How to take 50% market share in three years.

[yulgrhet]

The genius of Lisa Su.

The creme-de-la-creme of the server market is the big cloud players who buy chips by the millions. That is who AMD has already sampled EPYC 2 to.

AMD is offering a chip, EPYC 2, that will have a !!!!! ~3X !!!!! perf/watt/$ advantage over Intel and a roadmap that will maintain that level of advantage for the next three years vs the now stalled out Intel.

Perf/watt/$ is king. 3X perf/watt/$ is king of kings. That is literally an offer the big cloud players can't refuse. And won't refuse. They will wholesale turn to AMD for their new build outs.

Taking the first 50% of the server market won't be hard, the second 50% will be slower, but by then every OEM will have a full line of AMD servers and be pushing them because if they don't their competition will and take their business.

The further genius was switching Zen 2 to an uncore+chiplet architecture mid development cycle to double the number of CPUs available to service the anticipated demand, substantially cut EPYC 2's cost, seriously advance the performance of the roadmap and, through the uncore, provide for myriad customizations across the CPU market and even within a market.

AMD could have half a dozen custom uncores planned for Zen 2 and rapidly and easily steer their 7nm chiplets during the packaging step to whatever market segment needs more supply.

 

[UsandThem]

The taking of the server market is not going to be easy, regardless if it is the first 50% or not.

Saying so on a road-map is one thing, executing those goals is a whole another beast. If AMD can pull it off, great. But Intel has owned that market for a long time, and they aren't giving it up without a fight.

 

[yulgrhet]

Saying so on a road-map is one thing, executing those goals is a whole another beast. If AMD can pull it off, great. But Intel has owned that market for a long time, and they aren't giving it up without a fight.

Can't fight without ammunition. Intel is stuck with 28 cores on 14nm for at least the next two years. Among the TOC focused cloud and DC whales EPYC 2, with 64 cores drawing less power at half the price AND a roadmap that will continue to substantially outperform Intel for several years counts as Intel having no ammunition.

No server whale can afford to pass up that kind of competitive edge.

 

[epsilon84]

AMD is offering a chip, EPYC 2, that will have a !!!!! ~3X !!!!! perf/watt/$ advantage over Intel and a roadmap that will maintain that level of advantage for the next three years vs the now stalled out Intel.

Perf/watt/$ is king. 3X perf/watt/$ is king of kings. That is literally an offer the big cloud players can't refuse. And won't refuse. They will wholesale turn to AMD for their new build outs.

First of all, I'm interested in how you came up with this 3x performance/watt/$ figure. You've got 3 variables there, and I'm pretty sure Intel has a whole lot of server CPUs that don't all have the same performance/watt, or price/performance ratio. So do enlighten us how you came up with that stat.

Secondly, a year is a long time in tech, let alone 3 years. How can we possibly predict how AMD will compare with Intel in 3 years when neither products are available right now? Also, a simple price adjustment from Intel and your '3x' number goes out the window.

That being said, I'm sure AMD is gaining ground on Intel in the server space, but it's a VERY long road to a 50% slice of the pie...

 

[UsandThem]

Can't fight without ammunition. Intel is stuck with 28 cores on 14nm for at least the next two years. Among the TOC focused cloud and DC whales EPYC 2, with 64 cores drawing less power at half the price AND a roadmap that will continue to substantially outperform Intel for several years counts as no ammunition.

No server whale can afford to pass up that kind of competitive edge.

Well, I guess we can bookmark this thread, and revisit it in two years, and see if your prediction comes true.

IMO, if AMD gets a 20 - 25% share within two years, it would be a huge win for them (which would match their best server market share in their history). Intel has deep pockets, and they have shown in the past they are willing to offer $$$ to ensure their products are selected by those who make purchase decisions.

 

[NTMBK]

This is just a straight copy paste of your Semiaccurate post, you should show more variety

 

[yulgrhet]

First of all, I'm interested in how you came up with this 3x performance/watt/$ figure.

As Intel's 10nm woes deepened Lisa Su knew AMD had an unprecedented opportunity to take market share. The was however one major unavoidable constraint, 7nm would be ramping and 7nm wafers in intense demand. AMD couldn't just order up more wafers, it had to better utilize what would be available.

The only way to do that was to increase the number of dies on the wafer, hence the move to an uncore+chiplet architecture which was undoubtedly already in progress for use later in the Zen series, probably the 'next gen' architecture roadmapped after Zen 3.

Moving the uncore (and L3) to another chip would halve the size of the die PLUS increasing wafer yield due to the much smaller dies. and, calculating the increase in demand for Zen 2 2.0, also allowing increasing the CCX from cores to 8 cores. A triple whammy.

To cap:

An uncore arch would double the 7nm dies and so double the number of producible Zen 2 CPUs. Decrease the 7nm fabrication cost per CPU. As the uncore would be on an older process it would be far faster and cheaper to engineer and fabricate, opening the door to custom uncore chips for each market segment.

Supporting data:

1. AMD has an unprecedented opportunity to take market share from Intel.
2. Charlie D's recent article AMD dramatically changed Zen 2 more than 2 years ago increasing the CCX core from 6 to 8.
3. AMD has leapfrogging design teams.
4. Zen 3 changes would now be substantially simplified, freeing up manpower to work on Zen 2.
5. The infamous draining of RTC to work in the CPU division at that time.
6. And of course the fact that the ONLY way to take more market share from Intel was manufacturing more substantially more dies on a 7nm wafer to produce more Zen 2 CPUs and only an uncore architecture could accomplish that.

It all dovetails together in a logical manner.

The payoff will be staggering, not only with Zen 2, but with a substantial cost/performance increase over the original roadmap in the following years.

The market share picture will be even worse for Intel in the HEDT, high end gaming, entry and mid level computers and the entire mobile market. Zen 2 + Navi APUs will take over notebooks, laptops and X86 tablets.

The amount of pain Intel faces in the coming years borders on the inconceivable.

The 3x part:

The 7nm process = double the density and 40% more performance vs 14nm.
The uncore = further doubles the cores per wafer + 10-15% IPC increase, decrease in latency, substantially higher clocks and whatever additional efficiencies an uncore architecture brings to the table.
Zen 1's modular design already costs AMD 60% of what it cost Intel to manufacture an equivalent amount of performance with Intel on a better process.

Already 60% the cost of Intel + 64 cores vs 28 cores, double the density + double the cores = 4X increase in cores per wafer + 40 performance increase + increase in IPC + decrease in latency + various uncore efficiencies ... you get the idea. 3X cost/watt/$ advantage is, if anything, understating the case.

 

[yulgrhet]

This is just a straight copy paste of your Semiaccurate post, you should show more variety

See my next post, I always think of more to add as I work on a new post.

 

[Dayman1225]

increasing the CCX from cores to 8 cores.

Not happening, atleast not with Rome.

 

[epsilon84]

As Intel's 10nm woes deepened Lisa Su knew AMD had an unprecedented opportunity to take market share. The was however one major unavoidable constraint, 7nm would be ramping and 7nm wafers in intense demand. AMD couldn't just order up more wafers, it had to better utilize what would be available.

The only way to do that was to increase the number of dies on the wafer, hence the move to an uncore+chiplet architecture which was undoubtedly already in progress for use later in the Zen series, probably the 'next gen' architecture roadmapped after Zen 3.

Moving the uncore (and L3) to another chip would halve the size of the die PLUS increasing wafer yield due to the much smaller dies. and, calculating the increase in demand for Zen 2 2.0, also allowing increasing the CCX from cores to 8 cores. A triple whammy.

To cap:

An uncore arch would double the 7nm dies and so double the number of producible Zen 2 CPUs. Decrease the 7nm fabrication cost per CPU. As the uncore would be on an older process it would be far faster and cheaper to engineer and fabricate, opening the door to custom uncore chips for each market segment.

Supporting data:

AMD has an unprecedented opportunity to take market share from Intel.
Charlie D's recent article AMD dramatically changed Zen 2 more than 2 years ago increasing the CCX core from 6 to 8.
AMD has leapfrogging design teams.
Zen 3 changes would now be substantially simplified, freeing up manpower to work on Zen 2.
The infamous draining of RTC to work in the CPU division at that time.
And of course the fact that the ONLY way to take more market share from Intel was manufacturing more dies on a 7nm wafer to produce more Zen 2 CPUs.

The market share picture will be even worse for Intel in the HEDT, high end gaming, entry and mid level computers and the entire mobile market. Zen 2 + Navi will take over notebooks, laptops and X86 tablets.

You managed to write a long winded post based on theoreticals yet completely ignored my actual question - which was, I'll repeat, how did you work out the 3x performance/watt/$ figure?

 

[happy medium]

The genius of Lisa Su.

The creme-de-la-creme of the server market is the big cloud players who buy chips by the millions. That is who AMD has already sampled EPYC 2 to.

AMD is offering a chip, EPYC 2, that will have a !!!!! ~3X !!!!! perf/watt/$ advantage over Intel and a roadmap that will maintain that level of advantage for the next three years vs the now stalled out Intel.

Perf/watt/$ is king. 3X perf/watt/$ is king of kings. That is literally an offer the big cloud players can't refuse. And won't refuse. They will wholesale turn to AMD for their new build outs.

Taking the first 50% of the server market won't be hard, the second 50% will be slower, but by then every OEM will have a full line of AMD servers and be pushing them because if they don't their competition will and take their business.

The further genius was switching EPYC 2 to an uncore+chiplet architecture mid development cycle to double the number of CPUs available to service the anticipated demand, substantially cut EPYC 2's cost, seriously advance the performance of the roadmap and, through the uncore, provide for myriad customizations across the CPU market and even within a market.

AMD could have half a dozen custom uncores planned for Zen 2 and rapidly and easily steer their 7nm chiplets during the packaging step to whatever market segment needs more supply.

Honestly, should I wake you now?

 

[Dayman1225]

If 50% happens by 2020/21 I will eat 1, maybe even 2 hats.

 

[epsilon84]

If 50% happens by 2020/21 I will eat 1, maybe even 2 hats.

Easy there now, don't bite off more than you can chew!

 

[yulgrhet]

If 50% happens by 2020/21 I will eat 1, maybe even 2 hats.

I recommend beaver or rabbit.

 

[yulgrhet]

Not happening, atleast not with Rome.

Easily doable with an uncore architecture. Also easily doable is a 128 core EPYC 3. Imagine that vs. a 28 core Intel CPU.

 

[Dayman1225]

Easily doable with an uncore architecture. Also easily doable is a 128 core EPYC 3.

Doable yes, doing, no, at least not with Rome.

 

[whm1974]

Dude, I think it unrealistic to expect AMD to take 50% in three years. AMD just gotten back into the server market, and Intel has owned that market for long time. And come to think of it, AMD just also has gotten back to being competitive with Intel as well.

Lets us see if AMD can continue it's streak first.

 

[sandorski]

I hope it happens, but not sure it will. If they stay competitive for 5 years it's more likely.

 

[yulgrhet]

You managed to write a long winded post based on theoreticals yet completely ignored my actual question - which was, I'll repeat, how did you work out the 3x performance/watt/$ figure?

It was a long winded pertinent fact filled post.

I'll do the arithmetic for you for Intel's 28 core chip vs. EPYC 2's 64 core chip on a per core basis.

A Zen core is 60% the cost of an Intel core due to the modular vs monolithic design.

Costs:

Wafer cost 7nm vs 14nm = 60% more per wafer give or take.

Below is a relative comparison.

100 14nm dies = $1000 = $10 per die.
400 7nm dies = $1600 = $4 per die = 2.5X cost advantage per die Zen 2 vs. Zen.
Zen 14nm vs Intel 14nm per core cost = 1.7X advantage.
7nm Zen 2 vs 14nm Intel = 2.5 X 1.7 = 4,25.
It costs Intel >4 times more per core than Zen 2.

COST = 4x advantage. That alone exceeds my 3X perf/watt/$ calculation.

Watts - we'll make it easy and say AMD's chiplet 7nm core is twice as efficient per core as Intel's 14nm.
Watts = 2X advantage.

Performance - clock gains + IPC gains + fabric gains + uncore gains = 30% per core should be conservatively safe.
Performance = 1.3X advantage.

4.25 + 2 + 1.3 = 7.55 X advantage over Intel. Take a generous 40% off for various and sundry moderating factors like adding in the uncore cost ... 7.55 X 60% = 4.53.

Conclusion - AMD's Zen 2 actually has a ~4.5X perf/cost/$ advantage over Intel. FOR TWO YEARS MINIMUM.

Oh my.

 

[Burpo]

Just curious.. Where do you get your drugs?

 

[Markfw]

Just curious.. Where do you get your drugs?

I don't think he is in outer space, just a little too optimistic. Even Intel said they would try to keep it under 20%, and I think 25% in 2 years is realistic. But not 50%

 

[epsilon84]

I don't think he is in outer space, just a little too optimistic. Even Intel said they would try to keep it under 20%, and I think 25% in 2 years is realistic. But not 50%

25% is a far cry from 50%. That's like expecting AMD to grow at 3 times their best case forecast - 15% per year rather than 5%.

If that isn't pie in the sky stuff then I don't know what is.

 

[IEC]

50%? In three years? Lol.

I know orgs still using Xeon E5 v2 CPUs from the Sandy Bridge era. Plenty of Xeon v3 CPUs out there as well.

Cloud providers might be a different story, but it's going to be a long time before Epyc gains enough momentum (Rome wasn't built in a day, after all) to approach double digits, much less 50%. I somehow doubt that Intel with all its resources and cash on hand is going to sit still and let market share be taken. Even 20% would be a market-shaking impact.

 

[Markfw]

25% is a far cry from 50%. That's like expecting AMD to grow at 3 times their best case forecast - 15% per year rather than 5%.

If that isn't pie in the sky stuff then I don't know what is.

Didn't I just agree with you ??? I said 20-25% in 2-3 years.\

Edit: And I also said that what Intel expects, and is going to fight to stop. Jeesh, you can't even agree and be nice ?

 

[yulgrhet]

Just curious.. Where do you get your drugs?

Walgreens. Why?