Speculation: Ryzen 3000 series

[Mockingbird]

My point with that is that by that point they will have already salvaged these chips from their high margin product lines. Just as AMD will have a bunch of 4-6 core dies that they will make the same decision with. But I am guessing clock and leakage binning will lead core binning and that very quickly they will have an over run of good 8c dies.

But all that is my point all wrapped up together. If its not their high margin product ranges then it goes in their lower margin product ranges. That means less expected profit per chiplet. It's still a win because they found a use for it. Eventually they will still be getting way more good dies that for Epyc, ThreadRipper, Ryzen that they have to take good cores and speed and core bin them not on ability but on volume demand. Until AMD is selling them so quickly that they are 30 days behind on shipments, AMD's biggest goal is to sell as many chips at as high as they can. They can't do that with the AM4 chip if it's priced well into Threadripper range without the Threadripper bonus's. Then they have perfectly good 8c dies not selling, eventually they get backed up enough that AMD has to sell them as 8c solo chiplets for $300, or 200 for 6c or $150 for 4c. Where as they could get 90% of the value of 2 8c chips and sell a ton more at $500 and part of that near doubling in price is without having a second IO die, meaning adding the extra chiplet and not quite doubling the price might still net them more profit.

You forget something.

AMD doesn't own its foundry, unlike Intel.

For Intel, once the initial overhead (development cost of 14nm, 10nm, etc) is paid for, the cost to make each additional die is close to neglectable.

For AMD, each additional die still costs a fixed amount (that AMD agrees to pay to TSMC).

There are not going to be glut of dies that AMD can't sell because AMD would simply cut production.

It is impossible that AMD would drop prices to the floor and flood the market without AMD owning a foundry.

 

[Schmide]

Look you can still have an opinion on all the sources here. What you shouldn't do is continue to affirm or refute it as a source. Just state your opinion and get over it, stop trying and change peoples minds, especially if you have already stated your opinion on something.

Edit: Oh and feel free to go back in time and disagree with all my posts. I care not.

 

[trollspotter]

Pointless stuff removed...

 

[Schmide]

Comment removed. Discussion here.

http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/the-like-disagree-is-too-open-to-abuse.2560150/

 

[trollspotter]

Pointless arguing removed.

 

[Mockingbird]

removed

 

[ToKillAMockingbird]

Nvm.



You cannot have two accounts here.


esquared
Anandtech Forum Director.

 

[Mockingbird]

ignore this post and move along

 

[trollspotter]

It said "leaks" right in his post.

Maybe, you should learn how to use the find function in your browser.

I read it twice and still didn't see where he said leaks, until now. Sorry about that.

 

[esquared]

Mockingbird and trollspotter?

Stop with this nonsense.


esquared
Anandtech Forum Director

 

[Topweasel]

You forget something.

AMD doesn't own its foundry, unlike Intel.

For Intel, once the initial overhead (development cost of 14nm, 10nm, etc) is paid for, the cost to make each additional die is close to neglectable.

For AMD, each additional die still costs a fixed amount (that AMD agrees to pay to TSMC).

There are not going to be glut of dies that AMD can't sell because AMD would simply cut production.

It is impossible that AMD would drop prices to the floor and flood the market without AMD owning a foundry.

Sorry this is some nonsense. No one is suggesting AMD could even compete with Intel and flood the market with CPUs. But being a better option doesn't mean you sell the most units of the two market has shown that in the past.

The problem is the same when people talk about AMD increasing their ASP to survive. It will make things better but we can't base the AMD expectation of profits on Intel's. Even Intel isn't going to maintain that. Margins going to tank this year due to Coffee lake R, Comet Lake, and Cascade Lake are going to eat their profits alive. Point is that just because Intel gets X amount of profit doesn't mean AMD needs to get just as much.

In the end if you are AMD at this point of time your answer to success isn't to throw away marketshare and mindshare for margins. If that was the case they wouldn't even offer Ryzen. If AMD can sell more chiplets by lowering prices they are going to do that. Not tell TSMC to slow down their product. Obviously it has some limit. But it's going to be a lot cheaper than you think.

 

[amd6502]

That's part of the reason why I don't think 16 cores on AM4 is happening. Save that for Threadripper.

12c is enough for upper enthusiasts but 16c would find excellent niche use. Imagine a 65W (35W cTDP) 32t processor that allows you to use a cheap AM4 board as home or enterprise server.

With big.little aSMT threads (16 strong + 16 nice) and somewhat modest base frequencies (not too far north of optimum perf/watt) the two memory channels wouldn't even be a typical limitation.

Those who don't need the bandwidth and IO, and prefer matx form factors and small desktop cases, can save a few hundred $ and pass on the TR4 motherbird.

 

[amd6502]

To all who think that the leaked prices are unrealistic:

The 8 core chiplet is a universal part which will be widely used, produced in high quantities and it is small at the same time: it will be cheap.

The 6 core chiplet is a broken part, when counting 8c and 6c chiplets, yield will be approaching 100%.

Can they be binned before packaging?

Why would you rule out server products using 6c chiplets? In all previous epyc and opteron lines (in the past decade) you had lower priced server parts with disabled cores, as well as inferior frequencies and perf/watt versus their prime costliest models.

Did you see them give away Summit Ridge 6c dies because it was regarded as silicon refuse just because it wasn't suited for epyc? Did they hand out 6c PR because these dies were garbage as far as threatripper?

It is unrealistic and completely wishful thinking. Why set yourself up for disappointment?

 

[Mockingbird]

Sorry this is some nonsense. No one is suggesting AMD could even compete with Intel and flood the market with CPUs. But being a better option doesn't mean you sell the most units of the two market has shown that in the past.

The problem is the same when people talk about AMD increasing their ASP to survive. It will make things better but we can't base the AMD expectation of profits on Intel's. Even Intel isn't going to maintain that. Margins going to tank this year due to Coffee lake R, Comet Lake, and Cascade Lake are going to eat their profits alive. Point is that just because Intel gets X amount of profit doesn't mean AMD needs to get just as much.

In the end if you are AMD at this point of time your answer to success isn't to throw away marketshare and mindshare for margins. If that was the case they wouldn't even offer Ryzen. If AMD can sell more chiplets by lowering prices they are going to do that. Not tell TSMC to slow down their product. Obviously it has some limit. But it's going to be a lot cheaper than you think.

As I said before, AMD already dominates the ~$200 and under market.

AMD doesn't need to drop prices to compete with itself.

Intel dominates the $300+ market with processors such as the Core i7-8700K, Core i7-9700K and Core i9-9900K.

The $300+ market is where AMD can make inroads and get marketshare.

That's where the processors such as the $329.99 8C/16T Ryzen 7 3700X (comparable to Core i9-9900K in performance) and the $499 12C/16T Ryzen 9 2800 come in.

 

[somethingclever]

AMD doesn't own its foundry, unlike Intel.

Because running a foundry is expensive business.

For Intel, once the initial overhead (development cost of 14nm, 10nm, etc) is paid for, the cost to make each additional die is close to neglectable.

Yes, but you forget Intel also has to run/maintain its foundries which also introduces non-negligible cost per wafer (depending on your definition of negligible).

For AMD, each additional die still costs a fixed amount (that AMD agrees to pay to TSMC).

Is that how TSMC charges for it's foundry business? I was under the impression cost would be more along the lines of cost/wafer. But that does extend to cost/die in the end too. Just semantics that I want to clarify.

There are not going to be glut of dies that AMD can't sell because AMD would simply cut production.

It is impossible that AMD would drop prices to the floor and flood the market without AMD owning a foundry.

It wouldn't be considered flooding the market if AMD can sell all it's chips...
I also don't see how Intel owning it's own foundries makes it in any way immune to cost dynamics of companies without them. If anything its an extra cost burden on Intel, and we saw that play out with their delayed 10nm.

Let's not forget that AMD cut GF from it's business to cut costs.

 

[itsmydamnation]

Can they be binned before packaging?

Yes, even things like HBM are sold as known good die stacks

 

[Topweasel]

As I said before, AMD already dominates the ~$200 and under market.

AMD doesn't need to drop prices to compete with itself.

Intel dominates the $300+ market with processors such as the Core i7-8700K, Core i7-9700K and Core i9-9900K.

The $300+ market is where AMD can make inroads and get marketshare.

That's where the processors such as the $329.99 8C/16T Ryzen 7 3700X (comparable to Core i9-9900K in performance) and the $499 12C/16T Ryzen 9 2800 come in.

OKaaaaay I think we are talking the same thing. Why did you say the prices are too low. The price list you responded to had a 16c top clock "3900x" at $499 and you said it's too low. They should have between 3 and 6 SKU's that will fit between the 8c chips and the 16c chips (they might forgo a low power 8c chip if they do single chiplet imho due to the heavily reduced power requirements). Perfect for stair stepping between $300 and $500 in $25-$50 increments.

You have been making it sound like AMD is worried about competing against threadripper. I doubt that's a worry and in a way they might do better in terms of a total silicon cost (not counting the recoup on defective dies) by selling a 16c much cheaper then past 16c TR's. But if it's the lower cost CPU's you have an issue with I still have to ask why do you feel that way? I mean A.) AMD has to keep the value stuff going it's as important in the refreshes as the halo models without them they don't get any OEM penetration and mindshare is at this point as important as Market share. And B.) I am now lost on what you think AMD is competing against themselves on the low end. Just previous gen stuff? I doubt canabalizing previous gen parts plays much in their pricing. If anything they do what every company in every market cept Nvidia/Intel/Apple do and lower the MSRP and off vouchers to help people clear their shelves before new the new product launch. You just need to look at 1700x and 2700x and 1950 vs. 2950 pricing to see the extreme version of what AMD does.

 

[Mockingbird]

Because running a foundry is expensive business.

Yes, but you forget Intel also has to run/maintain its foundries which also introduces non-negligible cost per wafer (depending on your definition of negligible).

Well, yes, there's a cost to keeping the foundries running, but it's neglectable compare to the initial cost of setting it up in the first place (overhead).

Once the overhead is paid for, Intel is basically printing money.

Is that how TSMC charges for it's foundry business? I was under the impression cost would be more along the lines of cost/wafer. But that does extend to cost/die in the end too. Just semantics that I want to clarify.

Yes, TSMC charges cost/wafer.

I just simplified it down.

It wouldn't be considered flooding the market if AMD can sell all it's chips...
I also don't see how Intel owning it's own foundries makes it in any way immune to cost dynamics of companies without them. If anything its an extra cost burden on Intel, and we saw that play out with their delayed 10nm.

Let's not forget that AMD cut GF from it's business to cut costs.

Since AMD is paying a fixed cost to TSMC for each wafer, AMD is just going to order enough wafers to meet demands.

There's not going a be huge stacks of wafers (that AMD can't sell) sitting somewhere (which would have forced AMD to lower the prices).

________________________________________________________________________

On the other hand, (after the overheard is paid for), each wafer that Intel output is practically free.

Even if Intel has a surplus, it doesn't make sense to cut production output because the cost to produce is so low.

So instead, Intel would lower the prices of each processors to move more unsold units.

 

[Topweasel]

12c is enough for upper enthusiasts but 16c would find excellent niche use. Imagine a 65W (35W cTDP) 32t processor that allows you to use a cheap AM4 board as home or enterprise server.

With big.little aSMT threads (16 strong + 16 nice) and somewhat modest base frequencies (not too far north of optimum perf/watt) the two memory channels wouldn't even be a typical limitation.

Those who don't need the bandwidth and IO, and prefer matx form factors and small desktop cases, can save a few hundred $ and pass on the TR4 motherbird.

I can think of a dozen things just in my company where it would be a benefit that memory bandwidth plays almost no part in it.

I think people take old rules of thumb and forget where they came from when applying haphazardly. Very little of the consumer market even the prosumer market where they might have use for 16c for a single workload is really memory bandwidth sensitive. My software engineers could use more cores always need more memory, but that's because they are generally working on 2 plus projects at once, have 3-4 other projects open for reference, have the project management tool open, could be doing a subversion check in or check out and a packager going. Something a little more core intensive like compiling is more of an after thought. But at the end of the day they are using between 14 and 20GB of memory and could use more cores to minimize compile time and keep the system responsive. None of that requires 4 channel memory.

Or me I currently do mostly VM testing work with my work Ryzen (and host a couple Linux VM's on my home Ryzen). I would love an extra 8c but I could probably manage on a single channel for all memory access.

That's where AMD and Intel both have great strength and AMD can maximise their consumer space. They have products that cover a wide spectrum. Someone has heavy IO or memory bandwidth demands. They have X299/Xeon-W and TR to fill those spaces. But by giving people a 16c on AM4 AMD gives customers who don't have super sensitive workloads and option for increased and class leading compute power without requiring an almost $1k increase in cost (not just CPU, but Mobo, memory, even power supply). It was what AMD intended to do when the launched Ryzen and them treading water with Zen+ shouldn't change that as their primary focus for Ryzen with Zen 2 and beyond.

 

[Mockingbird]

OKaaaaay I think we are talking the same thing. Why did you say the prices are too low. The price list you responded to had a 16c top clock "3900x" at $499 and you said it's too low. They should have between 3 and 6 SKU's that will fit between the 8c chips and the 16c chips (they might forgo a low power 8c chip if they do single chiplet imho due to the heavily reduced power requirements). Perfect for stair stepping between $300 and $500 in $25-$50 increments.

You have been making it sound like AMD is worried about competing against threadripper. I doubt that's a worry and in a way they might do better in terms of a total silicon cost (not counting the recoup on defective dies) by selling a 16c much cheaper then past 16c TR's. But if it's the lower cost CPU's you have an issue with I still have to ask why do you feel that way? I mean A.) AMD has to keep the value stuff going it's as important in the refreshes as the halo models without them they don't get any OEM penetration and mindshare is at this point as important as Market share. And B.) I am now lost on what you think AMD is competing against themselves on the low end. Just previous gen stuff? I doubt canabalizing previous gen parts plays much in their pricing. If anything they do what every company in every market cept Nvidia/Intel/Apple do and lower the MSRP and off vouchers to help people clear their shelves before new the new product launch. You just need to look at 1700x and 2700x and 1950 vs. 2950 pricing to see the extreme version of what AMD does.

AMD, like any other company, responds to competition (which, in this case, is obviously Intel).

$499 12C/24T Ryzen 9 3800 already has 50% more cores/threads than Core i9-9900K. (assuming similar single-thread performance)

Unless Intel launches a proper competitor to the Ryzen 9 3800, AMD has no incentive to lower the price.

___________________________________________________________________________________________________________

Now, let's say that Intel actually launches a new $499 12C/24T processor.

That would give incentive for AMD to launch $499 16C/32T Ryzen 9 3900 to undercut Intel's pricing.

 

[Mopetar]

That's very hard to believe considering that AMD just launched the Ryzen Mobile 3000 series and desktop APUs use the same die.

Futhermore, for many users, integrated is enough. Without desktop APUs, AMD would be conceding all those users to Intel.

The Ryzen 3000 mobile chips are all on 12 nm, but I'm talking about their eventual 7 nm APUs. Maybe we don't get those until much later depending on wafer availability or when they can move to DDR5 since I would imagine making the graphics that much beefier becomes pointless if the memory throughput has become the main bottleneck.

You also have to consider that with the amended WSA, AMD may be simply viewing the 12 nm products at GF as the cost of being able to build more high-margin 7nm parts. In that case, they might not care as much about margins on the APUs. It also depends what they price R3 chips at or if they have a very inexpensive, but serviceable Navi card this year. At a certain price point, you'll just skip the APU (whether Intel or AMD offers the best option doesn't matter) because the cost of a better dedicated CPU and GPU isn't that much more.

Sure, not everyone wants or needs that, but AMD knows what it's like living in the bottom part of the market trying to scrape by. They might not mind letting Intel have it if it means that they can focus on all of the high margin segments.

For example, AMD's new 8C/16T supposedly can match Intel's $500 Core i9-9900K, yet AMD cut the price from ~$300 to ~$200

AMD already dominate the ~$200 and under market. It's the $300+ market that Intel dominate and AMD wants to cut into.

It doesn't matter that AMD cuts the price, because you have to look at their cost with Zen 2, not what things currently cost. If they can get 66% functional per wafer, that means that the cost of each chiplet only comes out to around $20 (since TSMC 7 nm is said to be around $10,000 wafer)

The $300+ market is ripe for the taking with any 12 or 16 core chips that they put out later. Those only take an extra ~$20 chiplet to make, so the largest cost for AMD to make that hypothetical R7 or R9 is the opportunity cost of not using that chiplet for a hypothetical R3 or R5.

Unless Intel launches a proper competitor to the Ryzen 9 3800, AMD has no incentive to lower the price.

Companies will adjust prices to maximize profit, even if there is no other competition. Sometimes you're better off selling at a lower price if it means that you have more sales overall. It really comes down to how supply constrained AMD is though since you can't sell CPUs you don't have.

 

[Mockingbird]

It doesn't matter that AMD cuts the price, because you have to look at their cost with Zen 2, not what things currently cost. If they can get 66% functional per wafer, that means that the cost of each chiplet only comes out to around $20 (since TSMC 7 nm is said to be around $10,000 wafer)

The $300+ market is ripe for the taking with any 12 or 16 core chips that they put out later. Those only take an extra ~$20 chiplet to make, so the largest cost for AMD to make that hypothetical R7 or R9 is the opportunity cost of not using that chiplet for a hypothetical R3 or R5.

What you (and other people on this forum) are doing wrong is that you are pricing things from the perspective of the consumer. (What I called "wishful thinking".)

e.g. The prices will be as such because, I, the consumer want them to be.

When you need to do (if you want realistic prices) is to price things from the perspective of the producer.

...and from the perspective of the producer, those prices make no sense.

Prices are dictate by market conditions, NOT what the consumers want.

 

[somethingclever]

Since AMD is paying a fixed cost to TSMC for each wafer, AMD is just going to order enough wafers to meet demands.

There's not going a be huge stacks of wafers (that AMD can't sell) sitting somewhere (which would have forced AMD to lower the prices).

Wafers for complex designs take months to go through the process, so if anything AMD is going to order more wafers than it estimates demand to be. It's worse off to underestimate than over estimate after all if production really is as cheap as you insist. Look at intel now struggling to meat 14nm demand.

So AMD may very well end up with wafers it has trouble selling. Thats probably also why there are still 1700s being sold for a third of msrp.

Even if Intel has a surplus, it doesn't make sense to cut production output because the cost to produce is so low.

So instead, Intel would lower the prices of each processors to move more unsold units

I don't remember a single instance where intel lowered prices much less to move unsold products.

AMD not having foundries does not exclude them from using the same strategy either.

 

[PotatoWithEarsOnSide]

You're guilty of thinking from the perspective of a market leader.

 

[somethingclever]

You're guilty of thinking from the perspective of a market leader.

What you (and other people on this forum) are doing wrong is that you are pricing things from the perspective of the consumer. (What I called "wishful thinking".)

Both are happening in this thread...
For the high end sure AMD might not have as much incentive to lower prices, but I and a lot of others think that isn't true for the low end.

We can argue all day about what the exact prices will be but in the end I'm betting they'll be lower than current ones.