gecko(16-core)-DDR3/SR(4xh), Basilisk(16-core)-DDR4/XV(5xh).

 

[inf64]

Bulldozer is dead and buried design. There will be no more cores based on any form of CMT after Zen launches, please give it a rest.

 

[SAAA]

Nosta, I love you, but where in Cthulhu's unholy name do you get this stuff? And, what happened to those Harvester and Crane cores you said were coming on 22nm FD-SoI?

Entropic data streams. Hail, Cthulhu's Might.

Pls, pls have a look at this: http://merzo.net/

A bunch of angry; laid off employees to employees about to get laid off. Linkedin, resume searches, very subtle leaks in pdf documentation, website cache, Czech{Dresden is right there next to the border} dark trade of ditched/engineering dies, etc. There are also interviews in Korean, Hindi, Simplified Chinese, and German that have leaked info. Like the whole Puma Phone SoC leak that was quickly shutdown and canned.Basilisk APU = Bristol Ridge & Stoney Ridge. The codename was for a CPU though; Viperfish(10-core)-DDR3/PD(2xh), Gecko(16-core)-DDR3/SR(4xh), Basilisk(16-core)-DDR4/XV(5xh).

That's quite... the digging work man. Are you 100% sure that some kind of 22nm is in the air? I'd love to see the module concept ported to a much more efficient and smaller node, merely out of scientific curiosity at this point.
Btw Zen should easily have higher single core performance but if they canned Puma succesors the Bulldozer line could turn to a lower price segment if shrunk to 14nm.

 

[NostaSeronx]

- Highest pin bandwidth available
* 4.8/6.4 GHz Hexadecimal Data Rate (16X Data Rate) signaling (DQ)

• Bi-directional Differential RSL (DRSL)
• 200mV voltage swing
• Sixteen data bits transmitted per clock

* 300/400 MHz differential reference clock (CFM)
* 600/800 MHz dual-edge address/command signaling (RQ)

• Uni-directional single-ended RSL
• 900mV voltage swing

- Flexible interfaces
* Internal

• CMOS signaling for internal TDATA/RDATA
• Synchronous to memory controller PCLK
• Data serialization ratio of 16-to-1

* External

• Support for variable width data interfaces
• Available in either wire-bonded or bumped flip-chip package versions

- Low power

• 1.0 Vdd; 1.2V VddIO
• Ultra-low voltage external signaling levels

- Performance features
* FlexPhase circuits for precise per-pin data phase alignment

• Transparent interface levelization (de-skewing)
• Elimination of system trace length matching

* Programmable on-chip termination with external resistor reference
* Current calibration to maintain constant signaling levels

New Xtreme Data Rate 1 from Rambus.

Support was apparently secretly added in Bristol Ridge, F15M65h.
AM4 = New XDR [300 MHz - 450 MHz]
FP5 = New XDR Mobile [200 MHz - 267 MHz] {This a direct cut from new XDR, it isn't really new XDR mobile // 260-pin XDR rather than desktop 288-pin XDR}

It was shown here in its pre-alpha state;
http://semiaccurate.com/2012/09/17/rambus-shows-a-best-of-both-worlds-memory-tech/

New XDR(16X Data Rate)/New XDR2(32X Data Rate) is twice as fast as XDR(8X Data Rate)/XDR2(16X Data Rate). With XDR fitting in DIMMs and other DDR3/DDR4 like objects, XDR2 aiming for Graphics (GDDR5/GDDR5X). If anyone is following me, XDR2 is in Polaris.

https://www.rambus.com/amd-renews-patent-license-agreement-with-rambus/ Up to 2015.
https://www.rambus.com/amd-extends-patent-license-agreement-with-rambus/ Up to 2020. [Specific terms of the agreement are confidential.]:whiste:

How to know the difference between new and old;
Data Rate is doubled.
Vdds are 1.0v to 1.2v. Rather than 1.8v Vdd and 1.5v Vdd.
2015 dumps in GDSII on 28SLP/14LPP. PDK for 22FDX Platform isn't complete but prototype for XDR PHY has appeared.

XDR => GDDR5M [4.8 to 8 Gbps] // XDR2 => GDDR5X [9.6 to 20 Gbps]
New XDR; https://www.rambus.com/emerging-solutions/beyond-ddr4/ <== Hinted here.
New XDR2; https://www.rambus.com/32x-data-rate/ <=== Hinted here.

 

[DrMrLordX]

Ugh, RAMBUS. Please no.

 

[hojnikb]

Any news on when bristol ridge is going to launch on desktop ?

 

[NostaSeronx]

Any news on when bristol ridge is going to launch on desktop ?

September for OEMs, October for DIYs.

---
I've been noticing some weird things about Bristol Ridge and Stoney Ridge. Like they both have full implementations of VCE/UVD/DCE from P10/P11 and a modified version of GFX from Carrizo. There is also hints of Integrated Voltage Regulation on both parts. AMD does not do it the same way as Intel, but rather of an independent IVR per CPU, NB, and GPU rails. Each rail is optimized per device and this allows for a near 95% efficiency rather than 8x% in Carrizo/Carrizo-L.

Supposed support of F15hM65h and F15hM70h;

UVD6.x;
4K H.264 10-bit / 60 fps
4K HEVC 10-bit / 60 fps
4K VP9 Profile 2&3 / 60 fps & 30 fps
4K MJPEG / 30 fps

VCE3.x;
4K 2-pass H.264/HEVC 10-bit / 60 fps

DCE11.x;
eDP 1.4b-HDR/DP 1.4-HDR/HDMI 2.0b-HDR support

GFX8_x;
3rd Generation GCN Revision 1 Graphics (Radeon 445M -> Radeon 405M w/ Radeon 455DX -> Radeon 435DX - Dual Graphics w/ GCN1r1 (Iceland))

Bristol Ridge is laid out the same as Carrizo, but is not using the Carrizo die as implied by AMD slides. The die is apparently between 280 and 300 mm² and retains rectangular appearance. Following above and a massive update to IO/uncore/memory controller.
12x PCI Express 3.1
2x memory buffer for increased perf.
etc.

Misc findings;
Kaveri = GF28SHP (Super High Performance) ((HP @ TSMC equiv))
Carrizo = GF28LPA (Low Power Advanced) ((HPC @ TSMC equiv))
Bristol Ridge/Stoney Ridge = GF28HPA (High Performance Advanced) ((HPC+ @ TSMC equiv))

 

[Insert_Nickname]

I've been noticing some weird things about Bristol Ridge and Stoney Ridge. Like they both have full implementations of VCE/UVD/DCE from P10/P11 and a modified version of GFX from Carrizo.

Its not that weird really. AMD did something similar with Trinity/Richland. Their IGP is based on TerraScale3 (VLIW4), but they have a UVD block identical to the first-gen CGN discrete GPUs.

I suspect the UVD block is pretty much port-and-play (yup. both puns intended) between GPU generations.

 

[Azuma Hazuki]

Fascinating stuff, Nosta...given it's on an HP rather than an LP process, will we see an unexpected performance boost over Carizzo/Kaveri?

I ask because the A8-7600 has been my go-to for almost 2 years for basic builds for my customers, with around 90% the power of a Haswell i3 for 65% the price, and am hoping the Bristolridge equivalent is similar in relation to the 6100/7100.

 

[NostaSeronx]

Fascinating stuff, Nosta...given it's on an HP rather than an LP process, will we see an unexpected performance boost over Carizzo/Kaveri?

It is more a TDP shrink than a performance boost. Where in the areas of the same TDP, yes there will be a boost.

I ask because the A8-7600 has been my go-to for almost 2 years for basic builds for my customers, with around 90% the power of a Haswell i3 for 65% the price, and am hoping the Bristolridge equivalent is similar in relation to the 6100/7100.

A8-9600 will have lower power consumption and a higher GPU clock. VCE/UVD/DCE of Polaris10/11 etc. It would also be on the AM4 socket.

----
http://dl.acm.org/citation.cfm?id=2934586
^--I found the FIVR thing!!!!

http://i.imgur.com/7IauksM.png
^-- FDSOI next year might lead to Bristol/Stoney shrink with even lower power consumption. Maybe even include GFX_IPv9 (FP5 support with DDR4-3600 single channel/DDR4-3200 dual channel?, etc)

 

[Azuma Hazuki]

Good to know

I've always been interested in maximizing bang for the buck, and while Intel rules the high end, AMD's been excellent at the low to mid-range. Given the choice between an A8 and a Pentium Gxxxx, I'd always choose the A8.

Honestly I think the i3 is overpriced for what it is. Intel ought to either stop making Celeron and Pentium CPUs on the Core uarch and drop the i3's price, or rearrange them so that the i3 is 4c/4t, i5 is 4c/8t, and i7 is 6c/12t.

 

[NostaSeronx]

So, I've been digging further. Disclaimer; This is classified as speculation, no indication that it is actually in product.

Basically what I have found is a discrete toggle for Mode 0 and Mode 1 in Bristol Ridge's and Stoney Ridge's Excavator core. This can only be toggled on by a supported operating system and new firmware. (Hint: Windows 10 Redstone, etc.)

The differences between operator mode 0 and mode 1 is specific to how the Scheduler/PRF/Execution units in Excavator(BriRidge/StoRidge) work.

Mode 0 (Legacy) => Scheduler sees 2 ALUs / 2 Logic / 2 AGUs. PRFs are replicated and does not support register renaming. Execution in any new pipes are virtually fused off. (112(r0-r111) available registers with two 112-entry register files)
Mode 1 (Plus) => Scheduler sees 4 ALUs / 2 AGUs. PRFs are banked and supports the use of register renaming. Execution in any new pipes are on. (224 (r0_b0-r111_b1 / r0-r223) available registers with two 112-entry register files.)

Any fixes in Bristol Ridge Mode 0 apply to Carrizo as well. While, Mode 1 is what differentiates Carrizo and Bristol Ridge. This also applies to UVD/VCE/GFX when in fact the units in Bristol Ridge are nothing like Carrizo. This technique has been used before specifically in Oland-derived GPUs;
Oland Mark 1 => Southern Islands with VCE/UVD/DCE enabled
Oland Mark 2 => Sea Islands with no VCE/UVD/DCE
Oland Mark 3 => Volcanic Islands with no VCE/UVD/DCE
Oland Mark 4 => Arctic Islands with no VCE/UVD/DCE (Apparently, this GPU hints at GMI over PCIe interface. This only turns on when Mode 3? is enabled for it)
Greenland with 4 GMI ports = 100 GB/s to 4096 ALUs shrinking that down to 384/320 ALUs outputs 9.375 GB/s and 7.8125 GB/s. PCIe 3.0 with x8 = 8 Gbit/s x 8 ÷ 8 => 8 GB/s * 128/130 = 7.87692308 GB/s
Going by memory speed = 100 GB/s to 512 GB/s shrinking that down to 73.6 GB/s to 16 GB/s outputs 14.375 GB/s and 3.125 GB/s.

====
Redstone breaks performance... uck // Ultimate sacrifice;

The Voltage and Frequency can fluctuate within modules. It appears to only have a 400 MHz/100 mV switch though. Cinebench was 2.9 GHz average @ single core, 2.8 GHz average @ quad-core.

VCE / UVD and the GPU is broken in Microsoft's Redstone Update. Only 2 CUs are enabled, the UVD is supposedly on but 1.3K60 to 4K60 are run on the CPU. Dropped frames galore...

P-State FID 0x16 - VID 0x10 - IDD 15 (19.00x - 1.450 V) // 3.8 GHz @ 1.45v
P-State FID 0x10 - VID 0x2E - IDD 11 (16.00x - 1.262 V) // 3.2 GHz @ 1.262v
P-State FID 0xD - VID 0x3E - IDD 9 (14.50x - 1.162 V) // 2.9 GHz @ 1.162v
P-State FID 0xB - VID 0x46 - IDD 8 (13.50x - 1.112 V) // 2.7 GHz @ 1.112v
P-State FID 0x7 - VID 0x58 - IDD 6 (11.50x - 1.000 V) // 2.3 GHz @ 1v
P-State FID 0x114 - VID 0x66 - IDD 4 (9.00x - 0.912 V) // 1.8 GHz @ 0.912v
P-State FID 0x10C - VID 0x6E - IDD 3 (7.00x - 0.862 V) // 1.4 GHz @ 0.862v

 

[ET]

September for OEMs, October for DIYs.

Thanks for all the posts. October is later than I had hoped, but still, could be worse. I hope that we do get it in October, for a good price, and coupled with inexpensive microATX motherboards.

 

[dooon]

AMD PRO A-Series APU "Desktop Bristol Ridge" vs "Desktop Skylake" Spec

 

[cbn]

AMD PRO A-Series APU "Desktop Bristol Ridge" vs "Desktop Skylake" Spec

Glad to see the dual core (A6-9500) get a bump up to 6 CUs (8 compute cores total). That is a first!! I am very happy AMD finally did this! Thank you AMD! (re: Normally the dual core (eg, A6-7400K/A6-7470K) gets 4CUs on the iGPU (6 compute cores total)).