Middle Generation Console Upgrade Discussion [Scorpio, 4Pro]

[dobwal]

Because it's already done! A solution exists already and is in a released product.

And? All that does is save MS money. Why not take MS's money to serve as an investment into low powered CPU high Tflop Gpu Zen based design, which has application in other spaces. If it's technically and financially feasible, AMD can simply sell it to MS as a modern design that will make compatibility easier going forward. I imagine after Scorpio is released all the enhanced Scorpio version of titles are going serve as the BC titles for future based Xbox consoles anyways.

The semi-custom business serves AMD better if it acts as a center of R&D for new technology.

Plus it would serve MS as the cost to move to newer nodes can be shared across a plethora of zen based products versus simply sharing that cost with Sony.

 

[dobwal]

Does it matter to AMD as long as they get the money ? If MS is willing to pay more to AMD for zen than jaguar than that is profit they wouldn't otherwise make. As far as I'm aware MS pays a liscesing fee and then has it produced at the foundry of their choosing. So I don't see it taking zen chips away from AMD to sell to servers and enthusiasts anymore than a jaguar chip would by simply taking up fab time.

First AMD from a business standpoint is fabless. They contract out chip manufacturing to TSMC and GF. Second, AMD doesn't license out x86 based designs. Their agreement with Intel forbids such actions. They sell APUs to both MS and Sony.

 

[Michellstar]

And? All that does is save MS money. Why not take MS's money to serve as an investment into low powered CPU high Tflop Gpu Zen based design, which has application in other spaces. If it's technically and financially feasible, AMD can simply sell it to MS as a modern design that will make compatibility easier going forward. I imagine after Scorpio is released all the enhanced Scorpio version of titles are going serve as the BC titles for future based Xbox consoles anyways.

The semi-custom business serves AMD better if it acts as a center of R&D for new technology.

Plus it would serve MS as the cost to move to newer nodes can be shared across a plethora of zen based products versus simply sharing that cost with Sony.

Moreover, we know that jaguars into PS4pro are almost at the upper level of their frequency spectrum, you won´t get much more headroom from this micro-architecture (Puma tops at 2 ghz into desktops going by the wiki).

Also at the time of the conception of both Orbis and Durango, they got the latest graphics ip, and i think jaguar was first used on both designs as well.

We don´t know what scorpio will get, but i´m not rulling out Zen, AMD intents to use it everywhere.

I hope MS is not fooled into buying some excavator, and get the real deal, lol


The reasons Ms might or might not use ZEN, will have more to do with cost, thermals, die-size, etc etc

Until we get real data when ZEN is introduced, some more knowledgeable than me (easy) will make an educated guess. Raven ridge, the 8 core zen has 16mb of L3 cache and HT, but is this necessary in the console space (the more the merrier i know, but at a cost)

We´ll see

 

[Deleted member 13524]

I hope MS is not fooled into buying some excavator, and get the real deal, lol

The Excavator cores in the latest Bristol Ridge APUs are pretty good, actually.
Sure, IPC still doesn't touch anywhere near Skylake, but the latest generation has the FX9800P
has 2 modules / 4 threads at 2.7GHz minimum / 3.7GHz boost with a 8CU/512sp GPU at 750MHz max. within a 15W TDP.

Make those cores in 14/16FF and I bet they could get 4 modules at >3.2GHz while consuming less than 20W (cpu only).
Sure, Zen would be much better but it's not like the latest Excavator iteration is terrible.
And it would still be a whole lot better than going with Jaguar which hasn't had any serious overhaul for ~4 years.


I think the sole reason why Sony went with Jaguar for the Pro was for compatibility reasons. On compatibility mode the CPU cores can be downclocked to 1.6GHz, GPU down to 800MHz and gets cut in half and all timings just fit in the old code.

Microsoft has already shown that the Xbone's software doesn't mind having higher clocks, since the Xbone S already clocks higher everywhere and it doesn't need any compatibility mode as the games just run faster. This is probably because everything runs in virtual machines.
OTOH in the PS4 some AI, Physics and even render engines may be tightly synchronized with CPU clocks, so suddenly changing those without changing the code for each game could cause havoc.

So Microsoft has already shown that their software ecosystem can afford slightly different hardware, whereas Sony's cannot.
That said, it would be pretty stupid for Microsoft to keep using Jaguar cores.
Excavator or Zen. Zen would be ideal, but Excavator would be pretty good too.

 

[turkey]

Microsoft has already shown that the Xbone's software doesn't mind having higher clocks, since the Xbone S already clocks higher everywhere and it doesn't need any compatibility mode as the games just run faster. This is probably because everything runs in virtual machines.
OTOH in the PS4 some AI, Physics and even render engines may be tightly synchronized with CPU clocks, so suddenly changing those without changing the code for each game could cause havoc.

So Microsoft has already shown that their software ecosystem can afford slightly different hardware, whereas Sony's cannot.
That said, it would be pretty stupid for Microsoft to keep using Jaguar cores.
Excavator or Zen. Zen would be ideal, but Excavator would be pretty good too.

Xbox One S they run the CPU at the same 1.75GHz and its the GPU and by association ESRAM that is overclocked.

 

[Michellstar]

@ToTTenTranz I know that latest incarnation of excavator are really decent.
What AMDs considers an 8 core excavator? a 4 module? With shared FPU?
That´s what i´m worried, with much higher clocks of course it shines past jaguar/puma. But with similar clocks, and highly optimised code, i´m not so sure that would be worth the hassle to change to excavator anyway. There was a reason that Sony pass on excavator for Orbis. I remember Sebbi talking about it.

 

[Kreten]

Why

No it doesn't. And stop saying, 'read it carefully' as if everyone's just being sloppy/lazy, especially when someone's gone to the effort of breaking it down. If you want to present an alternative reading that no-one else is getting, you'll have to spell it out.

"I talked about our Zen roadmap for our products, in terms of desktops, servers and notebooks, but one should expect Zen in our semi-custom roadmap as well as we look beyond 2017 into the 18/19 timeframe, and so we really do view this as developing foundational IP that can go into a number of different markets and we have good prospects in those areas."

Clause 1, recapping what's already been said.
Clause 2, these products are not the only products Zen will appear in
Clause 3, subordinate clause. "We look beyond 2017 into the 18/19 timeframe" means nothing without hanging off the previous clause.

If clause 3 didn't exist, you'd have a point - expect Zen in out roadmap - but clause 3 exists to clarify the timelines for the investors. That's the only purpose to clause 3 and the only interpretation that makes any logical or grammatical sense. That's the interpretation that everyone else is getting.

So again, if you disagree you'll have to break down what you're reading to spell it out for us thickies.

are you guys only quoting portion of the paragraph and deliberately leaving out sentence just b4 where your quote starts that says "BESIDES CONSOLES"
I'm on phone so can't highlight stuff but pay attention to sentence leading into what you keep disecting as it clearly is opposite of what you are saying.




Well, I think, if you look today at the semi-custom business, it's been primarily game console driven. And it’s been again very, very pleased with the franchise that we’ve built around that. We do have engagements outside of game consoles, and that will ramp, let’s call it, more in the 2018 timeframe. The other thing you should expect is I talked about our Zen roadmap for our products, so in terms of desktops servers and notebooks. But one should expect Zen in our semi-custom roadmap as well as we go – as we look beyond 2017 into the ’18, ’19 timeframe. And so, we really do view this as we’re developing foundational IP that can go into a number of different markets. And we have good prospects in those areas.

 

[iroboto]

So i'm still confused at this.

I get your story about the pickles but I don't see how it applies to AMD. AMD sells chips. The have a factory that makes chips. How is it different if this factory makes jaguar or zen chips ? Every jaguar wafer that AMD makes is one less Zen wafer. So if MS wants 1 million Jaguar wafers that is 1m zen wafers that AMD can't make and sell for servers or consumers. IF MS instead makes 1m Zen wafers ... its the same thing..

Actually IF AMD is making millions of Zens for MS it could be a benfit to AMD since they will keep the fabs running producing a chip and getting data back to improve yields. Also if MS needs 8 core zens and 6tflop gpus in a single APU package the ones that fail and may only be 8 cores but 4 tflops or 6 cores and so on can then be sold into the retail channel for AMD to make profits on instead of eating the costs

I figure that zen would cost more since it is bigger than jaguar. I get that but the APU in scorpio will still be larger and require different tooling than other jaguar apus anyway.


But I guess its not much longer till we find out.

Let's not get too caught up in the Walmart example, it was just an example of how a channel exploited a poor management decision in which Walmart would go on to dictate their supplier prices. That's about as effective as a buyout without ever having to buy them out.

I guess it comes down to market pressure. There are no such things as infinite resources, there are no fab facilities that can deliver infinite number of chips in a preset time. Other companies also fight for the same fabrication time. As a business you would always choose a higher margin product over a lower margin one until demand is exhausted and whatever remaining resources you would dedicate your facilities to making the high volume lower margin products. This only makes sense to maximize revenue.

AMD gains substantially more revenue from selling MS jaguar again, then to sell them a cheap Zen.

AMD had 3 chips running jaguar for 3 years, invariably it knows how to produce jaguar with minimal defects. AMD cannot sell semi custom chips that are low yield, MS and Sony had full ownership over their silicon that they paid for.

So the question comes why not a super cut down Zen? this comes down to ROI. Imagine you built this amazing product but it cost you 2-3 billions of dollars of shareholder money to produce. But you can sell these units at an absurdly high margin. Then a company comes along and says hey I want that and you haven't sold a chip yet. You say the price is $400. They say no no, too high, remove all this stuff here, I don't want or need them. Give it to me for $90 now. Your research indicated to you that you had a lot of demand at $400, are you going to waste your time with this buyer ?

If you say yes then why did you spend 2-4 B of shareholder money making a chip that's worth $400 only to cut it down to sell it at $90? Desperation must be extremely high to want to agree to this without even putting your product on the market. How many units would of have to sell in order to make up your investment ? Why not just spend significantly less and take your existing technologies and bring it up to performance that they required, it would have been significantly cheaper and the margins would go into older capital costs projects.

Then we can get into why MS would want a cut down Zen, how much are you cutting out.. etc. But lets say nothing was cut out, it was this amazing CPU. All of a sudden MS is taking your goods selling your 6. TFLOP VEGA + new Zen for $399, when you see the market leaders are selling their CPUs for $230+ for the CPU and $399 for the GPU (in the same performance categories lets say)... how much money are you losing out on?

AMD and MS are marketing to the same group, gamers. You want gamers to buy a Ryzen and a Vega, you don't want them buying Scorpio, because there's no margin in Scorpio for you (heck no margin for MS either), and you don't get any revenue from software sales, where MS does.

Anyway, not saying it's not Zen. Cost wise I see something else being a better fit.

 

[Shifty Geezer]

are you guys only quoting portion of the paragraph and deliberately leaving out sentence just b4 where your quote starts that says "BESIDES CONSOLES"

I'm not deliberately leaving anything out. Why would I do that?? I'm discussing what's been quoted. If there's more to it, that can be quoted too for its added info. Certainly if there's a clear caveat preceeding the quoted text that changes it's meaning, it should be referenced and should have been referenced much sooner.

 

[Silent_Buddha]

Sure, I wasn't going for accuracy. Just suggesting AMD may want to keep a premium brand if they've a chance of creating one. Think of it more like Ford unveiling a new Ferrari rival supersports car, and providing 90% the same experience in a cheap sports car for 10th the price that just sticks a new body over the same chassis and engine. They'd completely devalue the premium car doing that. Keeping it rare keeps the price high.

Of course, that all assumes availability. Even if AMD provided cheap Zen based consoles and expensive Zen based supercomputers, it's no like potential high-end customers will buy the consoles instead. But in terms of brand, they may want to keep Zen something special. I don't actually think that's the case though given other evidence.

But that's where core count and stuff outside of the core differentiates the market segments. Lets look at Intel for an example.

They use the same cores from Pentium (budget) to i3 (mainstream) to i5 (performance) to i7 (enthusiast/prosumer) to Xeon (enterprise). The number of cores, whether HT is enabled or not, presence of AVX and at what level, and non-core stuff (amount of cache, for example) are what differentiates things.

For an automobile analogy think of it as the engine block of the vehicle. IIRC, the Ford GT fused 2 engine blocks from their truck line. Those engine blocks are also resold to other automobile makers. Some of whom use them for el-cheapo vehicles and some of those use them for expensive track day vehicles.

BMW and Mercedes also sell their engine blocks to other car makers that often go into vehicles that are far far cheaper than the vehicles that BMW and Mercedes use them in.

Zen is just the core of the CPU/APU/SOC. Just like with the CPU cores of Skylake. Just like with Engine blocks on a vehicle.

I don't disagree with your post, AMD definitely wants to have a premium product. But whether it has a Zen core isn't what will differentiate their premium products.

Regards,
SB

 

[Shifty Geezer]

I don't disagree with your post...

I think neither you nor Jay read my post in its entirety, because even I disagree with my post.

 

[Deleted member 13524]

Zen is AMD's only x86 CPU architecture. AMD's "Ferrari to Ford" diversification will most probably come at number of cores, clocks and maybe enabling/disabling multithreading.

If Zen is all AMD has in the works for 2017, I don't see why Zen shouldn't be in Scorpio.

@ToTTenTranz I know that latest incarnation of excavator are really decent.
What AMDs considers an 8 core excavator? a 4 module? With shared FPU?
That´s what i´m worried, with much higher clocks of course it shines past jaguar/puma. But with similar clocks, and highly optimised code, i´m not so sure that would be worth the hassle to change to excavator anyway. There was a reason that Sony pass on excavator for Orbis. I remember Sebbi talking about it.

4 module with FPU, which is why I specifically mentioned 3.2GHz as 2*1.6GHz.
However, I should have said 3.5GHz because I keep forgetting the Xbone's cores run at 1.75GHz.

That way we'd have half the number of FP units running at twice the speed, so they'd get the same throughput in a worst-case scenario.

 

[Michellstar]

AMD gains substantially more revenue from selling MS jaguar again, then to sell them a cheap Zen.

AMD had 3 chips running jaguar for 3 years, invariably it knows how to produce jaguar with minimal defects. AMD cannot sell semi custom chips that are low yield, MS and Sony had full ownership over their silicon that they paid for.

So the question comes why not a super cut down Zen? this comes down to ROI. Imagine you built this amazing product but it cost you 2-3 billions of dollars of shareholder money to produce. But you can sell these units at an absurdly high margin. Then a company comes along and says hey I want that and you haven't sold a chip yet. You say the price is $400. They say no no, too high, remove all this stuff here, I don't want or need them. Give it to me for $90 now. Your research indicated to you that you had a lot of demand at $400, are you going to waste your time with this buyer ?

If you say yes then why did you spend 2-4 B of shareholder money making a chip that's worth $400 only to cut it down to sell it at $90? Desperation must be extremely high to want to agree to this without even putting your product on the market. How many units would of have to sell in order to make up your investment ? Why not just spend significantly less and take your existing technologies and bring it up to performance that they required, it would have been significantly cheaper and the margins would go into older capital costs projects.

Then we can get into why MS would want a cut down Zen, how much are you cutting out.. etc. But lets say nothing was cut out, it was this amazing CPU. All of a sudden MS is taking your goods selling your 6. TFLOP VEGA + new Zen for $399, when you see the market leaders are selling their CPUs for $230+ for the CPU and $399 for the GPU (in the same performance categories lets say)... how much money are you losing out on?

AMD and MS are marketing to the same group, gamers. You want gamers to buy a Ryzen and a Vega, you don't want them buying Scorpio, because there's no margin in Scorpio for you (heck no margin for MS either), and you don't get any revenue from software sales, where MS does.

Anyway, not saying it's not Zen. Cost wise I see something else being a better fit.

I´m not ruling out Zen either, but in my case it´s more a wish than anything.
Taking into consideration as well how the W10 part (dash, UPW apps, snap, etc) perform under jaguar. So Ms needs to improve this portion of the machine, with this premium console.

Xbox One estimated contract with AMD was valued at 3Billions for the lifespan of the console, not developing it, I mean that´s the thing with the semi custom division, they share the R&D, and all the work that could go into Scorpio would land into raven ridge, that´s how it went for them last time.

Anyway, good summary of what could had happened between AMD and Ms

 

[Jay]

I think neither you nor Jay read my post in its entirety, because even I disagree with my post.

lol sorry wasn't actually disagreeing, just saying we know it's not the case and why, adding on to your post.

I think in the link I posted it also mentioned about amd doing presentation at gdc on developing for zen. Which was also my point about their push to get their tech optimised for, and having it in consoles is a good way to help go about that.

how long we gotta wait until all is revealed..... Lol

 

[Rangers]

I had an idea today, if Scorpio is to be backwards/forwards compatible, why not just do away with Xbox One hardware after Scorpio launch? No more sold. Transition every new Xbox sold is a Scorpio. I wouldn't expect this day one, but maybe something that stealthily happens over the next months.

If MS feels XBO was just too underpowered which I think is plausible. PS4/Pro are much closer together.

Problems: No low priced option for the mainstream. XBO can be sold for 250 on special. Scorpio is likely 399 and up only. Software. No more XBox One hardware would likely mean XBO's days of getting ports of everything (the forwards compatible part) would possibly be numbered no matter how good MS's intentions. And the ports in the meanwhile might be increasingly shoddy.

 

[Jay]

I had an idea today, if Scorpio is to be backwards/forwards compatible, why not just do away with Xbox One hardware after Scorpio launch? No more sold. Transition every new Xbox sold is a Scorpio. I wouldn't expect this day one, but maybe something that stealthily happens over the next months.

If MS feels XBO was just too underpowered which I think is plausible. PS4/Pro are much closer together.

Problems: No low priced option for the mainstream. XBO can be sold for 250 on special. Scorpio is likely 399 and up only. Software. No more XBox One hardware would likely mean XBO's days of getting ports of everything (the forwards compatible part) would possibly be numbered no matter how good MS's intentions. And the ports in the meanwhile might be increasingly shoddy.

if I'm understanding you correctly, then I would say it's a bad idea.
I think the x1s is more important than the Scorpio. If had to get rid of one it would be Scorpio.
that doesn't mean Scorpio isn't important with a x1s there though.

 

[Kreten]

Sure, I wasn't going for accuracy. Just suggesting AMD may want to keep a premium brand if they've a chance of creating one. Think of it more like Ford unveiling a new Ferrari rival supersports car, and providing 90% the same experience in a cheap sports car for 10th the price that just sticks a new body over the same chassis and engine. They'd completely devalue the premium car doing that. Keeping it rare keeps the price high.

Of course, that all assumes availability. Even if AMD provided cheap Zen based consoles and expensive Zen based supercomputers, it's no like potential high-end customers will buy the consoles instead. But in terms of brand, they may want to keep Zen something special. I don't actually think that's the case though given other evidence.

Your comparison is a bit flawed would be more like Ferrarri unveiling a new engine and knowing that in their car they can only sell thousands and then Ford offers them to buy 30Million+ for a discount to put in new Ford Boat.

 

[Kreten]

Let's not get too caught up in the Walmart example, it was just an example of how a channel exploited a poor management decision in which Walmart would go on to dictate their supplier prices. That's about as effective as a buyout without ever having to buy them out.

I guess it comes down to market pressure. There are no such things as infinite resources, there are no fab facilities that can deliver infinite number of chips in a preset time. Other companies also fight for the same fabrication time. As a business you would always choose a higher margin product over a lower margin one until demand is exhausted and whatever remaining resources you would dedicate your facilities to making the high volume lower margin products. This only makes sense to maximize revenue.

AMD gains substantially more revenue from selling MS jaguar again, then to sell them a cheap Zen.

AMD had 3 chips running jaguar for 3 years, invariably it knows how to produce jaguar with minimal defects. AMD cannot sell semi custom chips that are low yield, MS and Sony had full ownership over their silicon that they paid for.

So the question comes why not a super cut down Zen? this comes down to ROI. Imagine you built this amazing product but it cost you 2-3 billions of dollars of shareholder money to produce. But you can sell these units at an absurdly high margin. Then a company comes along and says hey I want that and you haven't sold a chip yet. You say the price is $400. They say no no, too high, remove all this stuff here, I don't want or need them. Give it to me for $90 now. Your research indicated to you that you had a lot of demand at $400, are you going to waste your time with this buyer ?

If you say yes then why did you spend 2-4 B of shareholder money making a chip that's worth $400 only to cut it down to sell it at $90? Desperation must be extremely high to want to agree to this without even putting your product on the market. How many units would of have to sell in order to make up your investment ? Why not just spend significantly less and take your existing technologies and bring it up to performance that they required, it would have been significantly cheaper and the margins would go into older capital costs projects.

Then we can get into why MS would want a cut down Zen, how much are you cutting out.. etc. But lets say nothing was cut out, it was this amazing CPU. All of a sudden MS is taking your goods selling your 6. TFLOP VEGA + new Zen for $399, when you see the market leaders are selling their CPUs for $230+ for the CPU and $399 for the GPU (in the same performance categories lets say)... how much money are you losing out on?

AMD and MS are marketing to the same group, gamers. You want gamers to buy a Ryzen and a Vega, you don't want them buying Scorpio, because there's no margin in Scorpio for you (heck no margin for MS either), and you don't get any revenue from software sales, where MS does.

Anyway, not saying it's not Zen. Cost wise I see something else being a better fit.

I would argue that AMD selling millions of Zen chips to MS also ensures that their fab costs go down significantly faster where it leaves them in better position for server/pc/laptop as they would end up with even higher margins there and be in a position to lower price to get higher demand faster than competition.

Xbox and PC are not in same category it's same like comparing Boat to Car where boat has much more restrictions and can only go to certain places. Same thing with consoles and PC where console is closed ecosystem without demand from people looking to do work and school work and other things on it.

 

[iroboto]

I would argue that AMD selling millions of Zen chips to MS also ensures that their fab costs go down significantly faster where it leaves them in better position for server/pc/laptop as they would end up with even higher margins there and be in a position to lower price to get higher demand faster than competition.

Xbox and PC are not in same category it's same like comparing Boat to Car where boat has much more restrictions and can only go to certain places. Same thing with consoles and PC where console is closed ecosystem without demand from people looking to do work and school work and other things on it.

Possibly.

AMD outsources its Fabs, most companies will go to the big fab players, with the exception of Intel. I'm not entirely sure the experience of semi custom chip/cut down Zen APU, would improve the yield on full Zen CPUs. My assumption is that yield is the only cost savings there are, prices are dictated by the fabricator, or time spent renting the fabricator.

I imagine that going from APU to CPU would require major re-tooling, so I'm unsure if there are savings there either to be harvested.

 

[3dilettante]

I would argue that AMD selling millions of Zen chips to MS also ensures that their fab costs go down significantly faster where it leaves them in better position for server/pc/laptop as they would end up with even higher margins there and be in a position to lower price to get higher demand faster than competition.

AMD's first major consideration is the number of wafers it is obligated to buy as part of its agreement with Globalfoundries.
Penalties ensue below that. The details of what it costs below or above that threshold would be negotiated with fab costs in mind, but mainly by the foundry rather than the customer.

If GF cannot supply all of what AMD needs, AMD has the option of buying wafers from some other foundry--at the cost of paying GF what appears to be the margin GF would have gotten if the wafers were purchased there.
AMD's already paid for an exemption for "certain products" which was done before for APUs and the consoles, but which ones those are this time is unclear.

If overall demand requires taking the hit for using a second source, it is hopefully for a product that can be priced to pay the margins for two foundries.