AMD Bristol/Stoney Ridge Thread

[Asterox]

Will definitely not happen. But would be interesting to see, especially with a 'good' L3 cache.

Now is to late for that, it could have been done very simple and pretty cheap.AMD has made a decision, that it will not bake 6 or 8 thread Excavator for AM3+ socket.

https://forums.anandtech.com/threads/amd-bristol-stoney-ridge-thread.2463487/page-29#post-39044652

In reality no one expected it, but still AMD did shot one 4 threads Excavator CPU in dead FM2+ socket, or 2X deceased vs AM3+ socket.Athlon X4 845 is good and low-cost CPU, approximately similar CPU performanse as Phenom II X4, but with 50% lower TDP and power consumption.

In the same time period as with Athlon X4 845, AMD if they wanted it could do same thing with AM3+ socket.

Youtube TechEpiphany chanell, they have a lot gaming video tests with A12-9800 or Athlon X4 970.

https://www.youtube.com/channel/UCV_FbbkkWz4KHNzMlmYO04A/videos

 

[NTMBK]

A 14nm shrink would be more than competitive against Ryzen, moreover with a L3 cache as pointed by Formulav8, and also with a correctly sized L2, but on 28nm it would be hopless perf/watt wise.

It's a crap architecture, and putting it to 14nm would be a staggering waste of money. Never mind releasing it for a dead socket with terrible features like AM3. I don't think that the single digit sales that would make would cover R&D costs.

 

[ao_ika_red]

I know that I late into the party, but here's the price recap of Bristol Ridge APU that I've found in local e-marketplace. I convert these prices from IDR to USD using Google Currency Converter on Aug, 22th 2017.
A6 9500E (2T/4CU): $52.47
A6 9500 (2T/6CU) : $52.47
A8 9600 (4T/6CU): $67.47
A10 9700E (4T/6CU): $97.45
A10 9700 (4T/6CU): $97.45
A12 9800E (4T/8CU): $107.2
A12 9800 (4T/8CU): $107.2
All prices after VAT.
There's no Athlon to be found yet. I suspect they will be available on Sept.

 

[NostaSeronx]

It's a crap architecture, and putting it to 14nm would be a staggering waste of money.

Anything but, the issue is more actual physical design. The architecture without awkward decisions by AMD would rek pretty much anything on the market.

Remove the wasted PRF bindings => 2 PRFs in every processing unit -> Cores: 1 PRF per EX/AGLU, and FPU: 1 Lower 91-bit and 1 Upper 73-bit PRF(P0L 80-bit FMAC, P1L 80-bit FMAC, P0U 64-bit, P1U 64-bit, P2 128-bit MMX)
to Cores: 1 PRF to all units, and 1 >128-bit PRF to all FPU pipes. etc.

An overhauled architecture with mostly power/density/CPI oriented improvements can make 15h really, really fast and brainy. My calculations point to it being faster than Ryzen by a very small weighted margin. While, requiring no more than a Piledriver to Streamoller like change in units. Meaning to the normal person, it is completely compatible to bdver1 -> bdver4 optimization & 15h SOG.

14LPP and 28HPA have the same total cost by 2019, 22UHP is ~7/10ths of the former TCO. 28HPA with 8T to 22UHP with 8T will be like going from 8T to 12T in performance. With body biasing it is more like going from 8T to 14T. So, there is that in regards with 22FDX.

Release a SKU with the above in Bristol Ridge/Stoney Ridge SKU format (15h 65-7Fh compatibility). Do the optimizations in UHP, successor to the FX-9800P/A12-9800E would be looking at ~45-ish% boost in clock. Including the architectural enhancements, ideally ~60-ish% boost in performance. While remaining around the $99-$149 mark, and getting larger margins than BR/SR did in 2016.

 

[VirtualLarry]

If we compare Cinebench R15 scores(i added a couple of 4 thread processors for a better comparison), first Bulldozer CPU-s where garbage for a big severe headache.FX-4100 is barely faster than X4 Athlon, and obviously Piledriver is very similar story or a serious headache.

AMD is logicaly gone all in for Zen/Ryzen. But at least one 8 Threads Excavator CPU(on very cheep 28nm) for old AM3+ socket, well that would by a nice last greetings or goodbye.

Nice chart. Puts it all in perspective, between the Athlon II X4 640 that my friend uses for a daily-driver, the G4560 I sold him that he's not using yet, for a gaming rig, and the Ryzen R3 1200 CPUs that I ordered, that I plan on building. (They are a better deal than the i3-6100, which says a lot, really, considering that they are also cheaper. I don't understand those that defend Intel, in the face of the Ryzen release.)

Edit: Regarding my comment about the i3-6100 and Ryzen R3 1200 - it's quite possible that the i3-6100 has better single-threaded speed, due primarily to its higher clock speed out of the box. But MT performance advantage goes to the Ryzen 3 1200, which is a true quad-core with Broadwell-esque IPC, for less than Intel charges for a Skylake dual-core with HT.

 

[Abwx]

7.2

There's no Athlon to be found yet. I suspect they will be available on Sept.

Supply is different in Europe, actually the Athlon 950 is more available than the APUs although it s already out of stock in France...

https://geizhals.de/?cat=cpuamdam4&xf=12099_Desktop~820_AM4

http://www.ldlc.com/informatique/pi...00/p1e48t3o0a1+fb-C000000805+fv160-15394.html

 

[ao_ika_red]

Supply is different in Europe, actually the Athlon 950 is more available than the APUs although it s already out of stock in France...

It really paints the picture of AMD marketing priority. Because in my country, casual users tend to "buy and forget" their computer. As long as it's very cheap and capable to play e-sports smoothly (30fps+ is considered as smooth gameplay for majority), they will buy it.

 

[Asterox]

Supply is different in Europe, actually the Athlon 950 is more available than the APUs although it s already out of stock in France...

https://geizhals.de/?cat=cpuamdam4&xf=12099_Desktop~820_AM4

http://www.ldlc.com/informatique/pi...00/p1e48t3o0a1+fb-C000000805+fv160-15394.html

At the moment or now, "availability is very low or arriving soon" but this is expected.BristolRidge will soon sit in their place, as a example czech Alzashop who sells or delivers to all EU countries.

https://www.alzashop.com/amd-processors-with-socket-am4/18861238.htm

 

[amd6502]

If we compare Cinebench R15 scores(i added a couple of 4 thread processors for a better comparison), first Bulldozer CPU-s where garbage for a big severe headache.FX-4100 is barely faster than X4 Athlon, and obviously Piledriver is very similar story or a serious headache.

AMD is logicaly gone all in for Zen/Ryzen. But at least one 8 Threads Excavator CPU(on very cheep 28nm) for old AM3+ socket, well that would by a nice last greetings or goodbye.

Would've been nice for such an AM3+ update from consumer point of view, but I don't think board makes would be thrilled at this point to offer a BIOS update. (Also, the FX's have always been co-produced with server CPUs, and unless 22FDX excavator beats Zen in perf/watt there'd be no place for excavator in that market. So as far as AMD is concerned: not enough derivative products and they would be competing with their own Zen products. )

Piledriver was decent, especially for its time, and excavator beat all expectations of what a narrow 4 wide core could do (k10 core was 6 wide; 3+3 ALU and AGUs versus dozer's 2+2). In the chard above Piledriver 4350 scores the same as the a10-7890k: 331. http://benchmarks-tests.com/cpu_processors/amd_fx-4350/cinebench.php However, its wattage is about 30% higher than the 4100 and 7890k's, and it has twice the L3 of the 4300 and 4100. The 20% improvement in PD over BD, 331 vs 270 earned Bulldozer1 a funny nickname that stuck.

https://www.computerbase.de/2017-01...m-multitasking-test-the-witcher-3-plus-winrar

We can see that the integer load execution efficency is good on all CPUs when an FP load is fired at the same time but this latter is efficently executed (with said integer load also running) only on i7s and FXs.

Those are interesting test results. Is winrar very memory intensive by any chance. I'm suspecting the memory controller (which were never very strong on dozers) are maxed out. Looking at the FX 6300 vs steamrollers results, it looks like the L3 (and not just the extra 2 threads) might be of big help.

It's too bad the dual core Bristol Ridge's can't make use of healthy L2 extra cache (as say a victim cache like in zen) from the disabled faulty module. It does seem excavator's smaller yet faster L2 without L3 to back it up can sometimes be a handicap.

Anything but, the issue is more actual physical design. The architecture without awkward decisions by AMD would rek pretty much anything on the market.
[...]
Release a SKU with the above in Bristol Ridge/Stoney Ridge SKU format (15h 65-7Fh compatibility). Do the optimizations in UHP, successor to the FX-9800P/A12-9800E would be looking at ~45-ish% boost in clock. Including the architectural enhancements, ideally ~60-ish% boost in performance. While remaining around the $99-$149 mark, and getting larger margins than BR/SR did in 2016.

Given the narrow core limitation it would have a really hard time competing with the big cores. If excavator is going to continue it needs to carve out niche markets for which it makes sense, and in projects with lots of niche derivative products. I would really like to see excavator in action one more time at least. And the fact that stoney gets a refresh according to the road map, while bristol doesn't---that's completely mind boggling to me.

Yeah, dual-core is pretty-much nearly dead at this point, going forward.

The dual thread decade 2005 - 2015 had a pretty good run earlier on but I can only see it ending over the next few years; before this decade is over for sure. Acorn is doing the right thing with its big.little

 

[Ancalagon44]

So, I want to go to Ryzen eventually, and I was thinking of buying an AMD AM4 motherboard, DDR4 memory, and an Athlon X4 950 just so that I am on DDR4 and ready to buy Ryzen 2 when it comes out. I currently have a Thuban 1055T.

To my horror, I discovered that an Athlon X4 950 would be a sidegrade in terms of performance in most games and a regression in other areas. So, my Thuban is now 7 years old. Even with the advantage of a smaller and more efficient process node (28nm vs 45nm), and years of refinements to the Bulldozer architecture, and a higher clockspeed (2.8Ghz vs 3.5Ghz), an Excavator based CPU still performs around the same level as a Thuban CPU.

In certain benchmarks involving new instructions, it would be a lot faster. In everything else, it is about the same.

What I would get is 2 less cores in exchange for 30W less power used.

This to me illustrates just how bad the Bulldozer architecture is and why AMD abandoned it completely. You cannot fix it. AMD tried to fix it from Bulldozer to Piledriver to Steamroller to Excavator, and it still is not really much better than Thuban perhaps except in energy efficiency (and that only recently).

 

[Insert_Nickname]

So, I want to go to Ryzen eventually, and I was thinking of buying an AMD AM4 motherboard, DDR4 memory, and an Athlon X4 950 just so that I am on DDR4 and ready to buy Ryzen 2 when it comes out. I currently have a Thuban 1055T.

To my horror, I discovered that an Athlon X4 950 would be a sidegrade in terms of performance in most games and a regression in other areas. So, my Thuban is now 7 years old. Even with the advantage of a smaller and more efficient process node (28nm vs 45nm), and years of refinements to the Bulldozer architecture, and a higher clockspeed (2.8Ghz vs 3.5Ghz), an Excavator based CPU still performs around the same level as a Thuban CPU.

In certain benchmarks involving new instructions, it would be a lot faster. In everything else, it is about the same.

What I would get is 2 less cores in exchange for 30W less power used.

If you're on a budget, you can get a 1400/1500X to tie you over. Maybe a 1300X if you're really going cheap, but Zen really benefits from SMT. The 1200 isn't worth the money IMHO.

This to me illustrates just how bad the Bulldozer architecture is and why AMD abandoned it completely. You cannot fix it. AMD tried to fix it from Bulldozer to Piledriver to Steamroller to Excavator, and it still is not really much better than Thuban perhaps except in energy efficiency (and that only recently).

Exactly. But you really have to give credit to AMD's engineers, turning the BD disaster into something half decent with EX...

 

[DrMrLordX]

At this point, if you are seriously looking at the x4 950, you might want to look at R3s instead. The major downside to XV is that you will never get it in anything but a 2m configuration, making it a hard sell to someone with 6c Thuban. Maybe in certain SIMD-heavy workloads would you see some benefit. But an R3 would be a much better buy, as would R5 and R7.

 

[amd6502]

At this point, if you are seriously looking at the x4 950, you might want to look at R3s instead. The major downside to XV is that you will never get it in anything but a 2m configuration, making it a hard sell to someone with 6c Thuban. Maybe in certain SIMD-heavy workloads would you see some benefit. But an R3 would be a much better buy, as would R5 and R7.

Unless it's something temporary and experimental , why go from 6 to 4 threads? I agree with Insert_Nickname on SMT being very worth the small premium; either wait for RR AM4 or something like 1600 1500x.

Or if bored for another semi-sidegrade, trade in your thuban: http://cpuboss.com/cpus/AMD-Phenom-II-X6-1055T-vs-AMD-FX-8300

BR unfortunately inherited a no frills budget design given that it was designed around the fm2 budget platform and then made compatible with DDR4 (carrizo) and later AM4/FP5 (BR). Don't forget that more than half its die space is dedicated just to the iGPU. The die are would be smaller than Stoney's 125mm2 if it weren't for the iGPU. If following the FX's it would have twice the cores and and L3 had it ever made it to AM3 or AM4. I would've been thrilled at 6c/6t excavators with a lite iGPU. The reality is only a dual threaded Stoneys extreme ultra budget product is in the future.

 

[Abwx]

Those are interesting test results. Is winrar very memory intensive by any chance. I'm suspecting the memory controller (which were never very strong on dozers) are maxed out. Looking at the FX 6300 vs steamrollers results, it looks like the L3 (and not just the extra 2 threads) might be of big help.

It's too bad the dual core Bristol Ridge's can't make use of healthy L2 extra cache (as say a victim cache like in zen) from the disabled faulty module. It does seem excavator's smaller yet faster L2 without L3 to back it up can sometimes be a handicap.

That s it, the L3 cache make all the difference, just compare the i5s to i7s FI, unfortunately there s no FX43XX in thoses tests but i wouldnt be surprised if they are quite better than Kaveri for this test.

This to me illustrates just how bad the Bulldozer architecture is and why AMD abandoned it completely. You cannot fix it. AMD tried to fix it from Bulldozer to Piledriver to Steamroller to Excavator, and it still is not really much better than Thuban perhaps except in energy efficiency (and that only recently).

It illustrate that Kaveri and XV have too much throughput for the available cache, even Piledriver is sensitive to cache size to some extent since the 8MB L3 FX4350 has 5% better IPC than the 4MB L3 equipped FX4300, and that s for two modules only...

The Phenom 840 and 955 only difference is the presence of an L3 cache and the result is 10% higher perf/Hz...

http://www.hardware.fr/articles/913-7/cpu-performances-applicatives.html

 

[amd6502]

It illustrate that Kaveri and XV have too much throughput for the available cache, even Piledriver is sensitive to cache size to some extent since the 8MB L3 FX4350 has 5% better IPC than the 4MB L3 equipped FX4300, and that s for two modules only...

The Phenom 840 and 955 only difference is the presence of an L3 cache and the result is 10% higher perf/Hz...

I've definitely had issues with the weak memory controller (and/or lack of the L3 to compensate) on my A10 Kaveri laptop (4 threads). POGS is a BOINC project that seems to use a lot of memory access. Unless I downclock the 2 cores it runs on to close to the minimum clock rate, the system's responsiveness and Firefox performance suffers very noticeably. (Luckily with steamrollers the cores on same module can be clocked differently, so cores 0,3 can chug away very slowly and I'm left with two decently strong threads.)

BadReIgion/TechEpiphany has put out a great review of the A12. https://www.youtube.com/watch?v=uyScrZxYbwI Its CPU performance is a good bit weaker than the the newest 4 thread Pentium G due to lack of single thread oomph versus the big core rival. Mainly the real selling point of it is its advanced iGPU; making 900p gaming build cost super low for those who just need a work/school desktop with occasional casual or strategy themed gaming.

 

[Ancalagon44]

Unless it's something temporary and experimental , why go from 6 to 4 threads? I agree with Insert_Nickname on SMT being very worth the small premium; either wait for RR AM4 or something like 1600 1500x.

Or if bored for another semi-sidegrade, trade in your thuban: http://cpuboss.com/cpus/AMD-Phenom-II-X6-1055T-vs-AMD-FX-8300

BR unfortunately inherited a no frills budget design given that it was designed around the fm2 budget platform and then made compatible with DDR4 (carrizo) and later AM4/FP5 (BR). Don't forget that more than half its die space is dedicated just to the iGPU. The die are would be smaller than Stoney's 125mm2 if it weren't for the iGPU. If following the FX's it would have twice the cores and and L3 had it ever made it to AM3 or AM4. I would've been thrilled at 6c/6t excavators with a lite iGPU. The reality is only a dual threaded Stoneys extreme ultra budget product is in the future.

Buying another AM3 processor is pointless for me. I want to upgrade to AM4 - my worry is that I know Ryzen 2 or Ryzen+ comes out next year, and that is what I would rather have.

So, hence why I was thinking of biting the bullet on an AM4 upgrade now with a cheap AM4 CPU, to upgrade to a better Ryzen 2 or Ryzen + CPU next year once they are available.

Some of my thinking around this is that I am somewhat more patient than other tech enthusiasts. Or, perhaps I should say cheaper - I really don't like spending a lot of money on gaming hardware. So, perhaps I should just wait for Ryzen 2. I will have owned my Thuban for something like 5 years at that point.

The other thing is that the jump from Thuban to Ryzen (any Ryzen) is a lot bigger than the jump from Ryzen to Ryzen 2 will be. So, I could even upgrade to Ryzen now, and then upgrade to Ryzen 2+ ultra edition in a few years. Hopefully it stays compatible with socket AM4.

 

[Insert_Nickname]

The reality is only a dual threaded Stoneys extreme ultra budget product is in the future.

If they can get it up to 4GHz+, a Stoney refresh would be a very good budget product. But I doubt 28nm can do that, and maintain low power consumption. I also doubt AMD will port anything BD to 14nm.

I have briefly used a laptop with an A9-9420 (Stoney @ 3.0Ghz/3.6GHz with 15W TDP). Was very good for a budget product, far better then your average small core Intel counterpart.

I'd really like to see a budget "mini-CCX" Ryzen with just 2 cores + 4MB L3 for the really budget oriented market. The L3 would somewhat offset the OEM penchant for only using single channel memory. Add a basic 256SP Vega IGP, and you have a decent cat core replacement product. But I don't think that's happening.

 

[amd6502]

I have briefly used a laptop with an A9-9420 (Stoney @ 3.0Ghz/3.6GHz with 15W TDP). Was very good for a budget product, far better then your average small core Intel counterpart.

That is pretty impressive; sounds good for bare essentials or a mmc netbook. I wish they could throw the poor AM1 folks a bone with Stoney Athlons if such a thing were possible.

It looks like they're getting good results and nice yields from that old 28nm; that process is so much improved from the Kaveri days. These aren't all that many datapoints but both the Athlon 950 and the A8-9600 look to be very good overclockers (as in, just about almost A12 sort of frequencies): reddit.com/r/Amd/comments/6wefrk/i_guess_we_have_a_new_budget_king_the_a89600_is/

The other thing is that the jump from Thuban to Ryzen (any Ryzen) is a lot bigger than the jump from Ryzen to Ryzen 2 will be. So, I could even upgrade to Ryzen now, and then upgrade to Ryzen 2+ ultra edition in a few years.

Yes it's going to be some sort of jump. You might want to wait until around spring 2018. When Raven Ridge lands on AM4 early next year I think they could well have improved frequencies from the current Ryzens (although they'll be limited to 4c/8t max) from their new node maturing. At some point (zen+ maybe?) they'll have 14nm LPP+, and therafter I think it goes to 7nm.

A RR 4c/8t with minimal to no iGPU with very good frequencies would be a good fit for gamers and builders who want both lower budget and good performance.

I thougth some of the roadmaps did suggest a 2c/4t APU die was in the works (hopefully waiting for Zen+ and 14nm LPP+ to be ready).

 

[Insert_Nickname]

That is pretty impressive; sounds good for bare essentials or a mmc netbook.

It is. I would -think- it boils down to a severe frequency advantage compared to small core Intel. Bay Trail/Braswell/Apollo Lake simply can't match that nice 3.6GHz turbo boost. Compared to those, EX actually has an IPC advantage too, and huge L1 caches*.

I can't even begin to tell how much of an improvement on old Bobcat netbooks Stoney represents. It is really in a different class altogether.

*BT/BW/AL has 24KB data and 32KB instruction cache. Compare with 64KB data and 96KB(!) instruction on Stoneys EX module...

A RR 4c/8t with minimal to no iGPU with very good frequencies would be a good fit for gamers and builders who want both lower budget and good performance.

When RR lands, I wouldn't be surprised for the 4C/8T SKUs to migrate to that die. But that is pure speculation.

 

[ao_ika_red]

If they can get it up to 4GHz+, a Stoney refresh would be a very good budget product. But I doubt 28nm can do that, and maintain low power consumption. I also doubt AMD will port anything BD to 14nm.

I have briefly used a laptop with an A9-9420 (Stoney @ 3.0Ghz/3.6GHz with 15W TDP). Was very good for a budget product, far better then your average small core Intel counterpart.

I'd really like to see a budget "mini-CCX" Ryzen with just 2 cores + 4MB L3 for the really budget oriented market. The L3 would somewhat offset the OEM penchant for only using single channel memory. Add a basic 256SP Vega IGP, and you have a decent cat core replacement product. But I don't think that's happening.

SR was designed as a bridge between Puma+ core with BR. Heck, even my old Bobcat still trades blow with Apollo lake. But, because it only has 2T, Carrizo-L (Puma+) is still my preferred choice to somebody who need day-to-day cheap laptop.

There should be "small" RR called Banded Kestrel with 2C/4T, 192SP, and 6-15 W cTDP. The initial spec suggests it will land first on Enterprise market due to its ECC support. But because it has similarities with SR, there's no reason to not release it to the market.

 

[dark zero]

It's a crap architecture, and putting it to 14nm would be a staggering waste of money. Never mind releasing it for a dead socket with terrible features like AM3. I don't think that the single digit sales that would make would cover R&D costs.

Heck, even Intel with greater incomes ditched Broxton which would be great considering how badly ARM is doing...

 

[Asterox]

TechEpiphany did a quick A8-9600 overclocking, or 1ghz jump to 4.1ghz in snap of a finger.

 

[Arachnotronic]

Heck, even Intel with greater incomes ditched Broxton which would be great considering how badly ARM is doing...

Broxton lives on in Apollo Lake.

 

[kwalkingcraze]

Looks like A6-9500e, the new cheapest AM4 APU, is receiving identical benchmark scores with Celeron G1820. So, it's January 2014 all over again like Celeron G1820, except this time the name is A6-9500e with better graphics.

https://www.cpubenchmark.net/cpu.php?cpu=AMD+PRO+A6-9500E&id=2950

 

[AtenRa]

Looks like A6-9500e, the new cheapest AM4 APU, is receiving identical benchmark scores with Celeron G1820. So, it's January 2014 all over again with Celeron G1820, but this time the name is A6-9500e with better graphics.

https://www.cpubenchmark.net/cpu.php?cpu=AMD+PRO+A6-9500E&id=2950

A6-9500E and every BristolRidge E model is 35W only, Celeron G1820 is 53W with an iGPU only for playing movies.