Question Rethinking AMD consumer CPU lineup in the situation of chiplet scarcity

[Kocicak]

If we look at the higher end of the consumer CPU lineup by AMD (beginning with Ryzen 3600), and the fact that AMD is unable to deliver a highly clocked 6+6 core CPU and effectively offers no high core count CPU at this point, and the fact that the high quality chiplets are better used in server CPUs, I believe it is time to radically rethink the lineup of consumer CPUs.

I believe the higher end of the consumer CPU lineup can comprise of just 4-5 processors.

1) 6C 3.9/4.2 65W - the same as current 3600 - price 200 USD
2) 6+6C 3.6/4.2 95W - made from two chiplets which are used in 6C - price 300 USD

These two CPUs can cover 75-90% of the market demand, they are made from low quality chiplets which can have up to two nonfunctional cores. These chiplets are unusable in server processors anyway. They are essentially a waste product of server CPU production.

The market demand can be nearly all covered by chiplets that are not needed for server CPU production !!!

3) 8C 4.0/4.7 95W - very high quality chiplet which could be used in server CPU - price 400-450 USD
4) 8+8C 3.8/4.7 135W - made from two chiplets which are used in 8C - price 750-800 USD

5) 8+8C-super ?/? ?W - made from "miraculous chiplets", intended for those who want something extraordinary and special - price 1000 - 1200 USD. Existence of this product depends on the existence and quantity of those "miraculous chiplets".

What do you think?

 

[Kocicak]

This is why you probably flip burgers or clean the toilets.

I can assure you that I am flippin burgers like no other and clean toilets so well that YOU could eat from them.

 

[Kenmitch]

I think AMD should just stick with what they have announced.

I'd imagine they're better off just sticking to their game plan. I doubt going into panic mode like Intel would be of any benefit to the company.

I'd imagine the initial runs of offerings were dictated by some silly bean counters whom misjudged the markets wants and needs. They just missed on their sales predictions for the 3900x is all.

The cores vs clocks wars is probably what made the bean counters and AMD as a whole hesitant on offering a lower clocked higher core count offering.

 

[maddie]

I'd imagine they're better off just sticking to their game plan. I doubt going into panic mode like Intel would be of any benefit to the company.

I'd imagine the initial runs of offerings were dictated by some silly bean counters whom misjudged the markets wants and needs. They just missed on their sales predictions for the 3900x is all.

The cores vs clocks wars is probably what made the bean counters and AMD as a whole hesitant on offering a lower clocked higher core count offering.

I actually agree a bit with the OP.

One of the problems with 'sticking with the game plan' is that AMD appears to be on a 1 yr release cycle. Keeping to the plan works best the longer the timeframe as any delay is a smaller % of the total time, but Ryzen 4xxx is coming in roughly 9 months if all goes well. These releases in Nov/Dec will make it even worse.

On the other hand, selling a 12C low priced CPU will probably alienate a lot who bought the 3900X.

 

[UsandThem]

On the other hand, selling a 12C low priced CPU will probably alienate a lot who bought the 3900X.

When it comes to tech, usually what you buy this year will be both faster and cheaper the next year.

Take the 1700X ($399) and 1800X ($499) when they were launched. A year later the 2700 and 2700X hit, and was was faster and cheaper. It's just the name of the game when it comes to tech.

 

[Kocicak]

On the other hand, selling a 12C low priced CPU will probably alienate a lot who bought the 3900X.

Why? 3900X is a 3.8/4.6 CPU for $500, a premium product, proposed "double 3600" for $300 would be 3.6/4.2 - it has massivelly worse clocks! It a much worse product for less money, it is ok.

I believe that 3900X runs at higher frequency than 3.8 GHz while still in its TDP limit. The "double 3600" would probably really need to run at 3.6 to meet any reasonable TDP.

 

[Kedas]

Is there anyone buying a 3900X when the 3950X is out?
Almost everyone will take the 3950X like we saw with threadripper.
When you are in the high end game then many just take the top part.

So they shouldn't have a lot of 3900X in the field when 3950X launches.
And using the dies for EPYC gives a much bigger profit.

 

[Kenmitch]

Why? 3900X is a 3.8/4.6 CPU for $500, a premium product, proposed "double 3600" for $300 would be 3.6/4.2 - it has massivelly worse clocks! It a much worse product for less money, it is ok.

I believe that 3900X runs at higher frequency than 3.8 GHz while still in its TDP limit. The "double 3600" would probably really need to run at 3.6 to meet any reasonable TDP.

They have a custom order division at AMD. I'd imagine they'd be willing to make them. You'd probably need to order a minimum amount tho. Maybe only 10k, 20k ???

 

[Insert_Nickname]

Well, yeah, but I think that they should introduce an 8C/16T CPU, made from 4C+4C defective dies, with half-size L3 enabled in each die as well, as a dumping-ground for "bad chiplets". Maybe we'll see a 3700 soon enough.

Maybe a 2 core per CCX x4 with fully enabled L3? Then you'd have an 8C/16T CPU with 64MB(!) L3. Such could actually be a fun chip to play with.

But we're not going to get that I think. Except maybe, just maybe, as a third gen TR.

 

[maddie]

When it comes to tech, usually what you buy this year will be both faster and cheaper the next year.

Take the 1700X ($399) and 1800X ($499) when they were launched. A year later the 2700 and 2700X hit, and was was faster and cheaper. It's just the name of the game when it comes to tech.

We're talking here about a 4 month timeframe not an annual refresh affair.

Why? 3900X is a 3.8/4.6 CPU for $500, a premium product, proposed "double 3600" for $300 would be 3.6/4.2 - it has massivelly worse clocks! It a much worse product for less money, it is ok.

I believe that 3900X runs at higher frequency than 3.8 GHz while still in its TDP limit. The "double 3600" would probably really need to run at 3.6 to meet any reasonable TDP.

Massively worse clocks?

 

[Thunder 57]

Where to start here... This won't come off as nice but you just aren't convincing me as to why AMD should do this.

It needs much better chiplets than which go to 3600 and 3600X to clock higher and meet the TDP. And if they are efficient enough, it is actually more probable they are good overall and are fully functioning 8C usable in server CPUs.

You don't know that. You even admit this later. A 3900X needs two CCD's with at least six good cores, not 8.

We may never hear that, because all this information regarding production and sale numbers are trade secret. We do not know how is the binning process set up, what bins there are and how they can keep up with the demand.

But you just said that a 3900X needs good 8C chiplets! You don't know that.

What we know is that AMD fails to supply enough of high core count CPUs and pospones the release of already planned CPUs, while they apparently have abundance of lower quality chiplets which are being used in Ryzen 3600.

The solution is obvious and easy, right??? Just populate the damn free space in 3600 with a second chiplet and you are done!

Let's consider that. You say to do it and price it at $300. That is beyond ridiculous. I said there might be a spot for a 3900 at $400, but even that is too low. It would kill the 3800X. You would have to cut its price. So a 3900 would have to be $450. There it could actually make sense. I'll give you that the 3800X is in a bit of a weird spot as is, but that's how AMD went and they really can't deviate from that this early.

People with moderate heavilly multithreaded load needs would be satisfied enough with the "double 3600" which can be produced and sold cheaply while not consuming any good chiplets destined for server CPUs and people with higher such needs would buy the 3950X with the added benefit of higher boost frequency. Those wanting high boost frequency without any special heavilly multithreaded load needs would buy the 8C 3800X or how would be that called.

I am really sorry, but I fail to see any lowering of margins here anywhere.

You go on to propose a two chiplet CPU for less money than a single chiplet one. How does that not kill your margins, not to mention sales of the 3700X on up? Most people aren't going to care about a few hundred MHz. No one will pay more for less cores.

The only trouble for AMD would be in situation, that what they curently put in 3700X is so bad, that it cannot be used even in lower chiplet count server CPUs. Shift in sales from the $330 3700X to the $300 "double 3600" would mean, that they get less for CPUs in which they put two server CPU unfit chiplets than what they used to get for 3700X in which they put just one server CPU unfit chiplet.

When you look at the frequency and TDP of 3700X, if looks quite good to me, and I do not believe the chiplet in it cannot be used in any server CPUs.

The limited supply of 3900X indicates that what they put in it can be used in server CPUs and not selling it would just free up more chiplets for server CPUs. The demand for server CPUs fluctates too you know. A new security flaw of Intel CPUs comes and the demand just increases unexpectedly. A large portion of the demand for 3900X could be covered with the proposed "double 3600".

I think you try to explain it here, but I honestly just don't follow or see it. Lisa Su helped bring AMD back from the brink of bankruptcy. I'm pretty sure she can trust that she has people who know how to competently price these. Sometimes I question AMD's marketing, but everywhere else they seem to be spot on.

Hope I didn't come off as harsh, just trying to have a friendly back and forth. Let me ask you this, if AMD sold a $300 "double 3600", who would buy a 3700X, 3800X, or 3900X? I think if AMD did what you suggest, the market for the 8C CPU's dies off and the 3900X would have to sell for no more than $400, maybe even less.

 

[CHADBOGA]

Is there anyone buying a 3900X when the 3950X is out?

A lot will depend on pricing.

With what I believe the price difference will be, I'd happily go for the 12 core.

 

[DrMrLordX]

Is there anyone buying a 3900X when the 3950X is out?

Price is a factor. A 3900x can be had (retail) for the same price AMD charged for the 1800x over two years ago. That was one of the factors that made me not wait for the 3950X (which I easily could have done). Now that I've seen how the 3900X performs, I would be really worried about cooling a 3950X. Keeping those hotspot temps down will be a challenge! My 3900x is already toasty enough as it is.

 

[Markfw]

Price is a factor. A 3900x can be had (retail) for the same price AMD charged for the 1800x over two years ago. That was one of the factors that made me not wait for the 3950X (which I easily could have done). Now that I've seen how the 3900X performs, I would be really worried about cooling a 3950X. Keeping those hotspot temps down will be a challenge! My 3900x is already toasty enough as it is.

I guess I am not as ambitious as some. I am at 4100 mhz and 1.1 vcore for 68c full load. My empty case has a 280mm AIO and an x570, this is only a 240 MM AIO. I think I will be fine when I get my 3950x.

 

[DrMrLordX]

I guess I am not as ambitious as some. I am at 4100 mhz and 1.1 vcore for 68c full load. My empty case has a 280mm AIO and an x570, this is only a 240 MM AIO. I think I will be fine when I get my 3950x.

It all depends on workload, really. If you're folding then that isn't a terrible temp per se. I could see a 3950X getting worse though. Might have to back off on the clockspeed.

 

[beginner99]

Is there anyone buying a 3900X when the 3950X is out?

I will probably if it ever becomes available and the actual pricing is as it was announced. It's still $250 difference and reasliticaly I don't even need 12 cores. But from 8->12 it's a 50% increase. From 12->16 only 33% increase in cores.

The sorts of chips used for server CPUs versus consumer CPUs are pretty much on the opposite ends of the spectrum in terms of silicon characteristics.

Exactly. And that is why I don't think Epyc demand is the cause for the shortages.

I personally suspect the main issue is the target frequencies of the 12/16 core. Due to much higher demand (for 12-core) than expected, AMD doesn't have enough chiplets with a core that hits the needed max turbo speeds. It's basically a marketing blunder. Just putting 100mhz lower on the box would probably have solved the whole turbo-debate and the shortage.

And since they now can approximate demand, AMD delayed the 3950x as to not have the same problem as with the 3900x, eg. weeks-months of no availability. Better to delay the releases.

 

[marcUK2]

I have waited 2 months for a 3900x but just purchased a 3700x after suffering boredom attack....

I dont believe there is a shortage of working 6 core chips, there is no 3600x shortages....but AMD screwed themselves a bit by going after the gaming crowd with high boost speeds.....I would be more than happy with a vanilla 3900 or 3950 (non X) that has an all core clock of about 4ghz stock and boosted to 4.3. Surely people who buy huge multicore chips actually want them because they use software that makes use of the cores......

 

[fleshconsumed]

Yeah, I would have loved 3950 without the max single core boost of 3950x so long as it had the same all core boost. At this point it just might make sense waiting for 16 core Zen3 if it's less than a year away.

 

[Kocicak]

You don't know that.

You are correct, I speculate a lot. For example, my plan would increase consumption from the 3600 chiplet bin quite significantly. What if there is not enough of these chiplets to do it?

On the other hand, you know very little about the actual situation in AMD too. And IT IS POSSIBLE that changing the CPU lineup may be beneficial now, both financially and in production organisation. Only AMD knows. What we know is that there are some troubles executing the original plan and that a "double" 3600 should be easy and cheap to produce, its production does not need any HQ 8C chiplets needed for server CPU production and consumers would probably want it.

Let me ask you this, if AMD sold a $300 "double 3600", who would buy a 3700X, 3800X, or 3900X?

3700X and 3900X do not exist in my plan.

3600 and "double 3600" are made from slow 6C chiplets, unusable in servers, possibly able to cover a great deal of market demand. These processors are for people with "general computing needs".

3800X and 3950X are made from HQ 8C chiplets and have 400-500 MHz higher boost frequency than 3600 and "double 3600". 500 MHz is a significant difference. These are special processors for people who need high performance for lightly threaded load resp. high performance for both lightly and heavilly threaded loads.

Here is the overview:



I do not believe that you need more than 4 CPUs in the upper part of the consumer CPUs lineup. And the super cherrypicked limited edition for one grand is always possible too.

 

[guachi]

If AMD ever gets more 6-core chiplets than they think they can sell then I'd love to see a 12-core part made from them. I'd probably take that over the 3900X.

 

[Markfw]

If AMD ever gets more 6-core chiplets than they think they can sell then I'd love to see a 12-core part made from them. I'd probably take that over the 3900X.

Thats what the 3900x is made from, 2 - 6 core chiplets, but with a higher bin than the 3600x

 

[.vodka]

To be honest, the 3800x is kind of pointless at $400 over the $330 3700x. Bad deal for that price. Yes, I am aware of this:

This would make more sense IMO:

$200 - 3600
$250 - 3600x
$300-320 - 3700x
$400-420 - 3900 (2x3600 in cores and clockspeeds)
$500-520 - 3900x
$750 - 3950x

Such a 3900 would suit people who are after the 3900x for its MT prowess. Less pressure on the 3900x.

Chiplets fit for 3800x duty could proooooobably be turned into 3900x class if cherry picking the best six cores. If not, more stock for the rest of the lineup save for ⩾ 3900x.

This way the shortage could be easier to manage until TSMC further improves yields and produces more higher quality dies that are fit for 3900x and 3950x specs (despite their current capacity constraints for the 7nm node). Of course it doesn't help that EPYC demands 8 chiplets per CPU, it's understandable that there's a shortage.

-----------------

I'd gladly take 12 overall fast cores that can't do much more than ~4.2-4.3GHz peak ST boost at the 3800x price point, maybe a bit higher accounting for the extra manufacturing cost of including the second CPU chiplet.

I don't care if the factory boost curve tends to be a flat line like the 3600's.

Let me try my luck with PBO and whatever cooling I can manage to extract as much ST/MT clockspeed out of these two sub par chiplets. Let my motherboard's overengineered VRM stretch its legs and break a sweat.

I don't care if I unleash the limits and MT power consumption at ~4-4.2GHz gets out of hand relative to the 3900x because of the two sub par chiplets. At that price it's worth it.

There's room for a lower clocked 3950x in that price structure, too. I wouldn't be surprised if these hypothetical 3900 and 3950 parts become available at some point.

 

[Topweasel]

Price is a factor. A 3900x can be had (retail) for the same price AMD charged for the 1800x over two years ago. That was one of the factors that made me not wait for the 3950X (which I easily could have done). Now that I've seen how the 3900X performs, I would be really worried about cooling a 3950X. Keeping those hotspot temps down will be a challenge! My 3900x is already toasty enough as it is.

Yeah that was kind of my philosophy. Its 12c, an increase in cores I was interested in and at a general price of premium consumer platform CPU pricing. This was also for a completely new build replacing a 4c8t system.

With the 3950x its hard to say. I might get it as a cheap upgrade for my 1700 (X370) system. But it would be next year at earliest. But I can wait out a possible 3950 (no X) or see what Zen 4 is lining up to look like.

 

[Kocicak]

I believe that the 3900 non X as originally planned is just a little bit slower and cheaper than the X version (boost slower by 200 MHz, cheaper by 50-75 USD?).

The 3600x2 compared to that would be slower, power hungrier, could be as "cheap" as I am suggesting. BTW I do not think that 300 USD is cheap. AMD would be actually making a nice profit on these considering how slow chiplets (and just 6C chiplets!) would be needed for them.

 

[Topweasel]

This would make more sense IMO:

$200 - 3600
$250 - 3600x
$300-320 - 3700x
$400-420 - 3900 (2x3600 in cores and clockspeeds)
$500-520 - 3900x
$750 - 3950x

Such a 3900 would suit people who are after the 3900x for its MT prowess. Less pressure on the 3900x.

Chiplets fit for 3800x duty could proooooobably be turned into 3900x class if cherry picking the best six cores. If not, more stock for the rest of the lineup save for ⩾ 3900x.

This way the shortage could be easier to manage until TSMC further improves yields and produces more higher quality dies that are fit for 3900x and 3950x specs (despite their current capacity constraints for the 7nm node). Of course it doesn't help that EPYC demands 8 chiplets per CPU, it's understandable that there's a shortage.

AMD wouldn't kill the 3800x pretty much ever. It's the playfield leveler. 95% of a 9900k at $100-$150 cheaper. It's also what they need to stand up to the 9900KS and that's one it can't win against. Stepping down to a 3700x only would make it worse.

 

[fleshconsumed]

AMD wouldn't kill the 3800x pretty much ever. It's the playfield leveler. 95% of a 9900k at $100-$150 cheaper. It's also what they need to stand up to the 9900KS and that's one it can't win against. Stepping down to a 3700x only would make it worse.

Technically 9900KF is true 3800X competitor since both lack iGPU and KF version is a bit cheaper. 9900KF regular price is only $70 more than 3800X, and 9900KF has frequently been going on sale for only $420 making price difference pretty much moot.

I get what you're saying about AMD having 3800X halo product, they got to have that for marketing purpose. Unfortunately that leaves those of us who need/want more than 8 cores in a bind.