AMD Bristol/Stoney Ridge Thread

[hojnikb]

It makes no sense to develop on bulldozer arch.

It would sorta make sense to make a 2C4T bottom barrel part using zen cores and older 28nm fab (to lower costs). Doing anything with bulldozer for anything is just plain idiotic.
Only thing it make sense is to continue selling embedded parts for industrial use.

 

[DrMrLordX]

AMD is on the hook to take 28nm wafers or pay GF, so they will do something with those so long as GF continues to force them on AMD. What they will do with them is anyone's guess.

 

[NostaSeronx]

It makes no sense to develop on bulldozer arch.

It makes ton of sense to develop further into Bulldozer's architecture. There is a lot of caveats unexplored by AMD that retain execution compatibility to 15h. One special caveat allows both cores to execute as if logically as a single core. It also is included to the CXU/Cluster Execution Unit patent.

AMD is on the hook to take 28nm wafers or pay GF

GlobalFoundries has moved 28nm to "mainstream," unofficially. Latest consumer products at AMD tend to stay with the "leading edge."

Kaveri 28SHP => Carrizo 28A(High Density SHP) => Bristol/Stoney 28HPA(High Density SHP+), part of the whole continuous transistor enhancement thing.

---
A FX-9590 port could be done with emphasis on the 4x power decrease(2x 50% power drops, 1- from 22FDX and 2- from body biasing.) So, basically ~240 mm² ~ 8 core die @ ~5 GHz with a TDP of 65Ws. For $120 and less and is completely feasible as FDSOI is cheaper than FinFET at the end of the day.

 

[DrMrLordX]

Yeah that would have been great last year.

 

[NostaSeronx]

Seronx projective speculation(non-final, subject to change at any time as more info arrives):

- <30% IPC(instructions per cycle) increase, >30% EPI(energy per instruction) increase. (Dual IPC & Frequency boost.)
- Bulk to SOI, Ultra-wide frequency and voltage range(UWFVR), Panoptic adaptive voltage bias scaling(PAVBS), and PAVFS(panoptic adaptive voltage frequency scaling).
- Fusion of Bristol/Stoney SKUs @ sub-180 mm² on 22FDX
- Two modules and four NCUs @ sub-10 watt w/ Micro-FX/A APUs and 15W+ w/ normal FX/A APUs.
- Some CPU architectural speculation...
* 128 KB L1i // 2x 64 KB L1d // 512 KB L2 // 4 MB L3 (Shared between modules)

Largest assumption:
If it hasn't appeared in Zen, it will appear in next gen CMT/Bulldozer/15h derived architectures.

Closest previous AMD likeness;
AMD T-bird to Thorton.

 

[ET]

Thanks for all the interesting info, NostaSeronx.

Sadly I've given up on Bristol Ridge, and plan to get a Pentium G4560 + low profile RX 460 (underclocked and undervolted) for my HTPC. Bristol Ridge was a nice dream, but AMD even now isn't willing to commit to a release date, and also motherboards unfortunately don't support HDMI 2.0.

 

[NostaSeronx]

AMD even now isn't willing to commit to a release date, and also motherboards unfortunately don't support HDMI 2.0.

Release date for retail is soon... Ryzen 5 to Ryzen 3, very quiet.
- Lenevo M715q w/ Bristol Ridge launched Feb, 2017. (other models launched as early as October 2016.)

The IP used in Bristol Ridge/Stoney Ridge supports DP1.2a/HDMI 1.4b at baseline, while supporting DP1.4 & HDMI2.0b via patching. These require the drivers and firmware to be up to date to use. So, three DP1.2-1.4 or three HDMI 1.4b-2.0b: yep, hybrid controllers. Technically, HDMI 2.1 is supported as well, but without 48 Gbps mode...

 

[dark zero]

Thanks for all the interesting info, NostaSeronx.

Sadly I've given up on Bristol Ridge, and plan to get a Pentium G4560 + low profile RX 460 (underclocked and undervolted) for my HTPC. Bristol Ridge was a nice dream, but AMD even now isn't willing to commit to a release date, and also motherboards unfortunately don't support HDMI 2.0.

That seems that is comming with Raven Ridge. Definately it will support HDMI 2.0.

 

[NostaSeronx]

@dark zero from http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=thread...and-discussion.2499879/page-182#post-38795595

but... Why Stoney Ridge is still alive at 2018????
@NostaSeronx is right?
And the refresh will be at 22nm?

Stoney Ridge is sticking to the FDX 28nm platform: https://www.globalfoundries.com/technology-solutions/cmos/fdx
Stoney Ridge will be on FP5 and FT4.

Bristol Ridge is being replaced by another product that uses a new from scratch module&core architecture. Which in all calculations... imply, a fused Cat and Construction architecture to be used. It should appear on 22FDX and is mostly targeting EPI: <25% increase in work/IPC & >25% decrease in energy/EPI, etc. (Below Zen, but significantly to targeting the left: http://images.anandtech.com/doci/10591/HC28.AMD.Mike Clark.final-page-004.jpg)

Should look similar to Samsung's Mongoose architecture, but in a module approach and higher clocks. Execution compatibility of GP/FP with bdver1-bdver4 & btver1-btver2 & support of znverx FP compatibility.

 

[blackarchon]

I have a general question about the excavator cores: why does using faster RAM have nearly no effect on the performance in CPU-limited tasks? What is holding them back?

 

[amd6502]

I have a general question about the excavator cores: why does using faster RAM have nearly no effect on the performance in CPU-limited tasks? What is holding them back?

When you say memory speed, there are actually two very different ways that memory can be fast:

1. high frequency / high bandwidth

2. low latency time (CAS/frequency)

The first hardly ever significantly affects a computer's performance. (Possible exceptions purely memory oriented operations, such as copy a block of memory to another block). This type for speed (bandwidth) is what GPUs need, and CPUs could care less about.

The second actually greatly affects typical CPU performance. Exception is when the program or loops are so small that the CPU's cache is almost 100% effective.

 

[amd6502]

Bristol Ridge is being replaced by another product that uses a new from scratch module&core architecture. Which in all calculations... imply, a fused Cat and Construction architecture to be used.

What would be spectacular is if you could share the front end (including L1i) between a cluster of 2 puma and 2 excavator cores/threads. Take Zen's front end (https://upload.wikimedia.org/wikipe...cture.svg/576px-Zen_microarchitecture.svg.png), modify/double it so that it can handle 4 threads and dispatch 12 ops. Allow internal renaming of the 2+2 cores so that the dispatcher can rearrange the the workloads for optimal efficiency:

1) avoid internal tasksetting such that two float heavy threads end up on the dozer module.
2) highly branching or cache-missing threads get higher affinity for the pumas, which clock at 2/3 or 1/2 of the dozer's frequency. Idle/no-op/dumb loop threads get highest affinity to puma threads.

 

[DrMrLordX]

I have a general question about the excavator cores: why does using faster RAM have nearly no effect on the performance in CPU-limited tasks? What is holding them back?

Depends on the task. Doing something with a large memory footprint, such as SuperPi 32m or y-cruncher in the 1 or 2.5 gig range will chew up some memory, making system memory performance more important than some other benchmarks.

 

[ET]

I wonder if AMD will ever release Bristol Ridge for desktop DIY. I'm happy I caved in and bought an Intel G4560. Kind of sorry I bought an RX 460 with it (mainly due to MSI's inability to produce a matching low profile bracket). The GeForce GT 1030 seems like a better fit for my needs, but that would have been a bit too much leaving AMD behind on both CPU and GPU fronts, so I'm glad it wasn't available when I upgraded. Ended up with an imperfect solution, but at least I supported AMD. I wish AMD released stuff that fitted my needs.

 

[NostaSeronx]

I wonder if AMD will ever release Bristol Ridge for desktop DIY.

Not Bristol Ridge should be coming out sometime upcoming Q4-Q1. On the 22FDX*cough*16nm*cough* from GlobalFoundries*cough*TSMC*cough*.

...28nm planar, 20nm planar and 16nm FinFet <-- 2015(Excavator)
...28nm Bulk, 22nm FDSOI (22FDX) <-- post-2015(Excavator) (pre-12FDX roadmap)

===
March 2017 Refresh if I didn't post it here:
A12->E2 => xx20 = 2017 models of xx00 models of Bristol Ridge & Stoney Ridge w/ A9-9430 being the 2017 model of A9-9410. (Same TDP: Higher Performance)

Apparently, there will be a refresh of the FX and Desktop versions eventually. Which might get undercut because of the above shrink.

 

[formulav8]

I wonder if AMD will ever release Bristol Ridge for desktop DIY.

According to Anandtech's interview with Lisa Su in March she said its coming. Mobo makers for AM4 has also been adding BR support to their boards. AMD needs an AM4 APU before the Zen apu comes out next year.

 

[ET]

According to Anandtech's interview with Lisa Su in March she said its coming.

Yes, that was when I decided to give up, because it was obvious it would take some time until it's released. I think that Bristol Ridge was simply underwhelming for the desktop. It was a good chip for low power laptops, but not better in performance than Godavari for the desktop (and not that great on power draw either).

 

[formulav8]

but not better in performance than Godavari for the desktop (and not that great on power draw either).

Actually Carrizo wasn't to bad. I think it was looncraz that did an Athlon x4 to steamroller comparison and it did quite well even though the low clockspeeds hurt it. Half the L2 cache hurt it's IPC in some area's but Carrizo for the most part still did well. That was while it was on the low power 28nm process. BR is on a higher performance, higher clock-able 28nm process and should handily beat SR having higher clockspeeds than Carrizo. A Chinese overclocker IIRC clocked BR well over 4ghz.

AMD still needs an APU until the Zen APU comes out IMO.

 

[ET]

IIRC from the reviews I've seen, BR is not bad as a CPU, but loses in games by quite a margin to the A10-7870K. So it's a downgrade compared to what is available on FM2+, and therefore not really good for much except plugging the low end AM4. Presumably AMD is fine with that hole being filled by FM2+ and AM3 until it has something better than BR for AM4.

I did think like you that AMD needs something at the low end, but that was mainly as a stop-gap before Ryzen release, to get AMD fans to adopt AM4. At this point I think that it's better for AMD to keep AM4 an enthusiast platform, and BR will hurt that image. I know that Lisa Su said that BR will be made available, but at this point I don't see this happening, and don't see the point of it happening. The closer we get to Ryzen 3 and Raven Ridge release, the less point there is in releasing BR.

 

[Insert_Nickname]

Actually Carrizo wasn't to bad. I think it was looncraz that did an Athlon x4 to steamroller comparison and it did quite well even though the low clockspeeds hurt it. Half the L2 cache hurt it's IPC in some area's but Carrizo for the most part still did well. That was while it was on the low power 28nm process. BR is on a higher performance, higher clock-able 28nm process and should handily beat SR having higher clockspeeds than Carrizo. A Chinese overclocker IIRC clocked BR well over 4ghz.

Carrizo isn't half bad for what it is, and it is ********* efficient at low frequency. Even on desktop.

AMD still needs an APU until the Zen APU comes out IMO.

I disagree. Kaveri/Godavari performs very similarly, even if they're on an older dead platform*. Excavator isn't a gaming CPU either, Steamroller beat it in many games at similar frequency.

*Recent FM2+ boards, like f.x. the Asrock A88M-G/3.1, have all the bells and whistles needed.

IIRC from the reviews I've seen, BR is not bad as a CPU, but loses in games by quite a margin to the A10-7870K. So it's a downgrade compared to what is available on FM2+, and therefore not really good for much except plugging the low end AM4. Presumably AMD is fine with that hole being filled by FM2+ and AM3 until it has something better than BR for AM4.

I did think like you that AMD needs something at the low end, but that was mainly as a stop-gap before Ryzen release, to get AMD fans to adopt AM4. At this point I think that it's better for AMD to keep AM4 an enthusiast platform, and BR will hurt that image. I know that Lisa Su said that BR will be made available, but at this point I don't see this happening, and don't see the point of it happening. The closer we get to Ryzen 3 and Raven Ridge release, the less point there is in releasing BR.

As I have stated a few times before, I don't think AMD should drag the remains of the BD fiasco over on their new platform. It could damage perception before RR. So I pretty much agree with you.

 

[formulav8]

Excavator isn't a gaming CPU either

AMD has no entry level AM4 cpu with an IGP. You're required to get a discreet card even when the more-so entry level R3 is released. BR would fill the actual entry nitche well. But AMD doesn't see it as to important or it would have been here by now.

But its apparently coming at some point or planned. Just the last month or so my board has added BR CPU's to the supported list.

A10-9700 (AD9700AGM44AB, 3.5GHz, 4C, L2:2M, 65W,rev.A1,AM4)
A10-9700E (AD9700AHM44AB, 3.0GHz, 4C, L2:2M, 35W,rev.A1,AM4)
A12-9800 (AD9800AUM44AB, 3.8GHz, 4C, L2:2M, 65W,rev.A1,AM4)
A12-9800E (AD9800AHM44AB, 3.1GHz, 4C, L2:2M, 35W,rev.A1,AM4)
A6-9500 (AD9500AGM23AB, 3.5GHz, 2C, L2:1M, 65W,rev.A1,AM4)
A6-9500E (AD9500AHM23AB, 3.0GHz, 2C, L2:1M 35W,rev.A1,AM4)
A8-9600 (AD9600AGM44AB, 3.1GHz, 4C, L2:2M, 65W,rev.A1,AM4)

 

[Insert_Nickname]

AMD has no entry level AM4 cpu with an IGP. You're required to get a discreet card even when the more-so entry level R3 is released. BR would fill the actual entry nitche well. But AMD doesn't see it as to important or it would have been here by now.

I don't view that as a particular disadvantage. A simple GT710/R5 230 is sufficient for basic video output, if you don't care about gaming.

But its apparently coming at some point or planned. Just the last month or so my board has added BR CPU's to the supported list.

More like they're finally getting "official" support. My own Crosshair VI has had support for BR from day 1, and I doubt any AM4 boards are different.

 

[DrMrLordX]

I don't view that as a particular disadvantage. A simple GT710/R5 230 is sufficient for basic video output, if you don't care about gaming.

While I agree with this in principle, I saw something quite silly in a NewEgg review that opened my eyes on the matter:

https://www.newegg.com/Product/Product.aspx?Item=N82E16813144016

This review is from: MSI X370 XPOWER GAMING TITANIUM AM4 AMD X370 SATA 6Gb/s USB 3.1 HDMI ATX Motherboards - AMD

Pros: Great MB. It had issues with my RAM, and I was able to fix this by resetting the CMOS.

Cons: Integrated graphic card doesn't work. I have 1800x. It worked for about couple of days. I primarily used GTX 1080 TI and had to change my Linux configuration for passthrough. When I moved my HDMI (tried with display port as well) from 1080 TI to onboard, it didn't work. I even removed GTX1080 Ti and it still didn't work. Even in the BIOS, there are no option for Integrated Graphics. I'm using firmware v1.5. Integrated Graphic is no longer working

Other Thoughts: I really need to be able to use both 1080 ti and onboard/integrated graphic for my work. if this doesn't work, then I'll probably buy a different brand MB.

Seriously? People are so used to iGPUs on Intel rigs that they don't even stop to think that some systems have no iGPU!

 

[amd6502]

I don't view that as a particular disadvantage. A simple GT710/R5 230 is sufficient for basic video output, if you don't care about gaming.

Most people who don't care about gaming also will find that 4c/4t CPU horsepower of BR more than enough. BR on desktop already exists, is used in OEM PCs, and is available for DIY in Germany/Europe and China. The time and availability in the US is disappointing, but there is little doubt that eventually it will make its way here this year.

The only thing that makes sense now is concurrent official US launch with Ryzen 3.

This way US consumers will have it laid out: Strong GPU + slightly weaker CPU at mostly sub $100 (A10) prices, or slightly faster 4-thread quadcore CPU at slightly higher price, > $100, and with bring your own GPU.

Since Summit Ridge is an 8-core design and yields for 8 cores are good, there would not be enough quadcore production (without wasteful measures) without BR quads complementing Ryzen 3. So it is very much needed until native RR quad arrives for desktop.

 

[ET]

I think that BR is mainly a PR problem. It wouldn't get good reviews and it would be hard for reviewers to find a real use case for it. It would only dilute the the reputation of the AM4 platform.

Ryzen is already cheaper than I had expected, with the R5 1400 retailing for $169 and selling for even less. R3 will therefore fit in a price range where I thought BR would slot.

I'll wait and see what R3 turns out to be. There's not much sense in R3 having no GPU, and neither is there much sense in R3 using Summit Ridge dies.