Speculation: AMD's response to Intel's 8-core i9-9900K

[Spartak]

I'd be really surprised if Comet has any arch upgrades beyond any Smeltdown fixes. Just higher clocks and HT; with the i9 getting 10 cores.

It makes sense when you realize that Icelake won't clock that high and might even be lower than Zen 2.

The rumored clock speed issues were almost certainly related to the 10nm process and not the architecture. Newer architectures are almost always designed for similar or higher clocks. Only completely new architectures (like core2 or zen) can cause clock speed regressions.

If their next gen isn't based on ice lake it'd be a de facto admission their 10nm alternative isnt really 10nm. Even for the stumbling Intel we are witnessing right now that would be an admission of failure high profile enough to send their stock cratering.

Look, I'm not an Intel fanboy. My surface pro 4 is the first product I own that has an Intel chip inside and that's just because Microsoft refused to launch an ARM version of the Surface. I've been building PC's for 20+ years which have always been AMD (and Cyrix before that).

But the way people expect AMD to launch Zen2 in april/may of 2019 and Intel at the same time to pull yet another 14nm skylake refresh is a very distorted view of the actual competitive landscape.

 

[scannall]

People keep comparing Zen2 to the current 9th gen but Zen2 is slated to be released in 19Q3/4 around the same time as Ice Lake. And just like Zen2 Ice Lake will have serious IPC improvements.

There are no indications anywhere of either serious IPC increases, or of much improved clocks. I rather doubt that the initial 10nm will get close to the current 14+++ in clocks. On the other hand, we have no hard numbers on Zen 2 either. It all makes for an interesting 2019 in any case.

 

[PotatoWithEarsOnSide]

Delboy used to say "this time next year I'm going to be a millionaire"

Until we start to see evidence that Intel has properly got to grips with 10nm, then the discussion will continue to be as it is. Even if they have, they still said it'd take 2 generations of 10nm before parity with their best 14nm...and that was before their 9th gen pushed the 14nm bar higher. What has changed with their 10nm progress since then for us to believe that 7nm for AMD won't be miles ahead of it at launch?

 

[Spartak]

Delboy used to say "this time next year I'm going to be a millionaire"

Until we start to see evidence that Intel has properly got to grips with 10nm, then the discussion will continue to be as it is. Even if they have, they still said it'd take 2 generations of 10nm before parity with their best 14nm...and that was before their 9th gen pushed the 14nm bar higher. What has changed with their 10nm progress since then for us to believe that 7nm for AMD won't be miles ahead of it at launch?

Yes you are one of those people that believe AMD is able to improve their clocks while making a 2 generational process jump, while Intel isnt even able to break even to their current clocks on a redialed 10/12nm node...potato talk indeed.

 

[Spartak]

There are no indications anywhere of either serious IPC increases, or of much improved clocks. I rather doubt that the initial 10nm will get close to the current 14+++ in clocks. On the other hand, we have no hard numbers on Zen 2 either. It all makes for an interesting 2019 in any case.

I guess even checking Wikipedia for what Ice Lake architecture is known to be is too much effort? Nobody including me is talking about improved clocks.

I do concur 2019 will be interesting as Intel's roadmap has received a hefty shakeup this past six months, so much is sure.

 

[jpiniero]

If their next gen isn't based on ice lake it'd be a de facto admission their 10nm alternative isnt really 10nm. Even for the stumbling Intel we are witnessing right now that would be an admission of failure high profile enough to send their stock cratering.

Icelake is still getting released, but it's mobile only.

 

[scannall]

I guess even checking Wikipedia for what Ice Lake architecture is known to be is too much effort? Nobody including me is talking about improved clocks.

I do concur 2019 will be interesting as Intel's roadmap has received a hefty shakeup this past six months, so much is sure.

There's nothing about expected IPC in Ice Lake Wiki. I'm not suggesting Intel is doomed or anything. They aren't. But their roadmaps have been more of a pothole filled dirt road for several years now, so i take them with a large grain of salt. When the products are actually in the wild we can all evaluate them together. Same with the AMD parts. Speculation is always fun, but it's not really useful for planning builds. Buy what's on the market, and fits your needs right now.

 

[IEC]

Microcenter is selling the 1700X for $129.99 in-store and you can save another $30 with a motherboard:
https://www.microcenter.com/single_product_results.aspx?sku=404848

If you would have told me AMD would be selling a competitive 8-core, 16-thread CPU for the cost of an i3 in 2015, I would have said you were crazy. On a budget, it's pretty much impossible to beat this level of CPU power used, much less new.

 

[Spartak]

There's nothing about expected IPC in Ice Lake Wiki. I'm not suggesting Intel is doomed or anything. They aren't. But their roadmaps have been more of a pothole filled dirt road for several years now, so i take them with a large grain of salt. When the products are actually in the wild we can all evaluate them together. Same with the AMD parts. Speculation is always fun, but it's not really useful for planning builds. Buy what's on the market, and fits your needs right now.

There are changes to L1 and L2 caches implying other changes as well, alongside AVX512.

An early geekbench result seems to imply a 10-15% IPC increase but it's hard to go by just a single result.

Taking a roadmap with a grain of salt is something different than dismissing it altogether BTW.

 

[PotatoWithEarsOnSide]

Their roadmaps was 10nm in 2017 (or earlier).
That worked out well for them didn't it?

 

[Spartak]

Jesus Christ how old are you?

Yes they had major difficulties and from now on they wont be able to fix things, even their redialed 10nm will be an abject failure, and ice lake will be thrown in the bin alongside any other improvements. 14nm skylake will live on and on and on till Intel goes bankrupt. You will be able to claim victory and I will bow to your superior insights!

 

[dlerious]

But the way people expect AMD to launch Zen2 in april/may of 2019 and Intel at the same time to pull yet another 14nm skylake refresh is a very distorted view of the actual competitive landscape.

1st gen R7 launched March 2, R5 launched April 11. 2nd gen R7 and R5 launched April 19. AMD has already shown 7nm Epyc, so why would 6 months from now for 3rd gen R5/R7 be unreasonable?

 

[scannall]

There are changes to L1 and L2 caches implying other changes as well, alongside AVX512.

An early geekbench result seems to imply a 10-15% IPC increase but it's hard to go by just a single result.

Taking a roadmap with a grain of salt is something different than dismissing it altogether BTW.

Will they get their process figured out, and we finally get to see the end of endless Skylake refreshes? Of course. That's a given, being TSMC is already pumping out 7nm parts, there is no reason Intel can't overcome their missteps. And I'm certain they will. Will it fall in this current revision of their roadmap? I have no idea, being it has been revised so many times. And what will the product look like when they finally get it out the door? I have no idea, and neither do you. But it will be a refreshing change regardless. Progress, even if the first iteration is underwhelming is good, because at least there is something to build from.

 

[PotatoWithEarsOnSide]

@Spartak no need for you to go mental over someone having a different opinion to you.
Without any evidence supporting your view, I'd rather just stick with what we do know.

 

[DrMrLordX]

Zen2 is slated to be released in 19Q3/4 around the same time as Ice Lake.

You keep repeating this as fact, and yet all you have is Charlie. AMD has never said Matisse will be coming out that late. Why would that happen? I mean honestly?

Also, The IceLake that comes out in late 2019 is mobile only.

1st gen R7 launched March 2, R5 launched April 11. 2nd gen R7 and R5 launched April 19. AMD has already shown 7nm Epyc, so why would 6 months from now for 3rd gen R5/R7 be unreasonable?

Exactly.

AMD's being cagey about actually saying that, so my money's on the mobo OEMs having problems getting X370/X470 compatibility squared away. X570 (or whatever) will likely work "just fine" with Matisse. Doing a drop-in replacement on older motherboards may not be viable until May/June or later.

 

[maddie]

This is AMD's response to i9 9900k.

An early in 2019 release of Rome with Ryzen 3xxx following a month or two after. Who knows, maybe we'll also be able to buy a 9900k by then.


Anandtech's Ian Cutress interview with AMD's Mark Papermaster.

Quote:
"IC: There have been suggestions that because AMD is saying that Rome is coming in 2019 then that means Q4 2019.

MP: We’re not trying to imply any specific quarter or time frame in 2019. If we look at today’s event, it was timed it to launch our MI60 GPU in 7nm which is imminent. We wanted to really share with the industry how we’ve embraced 7nm, and preview what’s coming out very soon with MI60, and really share our approach on CPU on Zen 2 and Rome. We’re not implying any particular time in 2019, but we’ll be forthcoming with that. Even though the GPU is PCIe 3.0 backwards compatible, it helps for a PCIe 4.0 GPU to have a PCIe 4.0 CPU to connect to!"


Last sentence in reply sure seems to me that MP is strongly hinting at the early release of Rome, the only PCIe 4.0 CPU at present. Yet we have these Q3/Q4 rumors being stated as fact with absolutely no attempt to pin to any corroborating data.

 

[jpiniero]

Last sentence in reply sure seems to me that MP is strongly hinting at the early release of Rome, the only PCIe 4.0 CPU at present. Yet we have these Q3/Q4 rumors being stated as fact with absolutely no attempt to pin to any corroborating data.

Rome's public release obviously isn't in Q1, if only because of the annoucement of the 7371. They obviously wouldn't have bothered if Rome was a Q1 launch. Still believe Ryzen will be at least 3 months from Rome's public release.

 

[DrMrLordX]

Rome's public release obviously isn't in Q1, if only because of the annoucement of the 7371. They obviously wouldn't have bothered if Rome was a Q1 launch. Still believe Ryzen will be at least 3 months from Rome's public release.

You may be right that Rome will be directed to under-the-table ODM sales at first; however, I do not think that has any bearing on when Matisse hits the shelves.

 

[jpiniero]

You may be right that Rome will be directed to under-the-table ODM sales at first; however, I do not think that has any bearing on when Matisse hits the shelves.

It does, since the Cloud Guys are going to get the first shot at the dies. Then the OEMs for Epyc 2's public launch. Then Matisse. At the time of the presentation, AMD may not have known exactly how much either the Cloud Guys or the OEMs were going to buy, compared to what the yields look like and the wafer allocation.

 

[DrMrLordX]

It does, since the Cloud Guys are going to get the first shot at the dies. Then the OEMs for Epyc 2's public launch. Then Matisse. At the time of the presentation, AMD may not have known exactly how much either the Cloud Guys or the OEMs were going to buy, compared to what the yields look like and the wafer allocation.

That implies potential wafer shortages.

 

[maddie]

That implies potential wafer shortages.

Correct, and with Apple sales appearing to be "challenged" going forward, this might not be a problem, if it ever was.

 

[jpiniero]

That implies potential wafer shortages.

Not supply, more demand. They're only going to buy so many wafers, especially when it's so expensive. Once the Cloud Guys and the OEMs are satisfied, only then will you see Matisse.

Epyc 2 would have to be a bust to see Matisse released when you think it will be.

 

[DrMrLordX]

Not supply, more demand. They're only going to buy so many wafers, especially when it's so expensive.

Wait. We're talking about AMD here, aren't we? AMD that over ten years ago, failed to capture market share when they had the chance due to supply shortages? The last thing they're gonna do is let market share slip through their fingers because 7nm is "so expensive". They have the chance to drive a knife straight into Intel's kidneys, and they're gonna skip on it, because of wafer costs? No, I don't think so.

 

[Atari2600]

They're only going to buy so many wafers, especially when it's so expensive.

What? Of all the reasons for a delay, that is the most far-fetched.

If it did not pay AMD to start the wafers, then they would be better off just closing up shop!

If anything, ordering more wafers* means a lower cost per chip due to further amortisation of fixed costs over those extra chips.


*as long as they can sell the chips produced - which should be apparent to AMD based on feedback from sampling.

 

[PotatoWithEarsOnSide]

If there is spare capacity at TSMC, and it is available for AMD to take up, then you can be 100% certain that they will do just that.
Development costs are entirely upfront, with production costs being marginal in comparison, though directly related to potential revenues; it'd be foolish not to buy wafers if their product is as good as we're expecting, since demand will almost certainly outstrip supply.