AMD Bristol/Stoney Ridge Thread

[NTMBK]

ASRock are launching a mini-STX system for AM4: https://www.asrock.com/nettop/AMD/DeskMini A300 Series/index.asp#Specification Could be good for compact Bristol Ridge systems, depending on the price.

 

[amd6502]

It seems BR probably is close to not being in production anymore (they do need to fill whatever numbers Ataribox needs, which they might be able to do from whatever stock of BR they already have).

It would be nice if Stoney (1.2B transistors) could run in AM4 as a $30 starter APU, but it would probably only be able to use only half the memory slots, so it could only be sold as kits with the MB+RAM.

A BR/Stoney hybrid (say 1.8B transistors) at $40 would be as good, and fill the low end demand when Athlon 200GE stock runs out.

With the death of 7LP the foundry is a complete mess. GloFo is throwing everything at 22FDX/12FDX. Mainly, because the FinFET nodes didn't allow them to get the net profit to satisfy the owners. No matter the capacity of FinFET, it will not net a profit. However, there is hope for 22FDX/12FDX to net profit. Which is why the huge push in the above segment, all 300mm are being tooled to do 22FDX.

How expensive would a port be to 28FDS and how large would the perf. boost be at 10W and 20W?
1. for Stoney
2. for a new variant, namely a BR/Stoney hybrid

Then same question for 2. on 22FDX

 

[amd6502]

A heads up to all XV/A-series fans, the FM2+ A8-7680 now finally is available in the US, and at a wonderful bargain too (under $60 shipped). Supposedly the same specs as A10-9700 AM4 variant, but the FM2+ variant also has a 45W cTDP, making it quite good for cheap SFF builds. Like all FM2, I assume the iGPU frequencies are also unlocked, but it only has 6 CU of the 8CU working (384 SPs).

 

[VirtualLarry]

A heads up to all XV/A-series fans, the FM2+ A8-7680 now finally is available in the US, and at a wonderful bargain too (under $60 shipped). Supposedly the same specs as A10-9700 AM4 variant, but the FM2+ variant also has a 45W cTDP, making it quite good for cheap SFF builds. Like all FM2, I assume the iGPU frequencies are also unlocked, but it only has 6 CU of the 8CU working.

Thanks. That actually doesn't sound too shabby, and with the 6CU, might it be faster in some games than the similarly-priced A200GE?

Edit: How's the supply of mATX and mini-ITX FM2+ boards, preferably those with chipsets that support SATA6G and USB3.0 (and hopefully, maybe, with a front USB3.0 header too for modern cases? Even the $20 Rosewill specials have USB3.0 ports, some of them.)

 

[NostaSeronx]

Stoney 28FDS
Cost would be lower to the same as 28nm designs.
Ver A(STM28FD): 1.35x = A6-9220C-derivied shrink : 1.8 GHz to 2 GHz <-- From GlobalFoundries' 28SLP(HPP is 10%, SHP is at least 5-10% over HPP)
Ver B(SS28FDS): 1.25x = A6-9220C-derived shrink : 1.8 GHZ to 2.25 GHz <-- From Samsung's 28LPP(LPP is on par with HPP/SHP)

An in between design: 2C/3CU <-> 4C/8CU aka 4C/6CU
Cost would be lower than Bristol Ridge.
Ver A(STM28FD), the QC at 1.9 GHz at base.
Ver B(SS28FDS), the QC at 2.1 GHz at base.

Now 22FDX not in GloFo's marketing, since they aggressively underestimate it.
It should be 1.5X on top of 28FD/28FDS;
28FD A6-9220C => 2 GHz -> 3 GHz at base.
28FDS A6-9220C => 2.25 GHz -> 3.375 GHz at base.

1.3x to be safe:
28FD QC 1.9 GHz to 22FDX 2.47 GHz
28FDS QC 2.1 GHz to 22FDX 2.73 GHz

22FDX will be more expensive, but allows for a larger shrink than 28nm.
28FDS @ 7-track ≈ Bulk @ 12-track

Which would allow a 22FDX design to return to Piledriver/Steamroller level scaling at a lower track height:

If we expect a new core within a new module. Then, the frequency can be much higher.

How expensive would a port be to 28FDS and how large would the perf. boost be at 10W and 20W?
1. for Stoney
2. for a new variant, namely a BR/Stoney hybrid

Then same question for 2. on 22FDX

It would be flat at 10W-15W, while there will be an increase at 20W-25W(UHP) and 3W-7W(ULP).

The biggest factor is the Adaptive Body Bias/ABB scaling over Adaptive Voltage Scaling/AVS. ABB scaling is better than AVS, as it produces less leakage/power compared to AVS.

 

[amd6502]

Thanks. That actually doesn't sound too shabby, and with the 6CU, might it be faster in some games than the similarly-priced A200GE?

I think the GPU will be more powerful than 3 Vega . Mixed int+fp code multithread is pretty decent on dozers. Pure fp code heavy multithread is not going to be that good, and of course it will loose in 1 and 2 thread loads to big core dual cores.

I think I might be a build addict; I couldn't help myself and ordered one. Going to put the old DDR3 memory from my old laptop to good use. It's going to be a browsing box, that can do casual gaming. If had DDR4 on hand I'd build a 2200g or ge, but I won't for a while (unless DDR4 prices come down).

 

[VirtualLarry]

Link to site to purchase? Does Newegg have them currently?

Edit: I have an older MSI A55M FM2+ board, will it support this APU? Or do all FM2+ boards NEED a BIOS/firmware update to take this APU? (Or is it similar enough to certain prior APUs that it will just drop in to some boards?)

 

[NostaSeronx]

@VirtualLarry
So far only this MSI supports A8-7680/A6-7480.
-> A68HM-E33 V2

Yes, newegg does have them via external vendor.
https://www.newegg.com/Product/Product.aspx?Item=9SIA2F88NE3524
https://www.newegg.com/Product/Product.aspx?Item=9SIA2F88PG5905

 

[amd6502]

Link to site to purchase? Does Newegg have them currently?

Edit: I have an older MSI A55M FM2+ board, will it support this APU? Or do all FM2+ boards NEED a BIOS/firmware update to take this APU? (Or is it similar enough to certain prior APUs that it will just drop in to some boards?)

Yup newegg. You'll have to flash the new bios before using these new A8's for sure; so definitely need to check with the board manufacturer. My guess is no bios will exist for that. MSI tends to only support their newest (two?) A68 boards.

 

[amd6502]

Stoney 28FDS
Cost would be lower to the same as 28nm designs.
Ver A(STM28FD): 1.35x = A6-9220C-derivied shrink : 1.8 GHz to 2 GHz <-- From GlobalFoundries' 28SLP(HPP is 10%, SHP is at least 5-10% over HPP)
Ver B(SS28FDS): 1.25x = A6-9220C-derived shrink : 1.8 GHZ to 2.25 GHz <-- From Samsung's 28LPP(LPP is on par with HPP/SHP)

An in between design: 2C/3CU <-> 4C/8CU aka 4C/6CU
Cost would be lower than Bristol Ridge.
Ver A(STM28FD), the QC at 1.9 GHz at base.
Ver B(SS28FDS), the QC at 2.1 GHz at base.

Now 22FDX not in GloFo's marketing, since they aggressively underestimate it.
It should be 1.5X on top of 28FD/28FDS;
28FD A6-9220C => 2 GHz -> 3 GHz at base.
28FDS A6-9220C => 2.25 GHz -> 3.375 GHz at base.

1.3x to be safe:
28FD QC 1.9 GHz to 22FDX 2.47 GHz
28FDS QC 2.1 GHz to 22FDX 2.73 GHz

22FDX will be more expensive, but allows for a larger shrink than 28nm.
28FDS @ 7-track ≈ Bulk @ 12-track

Which would allow a 22FDX design to return to Piledriver/Steamroller level scaling at a lower track height:

If we expect a new core within a new module. Then, the frequency can be much higher.
It would be flat at 10W-15W, while there will be an increase at 20W-25W(UHP) and 3W-7W(ULP).

The biggest factor is the Adaptive Body Bias/ABB scaling over Adaptive Voltage Scaling/AVS. ABB scaling is better than AVS, as it produces less leakage/power compared to AVS.

Wow, so really 28FDS wouldn't be quite worth it, but 22FDX is another thing.

At the CES talk Lisa never mentioned the milestones where dozer contributed (first consumer 8 core, first consumer cpu to break 5ghz, first x86 to break 4ghz at stock?). All their (consumer) marketing is on Ryzen and none on the A-series. So other than the Chromebook announcements and a few new Stoney SKUs they seem to be not at all giving the old architecture any attention.

Probably any bets on these CMT APUs might have to have a really small budget.

I think it should stay well under 2B transistors, so much closer to Stoney than BR. Or pretty much Stoney but with 4 threads (double the execution units). It could even halve the L2 which would lower the IPC but save wattage and be a net gain under 10W.

A RR-L half the size of RR would be 2.5B transistors. Stoney v2 would have to stay well under 2B to compete. I'm not sure how many transistors another module would add to Stoney's 1.2B transistors. I'm guessing the APU would be 1.5B-1.8B depending on whether one keeps BR's 2MB L2 or halves it. Halving it, and running the module asymmetrically (like big.little) would preserve IPC at Stoney's level, and still run 2+2 threads well for 10W even using the old 28 bulk. So with 22FDX's efficiency advantages that would be some very good 6W APUs for laptops and tablets.

If it were a generation beyond Excavator, who would be designing it? Aren't about 100% of their CPU engineers working on Zen2+3+... while also pondering the generation beyond? How many people would you need and what budget?

 

[NostaSeronx]

If it were a generation beyond Excavator, who would be designing it? Aren't about 100% of their CPU engineers working on Zen2+3+... while also pondering the generation beyond? How many people would you need and what budget?

Three groups:
- Never went to architect Zen, but were working on cancelled or related to Zen projects: Amur(Israel), Nolan(India), K12(USA)
- Never went to architect Zen, but worked on released projects; Carrizo, Bristol, Stoney
- Went to architect Zen to Zen2, and will be moved to 22FDX/12FDX, rather than be released from employment.

First group will do 22FDX/12FDX, then hop on to Zen3/Zen4/Zen5.
Second group will do 22FDX/12FDX, and swapped to NG Zen (3-5) APUs.
Third group will return to Zen5, after 22FDX/12FDX.
and the now mentioned fourth group is the global team that works on all architectures, and is generally the glue for all projects.

I would expect the budget would be similar to Bobcat(40nm Mainstream) as 22FDX/12FDX(New Mainstream) vs Bulldozer(32nm Advanced) as N7/N3(Continued Advanced). It will generally be cheaper to do 22FDX/12FDX than N7/N3.

22FDX is a penny in the bucket of N7/N7+ combined.
12FDX is also a penny in the bucket but relative to GAA nodes.

It is vital to have some retained experience of Planar devices, because of GAA being stacked planar, etc.

 

[Insert_Nickname]

So, Bulldozer is not dead?
No please, no!

Apparently not. I'd sure like to see that too.

To be fair to Excavator, it's still 50% faster per clock then the Cat cores, and can reach quite high frequencies for an entry level CPU. So it's a good AL/GL competitor. Those announced Chromebooks using it actually look interesting.

ASRock are launching a mini-STX system for AM4: https://www.asrock.com/nettop/AMD/DeskMini A300 Series/index.asp#Specification Could be good for compact Bristol Ridge systems, depending on the price.

Why would anyone put a BR APU in one of those, when even the 200GE smokes every last BR in CPU performance? GPU performance? 2200G or bust.

 

[ET]

A BR/Stoney hybrid (say 1.8B transistors) at $40 would be as good, and fill the low end demand when Athlon 200GE stock runs out.

When 200GE stocks run out we'll get the 300GE, using the Raven 2 die. Raven 2 is probably about Stoney's size (but at 14nm). Once that's binned enough, we might get a 2C/2T 2CU $40 SKU, which would be better than a Stoney variant.

I don't see a reason for AMD to produce a Stoney derivative.

 

[dark zero]

The only way a Bulldozer is even decent is on Quad Core on a uArch that uses less than 28nm (22nm to 14)

 

[Abwx]

Why would anyone put a BR APU in one of those, when even the 200GE smokes every last BR in CPU performance

Actually it doesnt, in MT it is comparable to the 200GE, in ST the latter has better perf but with two threads active FI the gap, although substancial, is not that huge as there s no CMT penalty if each thread use only one core/module.

https://www.computerbase.de/2018-09/amd-athlon-200ge-test/2/#diagramm-leistungsrating-anwendungen

Not a massive victory considering that the 9800E is clocked 400MHz lower than the 9700, looking at the 7 Zip score the difference in ST with usual integer based softs shouldnt be as big as it is in Cinebench ST, beside CB seems to favour Zen as in Povray the 9800E is 15% better.

 

[NTMBK]

Why would anyone put a BR APU in one of those, when even the 200GE smokes every last BR in CPU performance? GPU performance? 2200G or bust.

Cheap GPU performance, basically. Otherwise yeah, the Athlon is the better option.

 

[amd6502]

Why would anyone put a BR APU in one of those, when even the 200GE smokes every last BR in CPU performance? GPU performance? 2200G or bust.

And if the 200GE doesn't the 2200GE sure will.

However the dual core A6 has a place in AM4 under $40 and the one thing it wins at against the 200ge is price (not by much though, but I think the price on the A6 can come down to $35 or slightly under).

The A12 would also have a place but price would have to come down to below $70.

I think the A8's would be better off on FM2+ where it isn't competing against the 200ge's.

 

[NTMBK]

The only way a Bulldozer is even decent is on Quad Core on a uArch that uses less than 28nm (22nm to 14)

Except that AMD has announced no plans to make such a part.

 

[bsp2020]

Except that AMD has announced no plans to make such a part.

https://www.techspot.com/articles-info/1776/images/2019-01-05-image-2.png

I agree, especially now that they have new dual core Ryzen mobile. It is a new die different from Raven Ridge. Notice 14nm process it is using. I read an article that says AMD confirmed that it was a different die but I can't find the link at the moment. The chip should be about 100mm2, not much bigger than Stoney. I'm sure AMD is just clearing their 28nm inventory for chrome books and will bring dual core Zen1/+ to low end market.

Has anyone opened heat spreader of newly announced Athlon 220GE? It maybe interesting to see...

 

[jpiniero]

Notice 14nm process it is using. I read an article that says AMD confirmed that it was a different die but I can't find the link at the moment.

That would be much better if they had a native 2C+3CU die.

It's about time AMD put Dozer out to pasture. Hell, with Goldmont Plus Atom is on a higher IPC level than Dozer.

 

[NostaSeronx]

I highly doubt AMD has the capabilities of making a FinFET product than can compete with a FDSOI part. The sheer amount of R&D that requires makes it a pointless endeavor.

It is better for AMD to go 22FDX:
$15(22FDX @ 100 mm squared) vs $60(14LPP @ 100 mm squared)
3 GHz at 6W TDP(Low gate-related capacitance) vs 1 GHz at 6W TDP(High gate-related capacitance)
ABB for boost(No aging effect) vs AVS for boost(Aging effect+Self-heating effect).

Getting Zen to have enough low leakage requires a new architecture. Which was the whole point of Zen-Lite after Jaguar, but AMD couldn't do it. So, everything was shuffled to Chengdu/Shanghai.

GlobalFoundries has only one Fab for FinFETs, while we are nearing the point where they will have two for FDSOI. The FinFET fab is amidst troubles, while the FDSOI fab active now is GlobalFoundries largest capacity fab. Two location HVM with higher capacity for FDSOI means it will be per wafer cheaper than FinFET.

 

[Insert_Nickname]

Actually it doesnt, in MT it is comparable to the 200GE, in ST the latter has better perf but with two threads active FI the gap, although substancial, is not that huge as there s no CMT penalty if each thread use only one core/module.

https://www.computerbase.de/2018-09/amd-athlon-200ge-test/2/#diagramm-leistungsrating-anwendungen

Not a massive victory considering that the 9800E is clocked 400MHz lower than the 9700, looking at the 7 Zip score the difference in ST with usual integer based softs shouldnt be as big as it is in Cinebench ST, beside CB seems to favour Zen as in Povray the 9800E is 15% better.

Okay, properbly a bit too much hyperbole. Should have added "for their price bracket". Here the 200GE is positioned against the 9500 / 9600. The competition for the 9700(E) / 9800(E) would actually be the 240GE or 1200 (requires discrete graphics) if you can still get it. If you're buying either one of those, you're a hairs breath of pocket change from the 2200G. There is simply no reason to choose BR what-so-ever, except if you need graphics power very, very cheaply.

BR performance is not even a fair contest against a quad Zen...

And if the 200GE doesn't the 2200GE sure will.

However the dual core A6 has a place in AM4 under $40 and the one thing it wins at against the 200ge is price (not by much though, but I think the price on the A6 can come down to $35 or slightly under).

The A12 would also have a place but price would have to come down to below $70.

I think the A8's would be better off on FM2+ where it isn't competing against the 200ge's.

I can accept (dual module) BR as a basic Celeron competitor. But since the Athlons launched not more.

Single module Excavator has no place outside basic Chromebooks.

Edit; oh, yeah. Check out the new A8-7680. Basically BR on FM2+. If you can get it, and a board that has support.

 

[VirtualLarry]

BR performance is not even a fair contest against a quad Zen...

Maybe so, but it can still run Windows 7, something that Raven Ridge cannot do. (Although there has been some progress on this, there is a Reddit with a picture of a Raven Ridge rig running Windows 7, but with a dGPU, which kind of defeats the purpose, in my mind.)

 

[amd6502]

https://www.techspot.com/articles-info/1776/images/2019-01-05-image-2.png

I agree, especially now that they have new dual core Ryzen mobile. It is a new die different from Raven Ridge. Notice 14nm process it is using. I read an article that says AMD confirmed that it was a different die but I can't find the link at the moment. The chip should be about 100mm2, not much bigger than Stoney. I'm sure AMD is just clearing their 28nm inventory for chrome books and will bring dual core Zen1/+ to low end market.

Has anyone opened heat spreader of newly announced Athlon 220GE? It maybe interesting to see...

I wouldn't consider this confirmed unless someone has done a delidding on these newer series dual cores (or taken off the heat sink on the mobile dual cores) to confirm a 14nm RR-L die.

To me both these 3000 generation dual cores (including the mobile Athlon) are essentially a 2200u except off by 100 MHz in either direction (2200u is 2.5base 3.4boost).

Their road maps don't include the bottom lower margin product line. It might be a strategic company secret that doesn't need to be announced far ahead of time.

I agree with Nosta that 22FDX would do better for the 10w budget category than big cores on 12nm finfet.

 

[ET]

I wouldn't consider this confirmed unless someone has done a delidding on these newer series dual cores (or taken off the heat sink on the mobile dual cores) to confirm a 14nm RR-L die.

I don't think they're available yet, so delidding isn't yet possible, but AMD had confirmed to Anandtech that the new mobile line is two new chips. Like bsp2020, I can't find it with a trivial search, but I'm sure a little more work would dig that up.

I imagine that the 200GE is purely binned first gen Raven Ridge. It was said that Raven 2 has 4x PCIe, and, assuming that's true, it would conflict with 200GE's specs. I figure we'll have to wait for the 300GE to see Raven 2.