Speculation: Ryzen 3000 series

[Mockingbird]

But AMD is also trying to take a larger piece of the market, which means expanding their own production volume. In order to entice new customers to purchase AMD CPUs, they naturally have to reduce their price.

There might be a particular stable price at which AMD can sell some number of CPUs, but if they want to sell twice or three times that number of CPUs then the only way to do so is at a lower price.

AMDs new chiplet strategy is going to allow them to produce a greater number of CPUs and to be more nimble in adapting to small changes and shifts in the market since the same chiplet can just as easily be used to make more server chips if demand is strong there, but shifted back to desktop parts if the server market weakens.

Further, the market rate for some product doesn't mean that AMD is necessarily seeing additional profit. During the mining boom, it was the retailers and people flipping cards who were absorbing a lot of the additional profit. Both AMD and NVidia may have seen additional sales, but they were not receiving all of the benefits of those additional sales.

Let's say that AMD's $200 processor costs $50 to make. (not including development cost)

Now, AMD cut the price in half to $100.

Are you telling me that AMD is going to sell at least 3 times as many processors at $100 as it would have at $200?

 

[somethingclever]

Intel's SSDs have just been flying off the shelves these days! /sarcasm

They might not be popular on the consumer market, but they've been doing quite well in Enterprise. 3D xpoint ring any bells?

 

[Mopetar]

Let's say that AMD's $200 processor costs $50 to make. (not including development cost)

Now, AMD cut the price in half to $100.

Are you telling more that AMD is going to sell at least 3 time as many processors at $100 as it would have at $200?

As long as they can supply three times as many processors, then yes.

Suppose it costs Intel $100 to make a 9900K. I don't know how close this is to accurate, but it doesn't matter because you can just adjust the other numbers. If they sold this processor at $233, do you believe that they wouldn't sell at least three times as many?

The processor that AMD demonstrated at CES is their mid-range part. You act as though simply because AMD wants to sell something like that for $200, they're behaving foolishly. They're still going to have a ~$500 CPU, only it's going to way more cores.

You're going to get a lot of people who had some kind of Intel quad-core, maybe even a more recent one to jump ship for an 8C/16T CPU at $200. If you could make enough of them to get away with charging $100, everyone and their grandmother would buy your product.

 

[OTG]

Intel's SSDs have just been flying off the shelves these days! /sarcasm

I DID just buy a 660p for my Steam library!

 

[maddie]

As long as they can supply three times as many processors, then yes.

Suppose it costs Intel $100 to make a 9900K. I don't know how close this is to accurate, but it doesn't matter because you can just adjust the other numbers. If they sold this processor at $233, do you believe that they wouldn't sell at least three times as many?

The processor that AMD demonstrated at CES is their mid-range part. You act as though simply because AMD wants to sell something like that for $200, they're behaving foolishly. They're still going to have a ~$500 CPU, only it's going to way more cores.

You're going to get a lot of people who had some kind of Intel quad-core, maybe even a more recent one to jump ship for an 8C/16T CPU at $200. If you could make enough of them to get away with charging $100, everyone and their grandmother would buy your product.

You act as though simply because AMD wants to sell something like that for $200, they're behaving foolishly. They're still going to have a ~$500 CPU, only it's going to way more cores.

This is a vital statement.

AMD will have products spanning the price spectrum. Selling an 8C for $200 - $300 does not prevent someone from buying a more costly model.

Want to spend $500, then choose this 12C or 16C model. Have a bigger budget? Then what about this TR2 24C or 32C. You get better mem & I/O for a higher MB price.

 

[Mockingbird]

As long as they can supply three times as many processors, then yes.

Suppose it costs Intel $100 to make a 9900K. I don't know how close this is to accurate, but it doesn't matter because you can just adjust the other numbers. If they sold this processor at $233, do you believe that they wouldn't sell at least three times as many?

The processor that AMD demonstrated at CES is their mid-range part. You act as though simply because AMD wants to sell something like that for $200, they're behaving foolishly. They're still going to have a ~$500 CPU, only it's going to way more cores.

You're going to get a lot of people who had some kind of Intel quad-core, maybe even a more recent one to jump ship for an 8C/16T CPU at $200. If you could make enough of them to get away with charging $100, everyone and their grandmother would buy your product.

So you really think if AMD cut its profit margin to 1/3, AMD would at least triple its sales?

If you were in the boardroom, you probably get laughed out of the room.

 

[Mockingbird]

You act as though simply because AMD wants to sell something like that for $200, they're behaving foolishly. They're still going to have a ~$500 CPU, only it's going to way more cores.

This is a vital statement.

AMD will have products spanning the price spectrum. Selling an 8C for $200 - $300 does not prevent someone from buying a more costly model.

Want to spend $500, then choose this 12C or 16C model. Have a bigger budget? Then what about this TR2 24C or 32C. You get better mem & I/O for a higher MB price.

Let's ask more general questions.

Do you really think that if AMD cut its profit margin to 1/2, AMD would at least double its sales?

Do you really think that if AMD cut its profit margin to 1/3, AMD would at least triple its sales?

Do you really think that if AMD cut its profit margin to 1/4, AMD would at least quadruple its sales?

Do you really think that if AMD cut its profit margin to 1/100, AMD would at least centuple it sales?

et cetera

 

[IEC]

People seem to confuse a desire for profits as being diametrically opposed to gaining marketshare.

They don't have to be opposed. The pitch for marketshare that AMD has been providing since Zen 1 is more cores and bang for the buck at each price point versus their competitor. That convinced enough users to adopt the AM4 (and sTR4) platforms. Despite this, they raised their ASPs and increased their margins - and this is reflected in their improving financials. Pricing is an optimization problem, and AMD will price their parts accordingly.

As a business heavily-dependent on R&D, AMD *needs* large profits to sustain the kind of R&D efforts that will lead to success 3, 5, or even 10 years down the road. So even if I fully expect AMD to beat their competitor at nearly every price point with a competitive Zen 2 product, it will be priced in such a way as to ensure adequate profits for future R&D efforts.

 

[Zucker2k]

Who is AMD looking to disrupt? Themselves?

If AMD have a comparable product to Intel, and AMD already sell it for significantly less than Intel, then what is gained by lowering it more? Why aren't they already dominating the CPU commodity market with superior price/performance being recognized?

Because it's more likely that AMD has a perception/marketing/education issue, not a pricing issue.

Simply dropping pricing to deal with perception issue, will leave barrels of money the table, and reinforce the perception of AMD as the cheap option, not the premium option.

AMD 6 and 8 core parts, will not decrease in price, when they introduce an even better Zen core, because Lisa Su is a lot smarter than wishful thinkers of forums. She will want premium margins for her premium products.

This. I also think AMD will be looking at two critical things:
1. If they go too low on pricing, they'll force intel to cut prices, which means they'll have to go even lower. Price-war. Not healthy at all for margins, and who knows where that'll end.
2. A cheap 16 core monster chip in the desktop segment is certainly going to eat well into TR2 sales. Again, not good for margins.
What this means is that it's in AMD's interest to not poke Intel into making price cuts. Thanks to Intel's high prices and AMD's node advantage, there's plenty of wiggle room to deservedly bump prices with their much improved chips. I believe the 12 core R9 is going to be the halo chip on desktop for AMD until Intel can release something more dominant. The 8 core R7 will target the HT-less 8c/6c + HT i7s, the 6c R5 will target the i5s, and the 4c R3 will target i3s. With these, AMD can comfortably bump up prices to present Intel levels in each segment.

 

[somethingclever]

Let me address the developmental cost aspect of zen2. And potentially another reason why core counts have doubled/gone up.

I doubt anyone would disagree that developing for 7nm is more expensive than 14nm. Production cost per die itself shouldn't be that much greater because of the increased density.

So far people have been saying that AMD has addressed higher cost with increased yield (using the chiplet + IO die strategy) which is quite reasonable and very convincing.

I think another way AMD has negated higher costs is doubling volume. Rome and Ryzen both have double the number of dies, which naturally eliminates development costs up to twice that of zen 1 on 14nm.

So, cost per die in the end will be around the same, if not lower (with better yields and higher volumes). That is why AMD may be able to move 6/8 core single die units into the R3/5 segment (no matter what price they choose in the end). The smaller die area and better yields eliminate the 4 core bin, while increased volume to produce chiplets for Rome and high end Ryzen keeps costs the same.

So why didn't AMD perform that shift when they eliminated the 4 core bin with ryzen 2k? I think it's either because they couldn't increase volume enough to offset the development cost, or they were saving that shift for zen2.

 

[maddie]

Let's ask a more general question.

Do you really think that if AMD cut its profit margin to 1/2, AMD would at least double its sales?

Do you really think that if AMD cut its profit margin to 1/3, AMD would at least triple its sales?

Do you really think that if AMD cut its profit margin to 1/4, AMD would at least quadruple its sales?

Do you really think that if AMD cut its profit margin to 1/100, AMD would at least centuple it sales?

et cetera

This is the thing, where are you reading that AMD will cut margins? This is a classic example of a strawman argument.

I've been arguing that their costs might actually drop. 1 tiny (smaller than Apple design), high yielding, admirably "die harvestable" 7nm chiplet with the development cost amortized over their entire production of processors is not going to be this huge cost item as you seem to think. In fact the more they produce, the lower this design segment of the cost becomes. The lower limit is TSMC production cost.

You keep harping about cutting margins and fabricating scenarios that no one is really advocating. You are stuck with the idea that cheaper automatically means reduced margins.

Here's mine.
What if reduced prices come with identical margins?

 

[Mockingbird]

This is the thing, where are you reading that AMD will cut margins? This is a classic example of a strawman argument.

I've been arguing that their costs might actually drop. 1 tiny (smaller than Apple design), high yielding, admirably "die harvestable" 7nm chiplet with the development cost amortized over their entire production of processors is not going to be this huge cost item as you seem to think. In fact the more they produce, the lower this design segment of the cost becomes. The lower limit is TSMC production cost.

You keep harping about cutting margins and fabricating scenarios that no one is really advocating. You are stuck with the idea that cheaper automatically means reduced margins.

Here's mine.
What if reduced prices come with identical margins?

Don't change the topic.

You have been saying that any drop in profit margin can be make up for in increased sales, have you not?

 

[somethingclever]

I've been arguing that their costs might actually drop. 1 tiny (smaller than Apple design), high yielding, admirably "die harvestable" 7nm chiplet with the development cost amortized over their entire production of processors is not going to be this huge cost item as you seem to think. In fact the more they produce, the lower this design segment of the cost becomes. The lower limit is TSMC production cost.

I don't necessarily think that cost has dropped, but AMD has found a way to keep them about the same while increasing the number of cores.

 

[somethingclever]

You have been saying that any drop in profit margin can be make up for in increased sales, have you not?

That is true, so I don't see your point. Obviously cutting prices in half wouldn't double sales, but reducing prices will have a positive influence on how many units you will sell. The question is how low can prices be before increased sales stop making up for it?

 

[Mockingbird]

That is true, so I don't see your point. Obviously cutting prices in half wouldn't double sales, but reducing prices will have a positive influence on how many units you will sell. The question is how low can prices be before increased sales stop making up for it?

Start by setting higher prices (MSRP), then put the products on sale to gauge consumers' reaction.

AMD's two best selling processors, Ryzen 5 2600 and Ryzen 7 2700X seem to fluctuating around $160 and $300, respectively.

Most likely, those are the optimal prices.

 

[OTG]

Don't change the topic.

You have been saying that any drop in profit margin can be make up for in increased sales, have you not?

"BUT YOU SAID LOWER PRICE MEANS MORE MONEY, WHAT IF AMD SELLS CPUS FOR $.001 WILL THEY SELL ENOUGH TO MAKE MONEY?
Why are you trying to make this argument?

Everybody with two brain cells to rub together knows that higher sales can make up for lower price to a point. Below that point, lower price won't bring in enough sales to compensate, and/or you're losing money on each sale.​

​

We just happen to disagree where that point is.​

Not that any of us has more than half a clue about the actual numbers.​

​

​

Start by setting higher prices (MSRP), then put the products on sale to gauge consumers reaction.

AMD's two best selling processors, Ryzen 5 2600 and Ryzen 7 2700X seem to fluctuating around $160 and $300, respectively.

Most likely, those are the optimal prices.

Which is why the midrange chips will continue to be priced in the $150-$300 range.
Even if 'midrange' has 8-12 cores.

​

 

[Mockingbird]

"BUT YOU SAID LOWER PRICE MEANS MORE MONEY, WHAT IF AMD SELLS CPUS FOR $.001 WILL THEY SELL ENOUGH TO MAKE MONEY?
Why are you trying to make this argument?

Everybody with two brain cells to rub together knows that higher sales can make up for lower price to a point. Below that point, lower price won't bring in enough sales to compensate, and/or you're losing money on each sale.​

​

We just happen to disagree where that point is.​

Not that any of us has more than half a clue about the actual numbers.​

​

​

you misquoted.

Which is why the midrange chips will continue to be priced in the $150-$300 range.
Even if 'midrange' has 8-12 cores.

Why would AMD add more cores when its products are already selling well at the current prices?

 

[somethingclever]

Why would AMD add more cores when its current products are already selling well at the current prices

Look at it from this perspective. They aren't deliberately "adding more cores". They literally find themselves having more cores at the same cost. They keep roughly the same margins of zen1 but because of 7nm improvements, they've found that they can eliminate the 4 core chiplet bin because it's so rare.

To increase competitiveness and market share, they decide not to artificially segment by disabling two working cores on chiplets just for a 4 core R3.

 

[maddie]

I don't necessarily think that cost has dropped, but AMD has found a way to keep them about the same while increasing the number of cores.

In their accounting to assign costs to products, don't you think that a much smaller # of sales of Zen1 & Zen+ products affected the internal accounting cost of production?

I sometimes wonder if some of these debates are caused by only seeing costs as fab and assembly.

We have basically the same company costs (pretty much fixed in the short term) now being spread over what is expected to be a much larger production and sales volume. The sales team costs can now drop per CPU sale. This applies to all departments. I really think that for internal accounts to assign costs to any product, AMD will be seeing lower costs per unit. If sales quadruple and their fixed costs increase by 50% then fixed cost drops to 150/400 = 37.5% of the previous gen. The next big point is that the more you sell the cheaper each unit becomes to produce. What's the term? It becomes a virtuous circle leading to even greater gains.

 

[PotatoWithEarsOnSide]

As long as they can supply three times as many processors, then yes.

Suppose it costs Intel $100 to make a 9900K. I don't know how close this is to accurate, but it doesn't matter because you can just adjust the other numbers. If they sold this processor at $233, do you believe that they wouldn't sell at least three times as many?

The processor that AMD demonstrated at CES is their mid-range part. You act as though simply because AMD wants to sell something like that for $200, they're behaving foolishly. They're still going to have a ~$500 CPU, only it's going to way more cores.

You're going to get a lot of people who had some kind of Intel quad-core, maybe even a more recent one to jump ship for an 8C/16T CPU at $200. If you could make enough of them to get away with charging $100, everyone and their grandmother would buy your product.

This is pretty much it.
What Mockingbird is missing with his 475th reference to sub $200 CPUs is that AMD's development costs are fixed, so if they insist on selling fewer CPUs then their cost per unit related to that overhead is increased, much like how economies of scale allow producing vastly more 7nm tiny chiplets for a lower per unit cost; it makes absolutely no sense to restrict your market share when you've invested so heavily in high volume sales, hence lower prices being a necessity. They aren't competing with any CPUs bought in the last 2 years at all, since their target is clearly to win over those long term Intel owners that have been put off buying marginal increases in performance for ever increasing prices. You do that by offering a step change in approach; cheap yet significantly improved. That's how you get them to your platform. It's a long term strategy that requires short term reductions in margins; win on all metrics, rendering all arguments in favour the competitor entirely invalid. That's the point...mindshare.

 

[maddie]

Don't change the topic.

You have been saying that any drop in profit margin can be make up for in increased sales, have you not?

Have I really said what you wrote? I can't remember stating this. Please show and allow me to respond properly.

 

[Mopetar]

So you really think if AMD cut its profit margin to 1/3, AMD would at least triple its sales?

If you were in the boardroom, you probably get laughed out of the room.

If something costs $1 to make and you can sell 100 of it at $4 dollars, you make $300 of profit. If instead you lower your price to $2 and can sell 1000 of it instead, you make over three times as much profit.

The only circumstances under which you don't sell at $2 is if you can only actually make 100 of that thing. AMD's chiplet strategy is going to allow them to make many times the processors that they could before. This should be obvious.

 

[PotatoWithEarsOnSide]

Another way to look at is in relation to AMD's WSA with GF.
They're committed to buying however many wafers, and at whatever cost. Now though, they have twice (at least) as many dies with Ryzen 3000 for the price of that WSA, so they're already committed to selling more CPUs by design choice. The economics just of the IO die tell you that.
Sure, 7nm dies cost more than 12/14nm, but they get 3x as many chiplets per wafer, so on a per unit basis there is likely parity. But then you consider that R&D overhead, and that it is better to spread over increased volume...

It's not difficult to see that AMD have to increase their sales volume, and to do that they have win on all metrics.

 

[OTG]

Has there been any info one way or another if the two-chiplet designs need to have the same # of active cores?
Zen 1 had to be symmetrical, but if Zen 2 doesn't, allowing even higher effective yields could only help.
Instead of throwing out a chiplet with 5 cores, maybe you stick it in with 7-core and boom, R7.

 

[maddie]

If something costs $1 to make and you can sell 100 of it at $4 dollars, you make $300 of profit. If instead you lower your price to $2 and can sell 1000 of it instead, you make over three times as much profit.

The only circumstances under which you don't sell at $2 is if you can only actually make 100 of that thing. AMD's chiplet strategy is going to allow them to make many times the processors that they could before. This should be obvious.

I have to indulge in some ribbing here.

But, but, but, margins.