News Can AMD hope to seriously challenge Intel's market dominance without their own Fabs?

InvalidError

Titan
Moderator
TSMC's lead times getting longer does not really affect AMD: AMD has a wafer purchase agreement with TSMC, which means that TSMC guarantees it will provide AMD with at least a certain amount of wafers worth of chips every however many days for however long those agreements last. AMD may also have a contract for optional wafer starts beyond that which TSMC would have to shift other products' delivery time to accommodate whenever AMD needs to use those options. The longer back-order list only affects extra production that AMD has no pre-existing contracts for.
 
TSMC's lead times getting longer does not really affect AMD: AMD has a wafer purchase agreement with TSMC, which means that TSMC guarantees it will provide AMD with at least a certain amount of wafers worth of chips every however many days for however long those agreements last. AMD may also have a contract for optional wafer starts beyond that which TSMC would have to shift other products' delivery time to accommodate whenever AMD needs to use those options. The longer back-order list only affects extra production that AMD has no pre-existing contracts for.

I've read that elsewhere too. But that really does very little to alter the underlying problem that TSMC itself only has finite manufacturing resources and if we're allowed to conjecture I'd be seriously doubtful they would hitch their wagon to one star only going forward. There has to be limits to what AMD can 'command' of TSMC and that article alerts us they are straining under the current production work load (to a degree at least).

What that tells me is AMD may be safe for now but beyond the current aggreements they have to be able to ship a LOT more processors each month to take even 10% more market share from Intel. It's just super hard to fathom anybody thinking AMD could take 10% market share from Intel three years ago, not even AMD. Hard to pound out an long-term agreement like that when nobody believes it so what's in it is something nobody knows. Add in that under this CEO, I'm pretty sure AMD is not interested in sitting still, not with two more spins of Zen already in the chute and they're talking of taking over the low-end desktop segment now with R5-3500(X). They want market share and that means they'll need more wafers & chiplets to do it and staying hitched to TSMC would have to limit them in that flexibility.
 

InvalidError

Titan
Moderator
There has to be limits to what AMD can 'command' of TSMC and that article alerts us they are straining under the current production work load (to a degree at least).
The limit is whatever is spelled out in AMD's contracts with TSMC. If AMD has agreed to purchase 80k 7nm wafers per month for the next two years with a 20k option, then AMD is on the hook for paying TSMC for a minimum of 80k wafers per month and TSMC is on the hook for being able to accommodate up to 100k wafers per month for AMD. Since the agreements are typically only about wafers, AMD can split its orders whichever way it wants between CPUs, GPUs and anything else it might want to make on 7nm, not too much of a worry about having too much of one thing there.

Also, TSMC isn't "hitching to a single star" as should be evident from part of the reason for TSMC's lengthening lead times is making Apple's A12 and A13 chips.

Who is screwed by the longer back-order time is companies like Nvidia who didn't have 7nm volume contracts yet.
 
....
Who is screwed by the longer back-order time is companies like Nvidia who didn't have 7nm volume contracts yet.

And us too, if we want 7nm hardware and can't get it because 80k wafers is no where enough to feed the market. Everybody talks about them a lot but AMD's been a very small player in both GPU and CPU market for a long time. I really wonder if they signed up long term agreements for enough wafers that will guarantee them to take back healthy market share from BOTH Intel and Nvidia. What if that means not 80K wafers, but 160K wafers...240K wafers (I really don't know, that's your example but I think you get the point).

Keep in mind I'm not trying to say they'll take like 90% of CPU and GPU and push them out. But considering the market share AMD had before even just a 10% share increase amounts to a HUGE number of chips. What if, after RX 5800/5900 (imaginary I know...isn't it??), Zen3, Zen4 come out AMD has the product that could take 50% share? What do they do to try and get production capability to do that?

Buy out TSMC?... who's got the bigger market cap. not likely, i bet. I see more likely Intel buy 'em out and get a piece of AMD.

There has to come a limit, and the only party sitting in the catbird seat is TSMC as I see it.
 
Last edited:

valeman2012

Distinguished
Apr 10, 2012
1,272
11
19,315
I think intels own fabs have held them back in some cases. The 14NM shortage was purely due their fabs not keeping up with demand, just like TSMC is having issues with now.
14nm+++++++++ going last long but they are able still maintain and somehow magically prevented 7nm from Ryzen from fully beating (in many games and many others benchmarks (excepts Cinebench) Intel 14nm processors. What kind of special design does Intel has?
 
What kind of special design does Intel has?
One that hits 5ghz?

Intel and AMD are very different. Intel fabs and designs their own CPUs. AMD designs their chips very differently and has other people fab them.

Really the fabs seem to be holding both back with production issues. Apples iPhone 10 and 11 are both based on TSMC 7nm, with the latter iPhone using EUV. I suspect Iphones are eating up some of TSMCs production resources. Intel has no such issus.
 
One that hits 5ghz?
.....
I suspect Iphones are eating up some of TSMCs production resources. Intel has no such issus.
AMD had a design that hit 5Ghz all-core too... for whatever good it did them. I'm reminded of an old axiom, changed a bit to fit the situation: it's not the gig's that matter, it's what ya do with them that counts. Just as AMD wasn't pumping out 5Ghz factory-overclocked excavators at launch it's taken a bunch of 'plusses' for Intel to get to there too.

I thought Intel DID have issues with production resources... didn't they outsource some of their 12nm chip production to TSMC? That must have cost a pretty penny redesigning those chips for TSMC's process, so it must have been a pretty desperate problem too!
 

InvalidError

Titan
Moderator
And us too, if we want 7nm hardware and can't get it because 80k wafers is no where enough to feed the market.
The 80k wafers/month figure was just a random number for some sense of scale. At upwards of 500ish good chiplets per 300mm wafer, that would be as many as 40M chiplets/month. If AMD achieved 100% market share on new sales, that would be enough chiplets to meet the whole year's desktop demand (about 90M units) in less than three months. 80k wafers may not sound like a lot but it is still a lot of chiplets and GPUs compared to the GPU and desktop x86 markets' sales volume. The figure is very likely in the correct general ballpark.

Also don't forget that TSMC isn't the only fab-for-hire in town, Samsung makes 7nm chips for many designers too.
 

InvalidError

Titan
Moderator
What kind of special design does Intel has?
Simple: AMD is brute-forcing its way to performance (ex.: 32MB of L3 per chiplet to offset the chiplets' 10-20ns worse memory latency) while Intel is finessing its architectural tweaks to extract whatever extra performance it can out of what it is stuck with until it has a new process it can push more aggressive changes with. AMD gets the lead for things that scale well with extra CPUs, Intel maintains its lead on things that require well-rounded performance out of fewer cores.
 
  • Like
Reactions: valeman2012
I've read that elsewhere too. But that really does very little to alter the underlying problem that TSMC itself only has finite manufacturing resources and if we're allowed to conjecture I'd be seriously doubtful they would hitch their wagon to one star only going forward. There has to be limits to what AMD can 'command' of TSMC and that article alerts us they are straining under the current production work load (to a degree at least).

What that tells me is AMD may be safe for now but beyond the current aggreements they have to be able to ship a LOT more processors each month to take even 10% more market share from Intel. It's just super hard to fathom anybody thinking AMD could take 10% market share from Intel three years ago, not even AMD. Hard to pound out an long-term agreement like that when nobody believes it so what's in it is something nobody knows. Add in that under this CEO, I'm pretty sure AMD is not interested in sitting still, not with two more spins of Zen already in the chute and they're talking of taking over the low-end desktop segment now with R5-3500(X). They want market share and that means they'll need more wafers & chiplets to do it and staying hitched to TSMC would have to limit them in that flexibility.

Intels production do s indeed dwarf amd's at tsmc. I think they have like 15% or something there abouts (dont quote me)

But none of that matters and here's why: intels vast amount of business comes from the lease market and servers. AMD is only the smallest inroads there. So supply is good.

Now lets say demand picks up in the business space. Servers moving to Epyc and big oem laptop contracts are indeed a big win for AMD. Fabrication would lean towards business products. And while that will lead to shortages and price hikes for zen 2/zen 3 on a consumer level it will make amd the money they need to grow. And eventually thats good for all of us.
 
Intel doesnt have the issues with iphones taking up production resources.

Intel did have issues with production, just not iphone related.
I understand now.

But then neither would AMD...no? I'm led to understand AMD uses a high power 7nm process while iPhone's using the low power 7nm. But then, I've no more idea how much commonality of resources exist between the processes, with the inevitable constraints creating a bottleneck there, any more than I do how much commonality of utilzation exist between 12 nm and 7 nm.
 
Intels production do s indeed dwarf amd's at tsmc. I think they have like 15% or something there abouts (dont quote me)

But none of that matters and here's why: intels vast amount of business comes from the lease market and servers. AMD is only the smallest inroads there. So supply is good.

Now lets say demand picks up in the business space. Servers moving to Epyc and big oem laptop contracts are indeed a big win for AMD. Fabrication would lean towards business products. And while that will lead to shortages and price hikes for zen 2/zen 3 on a consumer level it will make amd the money they need to grow. And eventually thats good for all of us.
That's where I disagree because that is where it matters to me, as a consumer of home desktop computing products.

The scenario I see playing out is AMD has to sign up for more long-term committments to consume TSMC wafers...probably wanting to lock in a deal for 5 years or more. This is your company you're talking about you have to present a 5 year plan at annual stockholders meetings and establish confidence you can actually execute it. Lacking agreements to a sole-source supplier for the most essential and complex piece of your product makes you look foolish to savvy investors.

If they sign up for those wafers...what 'flavor'? the flavor that most reliably produces chiplets for servers? or the flavor that most reliably produces chiplets for desktops? (they may start identical but engineering will come under intense pressure to maximize yield of the highest margin chiplets and you have to know they will find a way). I don't know if you have to get to such detail, but if they do it's a significant decision. Get it wrong and I'm pretty sure which AMD will opt to pay attention to: high-margin servers and HEDT markets.

Bad for me and even more bad press when your chips (high-end especially) are never available. Generating customer bad will at a time you want to grow and take away share from a supplier who's always been considered 'reliable' (even though you may have had to pay them a premium for it) is never a good thing.
 

Karadjgne

Titan
Ambassador
I think it's the software, and has been, that's been putting a damper on amd expansion. When you have popular software like Adobe CC that doesn't scale well at all past 8 threads, the higher single thread performance of the Intel chips is a bonus and AMD's cheaper high thread counts is meaningless. Been this way since FX. When time is money, and much more valuable than the pc, having software that just won't take advantage of AMD's resources and will take advantage of Intels single minded push towards single thread perfection, there's an imbalance and Intel gets the money.

And to me its a matter of point of view. Intel is quite small in scope, it's got pretty much most of its eggs wrapped up in pc's, with a few side flavors in consoles, tablets etc. AMD is into everything from pc's to consoles to phones to coke machines and coffee makers. So I guess it depends on the point of view you look at it as to who has more of what.
 
That's where I disagree because that is where it matters to me, as a consumer of home desktop computing products.

The scenario I see playing out is AMD has to sign up for more long-term committments to consume TSMC wafers...probably wanting to lock in a deal for 5 years or more. This is your company you're talking about you have to present a 5 year plan at annual stockholders meetings and establish confidence you can actually execute it. Lacking agreements to a sole-source supplier for the most essential and complex piece of your product makes you look foolish to savvy investors.

If they sign up for those wafers...what 'flavor'? the flavor that most reliably produces chiplets for servers? or the flavor that most reliably produces chiplets for desktops? (they may start identical but engineering will come under intense pressure to maximize yield of the highest margin chiplets and you have to know they will find a way). I don't know if you have to get to such detail, but if they do it's a significant decision. Get it wrong and I'm pretty sure which AMD will opt to pay attention to: high-margin servers and HEDT markets.

Bad for me and even more bad press when your chips (high-end especially) are never available. Generating customer bad will at a time you want to grow and take away share from a supplier who's always been considered 'reliable' (even though you may have had to pay them a premium for it) is never a good thing.
Supply agreements vary by vendor and AMD has been in this long enough to be savy with their contracts.

Btw: is usually based on a per wafer cost. It could be 100 small chiplettes or 5 monolithic server chips. Doesnt matter. Sometimes there are bonuses for high yields. The later is a win win for both companies.

At the end of it all we are peons (home users) to the global market share. If AMD wins the home marketshare mindset as they currently are, then its a win. If AMD moves toward more business oriented products it is a win.

As long as AMD presents competition to intel its a win for consumers in the end.

You worry way too much.
 
Supply agreements vary by vendor and AMD has been in this long enough to be savy with their contracts.

Btw: is usually based on a per wafer cost. It could be 100 small chiplettes or 5 monolithic server chips. Doesnt matter. Sometimes there are bonuses for high yields. The later is a win win for both companies.

At the end of it all we are peons (home users) to the global market share. If AMD wins the home marketshare mindset as they currently are, then its a win. If AMD moves toward more business oriented products it is a win.

As long as AMD presents competition to intel its a win for consumers in the end.

You worry way too much.
Ahhh... but I don't think I do!

You see, Intel has had a de-facto monopoly for quite some time and still has the market presence, and free-cash balance, to dominate the discussion in any area it wants. Back in the day when Athlons were kicking Intel's Pentium 4 butt they used that same dominance to lock up contracts with customers (pre-builts for growing business desktops primarily, Dell's bread-and-butter). In ways considered anti-competitive (by mostly EU regulators) to be sure so they may not be able to re-use those tactics but you can bet they're casting about for others.

If they can limit the wafers AMD can get it will have the same effect and AMD's competition will be effectively blunted.

I'm really thinking if AMD doesn't develop a second source for 7nm wafers, they're not ever going to challenge Intel effectively.
 

InvalidError

Titan
Moderator
And to me its a matter of point of view. Intel is quite small in scope, it's got pretty much most of its eggs wrapped up in pc's, with a few side flavors in consoles, tablets etc. AMD is into everything from pc's to consoles to phones to coke machines and coffee makers.
Intel also does networking equipment from backplane fabrics to modem and wireless chipsets, FPGAs, micro-controllers, semi-custom ASICs, power management, chipsets and probably a few other related incidental things.

It may be best known for x86 but it does a fair amount of stuff besides x86.

Open your coffee maker and you may actually find an Intel 8051 chip instead. One of my keyboards uses an 80186, didn't even know that was a thing before I opened it for cleaning many years ago.
 
  • Like
Reactions: digitalgriffin

Karadjgne

Titan
Ambassador
Well I did say a few side flavors, etc. Didn't know about the keyboards, that's a new idea lol. But for all that Intel does have, amd is spread out further. Whether that's a good thing, I dunno, it spreads a little cash in a very big area, so cpus get a proportionately smaller R&D fundage allowance. Just the addition of a gpu dept is a game changer all around, that's huge considering not just the consumer market but AMD's battle with nvidia for the professional market.

Gotta give amd kudos for just hanging in there for this long, when most others folded or sold out.
 
Simple: AMD is brute-forcing its way to performance (ex.: 32MB of L3 per chiplet to offset the chiplets' 10-20ns worse memory latency) while Intel is finessing its architectural tweaks to extract whatever extra performance it can out of what it is stuck with until it has a new process it can push more aggressive changes with. AMD gets the lead for things that scale well with extra CPUs, Intel maintains its lead on things that require well-rounded performance out of fewer cores.
Since the Ryzen 3000 has a higher IPC than current desktop Intel, that statement couldn't be more incorrect. Intel is able to have higher single threaded performance since they have been able to tweak 14nm for the past 5 years. At this point 14nm has a very high yield and has been worked on so much that clock speed allows it to have very high single threaded performance. Also outside of gaming, where there are tons of Intel optimizations, i9-9900k loses more often compared to Ryzen 3700X than it wins. The pushing of higher and higher clock speeds while on the same architecture would be the brute force method.
 
Ahhh... but I don't think I do!

You see, Intel has had a de-facto monopoly for quite some time and still has the market presence, and free-cash balance, to dominate the discussion in any area it wants. Back in the day when Athlons were kicking Intel's Pentium 4 butt they used that same dominance to lock up contracts with customers (pre-builts for growing business desktops primarily, Dell's bread-and-butter). In ways considered anti-competitive (by mostly EU regulators) to be sure so they may not be able to re-use those tactics but you can bet they're casting about for others.

If they can limit the wafers AMD can get it will have the same effect and AMD's competition will be effectively blunted.

I'm really thinking if AMD doesn't develop a second source for 7nm wafers, they're not ever going to challenge Intel effectively.

Intel in no way is going to soak up TSMC's wafer production. And in a few years 7nm+ and 5nm will be here.

Yes, AMD is dependent upon TSMC. But TSMC is expanding. I'm not worried. I'm more worried about AMD's ability to keep innovating enough to stay 1 step ahead of intel on the Price : Performance ratio without hurting their bottom line too much the way bulldozer did.

Intel is about 2 years off on Xe and Jim Kellers designs should show up in that time frame also. But competition is good.
 

InvalidError

Titan
Moderator
Since the Ryzen 3000 has a higher IPC than current desktop Intel, that statement couldn't be more incorrect.
IPC depends heavily on what sort of instruction mix is going through the CPU, optimization effort and how those instructions stress the rest of the system architecture. Ryzen often has higher IPC in productivity-oriented cases likely in large parts thanks to its 32MB L3 cache, Intel has higher IPC in most games due to higher optimization effort from Intel and developers.

Here's a comparison between the two at 4GHz: https://www.techspot.com/article/1876-4ghz-ryzen-3rd-gen-vs-core-i9/
 
Intel in no way is going to soak up TSMC's wafer production. And in a few years 7nm+ and 5nm will be here.

Yes, AMD is dependent upon TSMC. But TSMC is expanding. I'm not worried. I'm more worried about AMD's ability to keep innovating enough to stay 1 step ahead of intel on the Price : Performance ratio without hurting their bottom line too much the way bulldozer did.

Intel is about 2 years off on Xe and Jim Kellers designs should show up in that time frame also. But competition is good.

While I'd still like to be able to continue to afford inexpensive HPED-equivalent system in my home, I personally believe it isn't the Lisa Su's AMD goal to remain the 'value leader' forever though.

I don't have your confidence Intel won't use every means at their disposal to keep AMD from getting there even if it's not (simplisticaly) buying up all their wafers.

I'm as curious what Intel's new designs will bring as I am to see what Zen3 and Zen4 is about. I have a lot of doubt it's about raw performance even though people have suggested Zen3 will be basically an uplift in clock speeds as it's maturing. I think it's about more clever ways of getting performance only at the time it's needed. It's all about power efficiency now-a-days.
 

TRENDING THREADS