AMD Ryzen SKU and Price Information/Speculation.

[Justinbaileyman]

AMD won't have an answer for the $65 G4560 for at least half a year. It's a great little CPU for 90% of use cases out there.

It was a great move releasing that early. The rest of their lineup... well.. doesn't look so good. Not even close. Their saving grace is that AMD isn't releasing Ryzen 5 and Ryzen 3 along Ryzen 7 this March... that scenario would've been a bloodbath.

Instead those two lines come later, and the APUs mid year to actually counter Intel's iGPU lines... one step at a time I guess, this is a busy year for AMD and ATI

Thats not true at all.. AMD will have R3's to match the G4560 out in April or by May this year the very latest. that is only a month or two so where do you come up with half a year?

 

[.vodka]

The R3s don't have an iGPU (all the current Ryzen lineup is based on the same full Ryzen die just cut down) and if the rumored pricing is correct -which has been correct for the entire R7 lineup-, they start at $120. You still have to add an dGPU. You do get 4 real cores vs 2C4T from the G4560.

$65-70 buys you the 2C4T G4560 (equaling the i3 6100, almost the old i5 2500) in CPU power with an integrated DX12 GPU that's perfect for 90% of use cases out there for average joe.

You can build the G4560 rig for much less than a Ryzen R3 rig. This is my point.

AM4 Bristol Ridge (2 Excavator modules + GCN3 iGPU) could cover that pricing bracket until Raven Ridge arrives, but then 2M Excavator just can't really compete with two Kabylake cores at 3.5GHz plus HT. It sure has lots more GPU horsepower though. Raven Ridge (mid year) gives you at the highest end 4C8T CPU (an i7) and a Vega iGPU. That's going to be insane, even when cut down to go below the $100 line.


Overall, no, they don't have an answer in the <$100 bracket, unless the rumored pricing is wrong and AMD is willing to at least price match the G4560 with a further cut down/downclocked 4C CPU, or they use Bristol Ridge in there.


The G4560 is the only decent product to come out of Kaby Lake for the desktop. The rest is meh.

 

[Topweasel]

Overall, no, they don't have an answer in the <$100 bracket, unless the rumored pricing is wrong and AMD is willing to at least price match the G4560 with a further cut down/downclocked 4C CPU, or they use Bristol Ridge in there.
The G4560 is the only decent product to come out of Kaby Lake for the desktop. The rest is meh.

My bet is Bristol Ridge. Even though the GPU's will eat more die space than cores. They know they can't charge an arm and a leg for them. I suspect the sub 100 sector to be one or two 4c4t APU's and a probably some 2c4t APU's.

 

[DrMrLordX]

You guys are forgetting Socket 939 Opterons. Some of them served as cheaper alternatives to those $400+ Athlon 64's and were very popular.

Who can forget the Opteron 165? That thing was legendary.

I still have my zipzoomfly.com receipt somewhere from when I bought my 3800+ x2 s939 cpu many years ago. IIRC it was $328 shipped.

Edit: I also still have the Asrock mobo and probably the cpu somewhere I think. The mobo doesn't work anymore I don't believe.

I remember when ZZF was a credible alternative to NewEgg that (for some of us) provided a tax-free way to shop. Their selection wasn't as good and their site layout was clearly inferior, but prices were acceptable. I miss them and the old (not crappy) Monarch.

My bet is Bristol Ridge. .

Mine too. Bristol Ridge is going to have to be sub $100 across the board. AMD's own product stack puts it there by default.

 

[Topweasel]

Who can forget the Opteron 165? That thing was legendary.



I remember when ZZF was a credible alternative to NewEgg that (for some of us) provided a tax-free way to shop. Their selection wasn't as good and their site layout was clearly inferior, but prices were acceptable. I miss them and the old (not crappy) Monarch.



Mine too. Bristol Ridge is going to have to be sub $100 across the board. AMD's own product stack puts it there by default.

Not true I could see the $100-$200 scattered with tons of 4c8t Bristol ridges with varying GPU modules and clock speeds. I7 level CPUs with better than irdis iGPU do have a market and the demand for $200 get a fast CPU with a capable iGPU does exist as some level. If the Ryzens chart rumors hold up AMD is obviously targeting just about every realistic and even some niche price brackets as a way to push market share instead of Nvidia and Intel's one size fits all price bracketing per silicon.

 

[unseenmorbidity]

Thats not true at all.. AMD will have R3's to match the G4560 out in April or by May this year the very latest. that is only a month or two so where do you come up with half a year?

r5 is q2, and r3 is q3.

Likely very early on the quarters, but not may.

 

[DrMrLordX]

Not true I could see the $100-$200 scattered with tons of 4c8t Bristol ridges with varying GPU modules and clock speeds.

There will be no 4m/8t Bristol Ridge CPUs, ever. In fact, I doubt they'll release anything more/better than the A12-9800.

 

[.vodka]

There will be no 4m/8t Bristol Ridge CPUs, ever. In fact, I doubt they'll release anything more/better than the A12-9800.

Correct.

<$100 market band aided by 2M4T Bristol Ridge until Raven Ridge arrives, then AMD can drop a truckload of different models to serve everything that requires the iGPU.

I reckon they'd have to be released under different naming to fit Ryzen 3/5/7, the current APU naming just can't be used anymore. These can't be named Ryzen either because they're construction core based.

Are these even available to be bought, or OEM only so far? Bristol Ridge was announced a while ago yet there wasn't any followup..

 

[Topweasel]

Crap meant Raven Ridge. Or more specifically whatever their Zen based APU will be called.

 

[Ajay]

Correct.

<$100 market band aided by 2M4T Bristol Ridge until Raven Ridge arrives, then AMD can drop a truckload of different models to serve everything that requires the iGPU.

I reckon they'd have to be released under different naming to fit Ryzen 3/5/7, the current APU naming just can't be used anymore. These can't be named Ryzen either because they're construction core based.

Are these even available to be bought, or OEM only so far? Bristol Ridge was announced a while ago yet there wasn't any followup..

Why not wait to Raven's Ridge for a 2C/4T. What an amazing waste of die space for Summit Ridge.

 

[DrMrLordX]

Are these even available to be bought, or OEM only so far? Bristol Ridge was announced a while ago yet there wasn't any followup..

The only places that I ever saw the AM4 A12-9800 for sale were some EU/JP stores selling a chip + Asus A320 mobo combo. Now that there are some proper AM4 boards out there, you could buy the combo, ditch/sell the A320 board and . . . wish that you had bought Ryzen instead?

 

[Glo.]

I have been thinking few days ago about Raven Ridge, and the total picture of AMD lineup. Ryzen 3 chips, that are supposed to be 4C/8T are coming 2H/2017. I have asked if this would mean that they are only APUs.

Somebody has answered that they would have offer something more for buyers, if they don't need the iGPU.

And boy...

What do you think of a possibility that the APU is not going to be disabled, but simply, will increase the performance of Vega GPU installed in your computer?

Imagine a stituation. You are buying, a 4C/8T, 16CU+4GB HBM2 APU. Then, after few weeks you end with Vega 11 GPU. You place it in your computer, and you are not disabling the GPU from the APU. Performance from those GPUs adds up!

For that very reason you need HBM on the APU.

 

[Mopetar]

I have been thinking few days ago about Raven Ridge, and the total picture of AMD lineup. Ryzen 3 chips, that are supposed to be 4C/8T are coming 2H/2017. I have asked if this would mean that they are only APUs.
.
.
.
Imagine a stituation. You are buying, a 4C/8T, 16CU+4GB HBM2 APU. Then, after few weeks you end with Vega 11 GPU. You place it in your computer, and you are not disabling the GPU from the APU. Performance from those GPUs adds up!

I don't think the R3 chips are APUs, just the lowest tier of salvage parts for Ryzen. I think that part of the reason they won't come out for a while is that yields are good enough that AMD probably doesn't have a lot of these dies. I also suspect they'd rather have people buy the high-end parts first and if there were R3 and R5 chips available, a few sales would be diverted to those.

Also, AMD has had the ability for their APU graphics to work with a discrete card for a while now: http://www.amd.com/en-us/innovations/software-technologies/technologies-gaming/dual-graphics.

I question if we'll be seeing a 16 CU APU though. That's a lot of die area (expensive) and that many shaders absolutely requires HBM in order to not be severely bottle-necked which means adding expensive HBM to the APU. It's entirely possible for AMD to make such an APU, but the die could end up larger than Ryzen and the margins wouldn't be nearly as good.

I think an 8 CU chip paired with a single Ryzen CCX would make for a reasonable chip that can do well both in the desktop and notebook market. I don't doubt that we'll eventually see a 16 CU APU with HBM as that's the obvious point where they really start to shine, but that means a larger chip and more expensive memory. I think AMD is going to want to wait as 14 nm yields improve (along with additional capacity becoming available) and the cost of HBM is driven down due to adoption and use in other product lines.

However, having the option to drop in a relatively inexpensive discrete GPU and having it run in tandem with the iGPU offers a nice upgrade path for people who want to add a little boost to their system. I think DX12 was contains some features (explicit multiadapter) to make multi-GPU support easier.

 

[Ajay]

I'm very dubious of HBM being added right now. Look at Intel and Iris Pro - the extra cost of the L4$ didn't turn out to be worth it to most buyers.

 

[Glo.]

Margins do not have to be as good, as they are with Ryzen 5 and 7 chips. The market for APUs is MUCH BIGGER, than just for the CPUs.

4C/8T with 16 CU's and 4 GB of HBM2, in two stacks, in 95W TDP package, if performance is right would presumably kill everything on the mainstream and low-end market, even if the APU would cost 300$.

4C/8T+16 CU's die size would be actually around 210 mm2. 44mm2 is CCX, 123 mm2 is die size of Polaris 11, with 4 memory controllers. With HBM2 - you have only 2, and they should be integrated in IMC, that should account for both RAM and HBM2(APUs usually have access to RAM, also).

I'm very dubious of HBM being added right now. Look at Intel and Iris Pro - the extra cost of the L4$ didn't turn out to be worth it to most buyers.

Even if the end result in 16CU Vega GPU would be clock for clock around 40% faster than 1024 GCN core Polaris 11 chip?

We are talking about a situation in which, in games at least, that Vega APU is between GTX 1050 Ti, and RX 470.

There is a very good reason for this whole idea of HBM2.

Lets look at this this way. 4C/8T+16CU APU+4 GB's HBM2, in 95W TDP, and cost of 299$.
Core clocks of CPU 3.4/3.8 GHz, core clocks of GPU 1250 MHz, for example. GPU arch is for example faster by 40% from 1024 GCN core Polaris GPU(1024 GCN core Sapphire RX 460). Idea of APUs is when they bring significant benefits. You end up with 199$ worth of CPU, with 150$ worth GPU, in 50% thermal envelope. For slightly lower cost.

And you can also add to this idea another level of graphical performance. By adding discreet GPU, from the same architecture.

 

[Vesku]

I'm very dubious of HBM being added right now. Look at Intel and Iris Pro - the extra cost of the L4$ didn't turn out to be worth it to most buyers.

A large CU count HBM APU would be marketed first to Console makers and server farms not consumers.

 

[JDG1980]

A large CU count HBM APU would be marketed first to Console makers and server farms not consumers.

Apple might be interested as well.

 

[imported_jjj]

Margins do not have to be as good, as they are with Ryzen 5 and 7 chips. The market for APUs is MUCH BIGGER, than just for the CPUs.

4C/8T with 16 CU's and 4 GB of HBM2, in two stacks, in 95W TDP package, if performance is right would presumably kill everything on the mainstream and low-end market, even if the APU would cost 300$.

4C/8T+16 CU's die size would be actually around 210 mm2. 44mm2 is CCX, 123 mm2 is die size of Polaris 11, with 4 memory controllers. With HBM2 - you have only 2, and they should be integrated in IMC, that should account for both RAM and HBM2(APUs usually have access to RAM, also).

Even if the end result in 16CU Vega GPU would be clock for clock around 40% faster than 1024 GCN core Polaris 11 chip?

We are talking about a situation in which, in games at least, that Vega APU is between GTX 1050 Ti, and RX 470.

There is a very good reason for this whole idea of HBM2.

Lets look at this this way. 4C/8T+16CU APU+4 GB's HBM2, in 95W TDP, and cost of 299$.
Core clocks of CPU 3.4/3.8 GHz, core clocks of GPU 1250 MHz, for example. GPU arch is for example faster by 40% from 1024 GCN core Polaris GPU(1024 GCN core Sapphire RX 460). Idea of APUs is when they bring significant benefits. You end up with 199$ worth of CPU, with 150$ worth GPU, in 50% thermal envelope. For slightly lower cost.

And you can also add to this idea another level of graphical performance. By adding discreet GPU, from the same architecture.

And will anyone pay 500$ for that APU? Even at 500$ it would be lower margins than Summit Ridge in retail as the costs would be 2x.
And why do this to begin with, They don't even invest for a native APU in server.
Then, how does it compare to discrete GPUs? Because those evolve too and would be both faster and cheaper with GDDR.
Plus ofc, how does it compare to a 11-12CPU APU that costs less than half.

In laptop maybe, if they got a large order but even then, they need cheaper packaging solutions not a Si interposer.
They could just do a traditional MCP with a CPU or APU plus a GPU and maybe the GPU would be on an interposer but not the CPU - so like in server.
And ofc, if the OS can use the HBM as system memory, there would be some cost savings as you can use less DRAM.

Bottom line, advanced packaging costs too much for now and some more time is needed before cheaper alternatives are avilable. Right now Intel's silicon bridge is limited but cheap. Access to it is complicated for AMD.
The only way you see large APU now is if they bring the server product in consumer or if they have someone like Apple in laptop insisting for such a product but rumors indicate that Intel CPU+AMD GPU with HMB is more likely at Apple.
In 1-2 years it will become feasible as cheaper alternatives are available but right now it's too costly.

And don't forget that memory bandwidth is only good if you need it.
Plus, the main limitation is TDP not BW.

 

[R0H1T]

Apple might be interested as well.

Apple & ultrabook/gaming notebook makers or even MS.

 

[Glo.]

And will anyone pay 500$ for that APU? Even at 500$ it would be lower margins than Summit Ridge in retail as the costs would be 2x.
And why do this to begin with, They don't even invest for a native APU in server.
Then, how does it compare to discrete GPUs? Because those evolve too and would be both faster and cheaper with GDDR.
Plus ofc, how does it compare to a 11-12CPU APU that costs less than half.

In laptop maybe, if they got a large order but even then, they need cheaper packaging solutions not a Si interposer.
They could just do a traditional MCP with a CPU or APU plus a GPU and maybe the GPU would be on an interposer but not the CPU - so like in server.
And ofc, if the OS can use the HBM as system memory, there would be some cost savings as you can use less DRAM.

Bottom line, advanced packaging costs too much for now and some more time is needed before cheaper alternatives are avilable. Right now Intel's silicon bridge is limited but cheap. Access to it is complicated for AMD.
The only way you see large APU now is if they bring the server product in consumer or if they have someone like Apple in laptop insisting for such a product but rumors indicate that Intel CPU+AMD GPU with HMB is more likely at Apple.
In 1-2 years it will become feasible as cheaper alternatives are available but right now it's too costly.

And don't forget that memory bandwidth is only good if you need it.
Plus, the main limitation is TDP not BW.

First, the aforementioned APU would cost 299$, not 500$.

Secondly, the die cost would be extremely similar to the cost of Ryzen 8C CPU, because of similar die size, for 4C+16CU design.
So pretty much around 20$ per die we are talking about wafer costs.

Thirdly, what APU we are talking about? 4C/8T+12CU is mobile design, with up to 35W TDP. It is called Raven Ridge.
4C/8T+16CU+ HBM2 is most likely Horned Owl APU with TDPs up to 95W, that is targeted for not only Mainstream market, but also embedded, professional, server, and Machine Learning Markets. Those are the markets where the margins can be higher.
Lastly we have Snowy Owl design: 16C/32T, 64CU's, 16 GB of HBM2. Note that it is exactly 4 times bigger design than Horned Owl. Coincidence? Or perfect scaling?

Next is the cost. Whole APU+2 stacks of HBM2 package will not cost more than 40$. Price it at 300$ and you still have huge margin, in a market that is much bigger than CPU only. You are targeting the market where you have BOTH CPU and GPU.

Lastly. Which of the markets would explode with such a product? Small-Form-Factor, efficient designs. NUC's. The markets in which the money lies, currently.

It is the funniest part, that this APU design can be a cure for dying mainstream market.

Im sure we will see this design this year, but it will be at least Q4 2017.

 

[lolfail9001]

Secondly, the die cost would be extremely similar to the cost of Ryzen 8C CPU, because of similar die size, for 4C+16CU design.

It is the HBM2 and interposer it involves that skyrockets the cost, don't you realize that? No, these are NOT getting cheaper.

Lastly we have Snowy Owl design: 16C/32T, 64CU's, 16 GB of HBM2. Note that it is exactly 4 times bigger design than Horned Owl. Coincidence? Or perfect scaling?

Coincidence, because there is like no evidence that such Snowy Owl is even a single package product. AMD's exascale suggestion paper states that there are 2 CPUs based on 2 parts each, so Snowy Owl is certainly a 16C part made out of 2 Zeppelins. But then it specifies that GPU parts are by themselves on the node, ruining any hypothesis about Greenland APU.

Next is the cost. Whole APU+2 stacks of HBM2 package will not cost more than 40$.

Go, provide evidence for that. Because that is funnier to read than bjt's past speculations on clocks.

 

[Glo.]

Go, provide evidence for that. Because that is funnier to read than bjt's past speculations on clocks.

Each Ryzen CPU costs AMD 18-20$ depending on the yield. Fottemberg have said that before, it even is correct if you get the wafer calculator to hand, and try to estimate the yield by yourself alongside with the die size. BitsAndChips also have provided rounded die sizes about the designs for mobile and desktop parts. Mobile part is 170mm2 according to their sources, and desktop is 210mm2, so around Polaris 10, and Ryzen CPUs. Cost of Interposer is 1$ per 10mm2, IIRC. Fiji Interposer cost AMD around 10$, from what we knew about it. So how much HBM2 costs that it would stop AMD to sell it for 299$? 500$ per memory cell?

AMD can sell Polaris 10 GPU with such low markup because the market is bigger than CPU market. If you consider this, and look what APU is, and think about what markets it would be targeted, the market is much bigger, because it combines both: CPU and GPU.

Funniest part IMO. Considering packaging, board, memory, and package of the die itself, 95W APU can be cheaper with 16CU APU can be cheaper to make than Polaris 10 GPU. Because its much less complex, if you think about it.

And you completely forgot the volume, you would produce, which would drive the costs down.

Once again, 16CU Vega GPU with HBM2 on package will be around 40% faster than similar core count and similar core clock Polaris GPU. If we have 1024 core Sapphire RX 460, and it is 10% faster than RX 460, you end with performance exactly between GTX 1050 Ti, and RX 470.

 

[french toast]

Each Ryzen CPU costs AMD 18-20$ depending on the yield. Fottemberg have said that before, it even is correct if you get the wafer calculator to hand, and try to estimate the yield by yourself alongside with the die size. BitsAndChips also have provided rounded die sizes about the designs for mobile and desktop parts. Mobile part is 170mm2 according to their sources, and desktop is 210mm2, so around Polaris 10, and Ryzen CPUs. Cost of Interposer is 1$ per 10mm2, IIRC. Fiji Interposer cost AMD around 10$, from what we knew about it. So how much HBM2 costs that it would stop AMD to sell it for 299$? 500$ per memory cell?

AMD can sell Polaris 10 GPU with such low markup because the market is bigger than CPU market. If you consider this, and look what APU is, and think about what markets it would be targeted, the market is much bigger, because it combines both: CPU and GPU.

Funniest part IMO. Considering packaging, board, memory, and package of the die itself, 95W APU can be cheaper with 16CU APU can be cheaper to make than Polaris 10 GPU. Because its much less complex, if you think about it.

And you completely forgot the volume, you would produce, which would drive the costs down.

Once again, 16CU Vega GPU with HBM2 on package will be around 40% faster than similar core count and similar core clock Polaris GPU. If we have 1024 core Sapphire RX 460, and it is 10% faster than RX 460, you end with performance exactly between GTX 1050 Ti, and RX 470.

what TDP are you basing that performance on?
Edit, dumb question, missed the 95w part.

 

[Glo.]

what TDP are you basing that performance on?

Clock for clock, core for core, based on the architecture analysis. Its average, because it will depend on situation.

IMO in game like Overwatch Vega will be 25% faster than Polaris, clock for clock, core for core, but in game like Doom Vulkan it will be up to 75% faster. Final performance depends on core clocks of the GPU. However.

Bristol Ridge 65W TDP APU has max core clock of 1108 MHz, with 512 cores, on 28 nm process.
Here we have 1024 GCN core chip, with 95W TDP, that is optimized for higher core clocks, and higher IPC at the same time. What this means? AMD changed the thing that pushed down, efficiency, and achievable core clocks for their GPU architectures. Registry Files available to CUs. They have changed how they are handled, and this allowed them to push the core clocks higher, and also will result in better efficiency. This is just my opinion, but I do not see at this moment any other option how they were able to push the core clocks, without increasing the inefficiency.

 

[raghu78]

imo if Raven Ridge comes with anything more than 512 sp (8 Vega NCU) AMD better equip it with 2 GB High bandwidth cache. A 2 Hi stack should be easy to manufacture at good yields and low cost. More importantly a RR APU at 95w with 3.4/3.8 Ghz 4C/8T and 1024 sp can sell for atleast USD 300 if AMD's Vega GPU is allowed to perform to its full potential. I think AMD will get very close to Nvidia in terms of perf/sp wrt perf/cc with the new NCU and Vega architecture. I would not be surprised to see a 1024 sp Vega GPU perform very close to a Rx 470 or even better due to a combination of higher clocks, higher IPC and significant architectural improvements. RR APU is truly capable of being a disruptive technology product.