Jump to content

Intel Offers More Cascade Lake-AP Performance Numbers

1 minute ago, CarlBar said:

 

This too, though the idea they'll limit Ryzen to 1 chiplet strikes me as unlikely. It just dosen;t fit how AMD's been going with the core/thread count.

Without knowing AMD's internal numbers, this is just speculation, but here's what would make the most sense for the 3rd Gen Ryzen stack.

 

12nm Raven Ridge Refresh:

3200G, 3400G

 

These will launch in January, likely. They're actually AMD's laptop parts, but they'll show up none the less.

 

1 7nm Chiplet + RomeIO die + Vega Chiplet:

3300X, 3400X, 3500X, 3600X, 3700X

 

These will be the normal range replacements.

 

2 7nm Chiplet + RomeIO die, no Vega:

 

3800X, 3850X

 

12c and 16c parts on AM4. This will act like the WX line on Threadripper 2nd gen. They're a drop-in upgrade for Render boxes. Lower clocks, non-gaming focused. Intel has 10c Desktop parts coming in 2020, per the rumors, so this will also let AMD head them off in that part of the market.

 

AM4 doesn't have a lot of space for chiplets on the package. They'll be able to sell a lot more OEM parts with some onboard graphics. Even if it's some 50mm2 Mini-Polaris part. The other thing it does is allow AMD to refresh for a 4th Gen Ryzen with a Navi Chiplet for graphics (smaller, cheaper, less power) in 2020 while also finding improvements by refreshing the IO die.

Link to comment
Share on other sites

Link to post
Share on other sites

53 minutes ago, GoldenLag said:

except binning isnt all about core clocks. also Ryzen didnt overclock better than threadripper. and as far as i know, Epyc didnt overclock any worse than neither. if they were making 4 core chiplets, they would be cutting down an allready small die, they would probably run into solder issues. 

doublechecked, i might have mixed some slides up, but at least the general consensis is some form of l4 cache as it is the only way that we can make sence of it. the infinity fabric now communicates through the IO die. which in turn dictates some sort of cache should be in place, though we dont know entirely

 

Except the biggest advantages of chiplet architecture are the clock binning, and higher yields.

 

Also L4 cache is still speculation. They went out of their way to emphasise massively reduced IF fabric latency.

 

50 minutes ago, Taf the Ghost said:

Without knowing AMD's internal numbers, this is just speculation, but here's what would make the most sense for the 3rd Gen Ryzen stack.

 

12nm Raven Ridge Refresh:

3200G, 3400G

 

These will launch in January, likely. They're actually AMD's laptop parts, but they'll show up none the less.

 

1 7nm Chiplet + RomeIO die + Vega Chiplet:

3300X, 3400X, 3500X, 3600X, 3700X

 

These will be the normal range replacements.

 

2 7nm Chiplet + RomeIO die, no Vega:

 

3800X, 3850X

 

12c and 16c parts on AM4. This will act like the WX line on Threadripper 2nd gen. They're a drop-in upgrade for Render boxes. Lower clocks, non-gaming focused. Intel has 10c Desktop parts coming in 2020, per the rumors, so this will also let AMD head them off in that part of the market.

 

AM4 doesn't have a lot of space for chiplets on the package. They'll be able to sell a lot more OEM parts with some onboard graphics. Even if it's some 50mm2 Mini-Polaris part. The other thing it does is allow AMD to refresh for a 4th Gen Ryzen with a Navi Chiplet for graphics (smaller, cheaper, less power) in 2020 while also finding improvements by refreshing the IO die.

 

AMD didn't include integrated graphics on their non-G chips this time around, and i don't think thats going to change, people don't buy high end desktop CPU's to use the integrated graphics. They might release some Ryzen 5 series G chips for OEM's, but thats the most i'd expect.

Link to comment
Share on other sites

Link to post
Share on other sites

2 minutes ago, CarlBar said:

 

Except the biggest advantages of chiplet architecture are the clock binning, and higher yields.

 

Also L4 cache is still speculation. They went out of their way to emphasise massively reduced IF fabric latency.

 

 

AMD didn't include integrated graphics on their non-G chips this time around, and i don't think thats going to change, people don't buy high end desktop CPU's to use the integrated graphics. They might release some Ryzen 5 series G chips for OEM's, but thats the most i'd expect.

OEMs are still the biggest part of the market. The IO die should be able to handle it without much issue, seeing as the Raven Ridge actually connects over PCIe anyway. But that OEM market matters a lot. All of those Ryzen Pro parts are without a GPU, thus they need a dGPU. Adding a 60mm2, 6 CU Polaris/Vega chiplet would open up around 50% of the Desktop market for AMD. 

 

I'm not making the enthusiast case for adding the GPU. I'm making the obvious business case.

Link to comment
Share on other sites

Link to post
Share on other sites

3 minutes ago, Taf the Ghost said:

OEMs are still the biggest part of the market. The IO die should be able to handle it without much issue, seeing as the Raven Ridge actually connects over PCIe anyway. But that OEM market matters a lot. All of those Ryzen Pro parts are without a GPU, thus they need a dGPU. Adding a 60mm2, 6 CU Polaris/Vega chiplet would open up around 50% of the Desktop market for AMD. 

 

I'm not making the enthusiast case for adding the GPU. I'm making the obvious business case.

 

It only makes sense if the OEM's will actually buy it though, and for that they need demand from consumers for AMD chips, and that comes down to brand appeal and mindshare. Intel have a pretty big lead here which isn't helped even now by the fact that they have the best enthusiast chips on the market. Saying their not the best but they're really cost effective is all well and good. But in practise thats just damming with faint praise to your average Joe or Jane. They need to hear the tech nerds in their social circles going on about how unbelievably good they are compared to anything else out there. And they need that to be going on for more than a half a generation or generation of chips to give it time to work at Intels lead, because it isn't going to happen overnight.

 

Thats kinda why i raised the Ryzen 9 idea. I don't think AMD will do it as they're not as fond atm of runaway thermals. But on paper they could do it and produce an absolute monster chip that would trash even the 9980XE. And that would get them the kind of press that would start to make a real dent in Intel's stranglehold on the desktop market. That's exactly what they've gone out and done with EPYC2. They took a rock solid architecture that was giving Intel stuff competition and they iterated on it whilst expanding the raw grunt under the hood in every way possibble. Even with 10nm up and running Intel would really be struggling to compete with EPYC 2, without it their dead in the water. But AMD need to bring that same level of killer punch to the desktop and badly if they're going to make real inroads there.

Link to comment
Share on other sites

Link to post
Share on other sites

Just now, CarlBar said:

 

It only makes sense if the OEM's will actually buy it though, and for that they need demand from consumers for AMD chips, and that comes down to brand appeal and mindshare. Intel have a pretty big lead here which isn't helped even now by the fact that they have the best enthusiast chips on the market. Saying their not the best but they're really cost effective is all well and good. But in practise thats just damming with faint praise to your average Joe or Jane. They need to hear the tech nerds in their social circles going on about how unbelievably good they are compared to anything else out there. And they need that to be going on for more than a half a generation or generation of chips to give it time to work at Intels lead, because it isn't going to happen overnight.

 

Thats kinda why i raised the Ryzen 9 idea. I don't think AMD will do it as they're not as fond atm of runaway thermals. But on paper they could do it and produce an absolute monster chip that would trash even the 9980XE. And that would get them the kind of press that would start to make a real dent in Intel's stranglehold on the desktop market. That's exactly what they've gone out and done with EPYC2. They took a rock solid architecture that was giving Intel stuff competition and they iterated on it whilst expanding the raw grunt under the hood in every way possibble. Even with 10nm up and running Intel would really be struggling to compete with EPYC 2, without it their dead in the water. But AMD need to bring that same level of killer punch to the desktop and badly if they're going to make real inroads there.

OEMs will buy whatever fits their criteria and price points. AMD actually has pretty good penetration in low-cost OEM markets in Eurasia & Africa. DDR3 still being important in those areas. Intel would have never added the iGPU if it didn't make business sense. Now, Intel also illegally kept AMD out of the OEM market for a long while, but the recent management teams haven't been quite as illegal about their actions.

 

Given that AMD's CTO said that Zen3 server parts will maintain the same socket until they need to move for DDR5, it's also extremely possible AMD will be able to sell OEMs on maintaining the same motherboards for several years. Keeping costs down like that makes OEMs happy. 

Link to comment
Share on other sites

Link to post
Share on other sites

5 minutes ago, Taf the Ghost said:

OEMs will buy whatever fits their criteria and price points. AMD actually has pretty good penetration in low-cost OEM markets in Eurasia & Africa. DDR3 still being important in those areas. Intel would have never added the iGPU if it didn't make business sense. Now, Intel also illegally kept AMD out of the OEM market for a long while, but the recent management teams haven't been quite as illegal about their actions.

 

Given that AMD's CTO said that Zen3 server parts will maintain the same socket until they need to move for DDR5, it's also extremely possible AMD will be able to sell OEMs on maintaining the same motherboards for several years. Keeping costs down like that makes OEMs happy. 

 

And having IGPU works for intel,l many users of systems don't want to do much the IGPU can't use and it lets them charge the OEM's a bit more whilst still saving the OEM's money overall.

 

And yeah they have penetration in the low cost market. Guess what, intel basically has nothing in that market for OEM's to use. Selling some system in those area's at those price points is better for OEM's than selling no systems. Selling well n a market where you have zero competition is nothing impressive.

Link to comment
Share on other sites

Link to post
Share on other sites

Intel Benchmarks: Created by Intel, for Intel. 

Link to comment
Share on other sites

Link to post
Share on other sites

2 hours ago, CarlBar said:

And having IGPU works for intel,l many users of systems don't want to do much the IGPU can't use and it lets them charge the OEM's a bit more whilst still saving the OEM's money overall.

In terms of global PC sales dedicated or at least above low end GPUs would be around 10% of sales (number out of my ass). Business computer sales far, far, outstrips all other sales in volume and those almost always use iGPU.

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×