Speculation: Ryzen 4000 series/Zen 3

[Glo.]

The memory bus width is directly linked to GPU cost. Wider bus means

-more pins on the GPU, so larger minimum die size
-more traces on the PCB, so probably more layers to be able to route them all
-more memory chips
-higher power consumption, so a larger cooler and more expensive power delivery circuits

I think it makes sense for a premium laptop APU, because it's replacing a 128-bit CPU and 128-bit GPU with a 192-bit APU- an overall reduction. But for low-end GPUs, it's going to be hard to make the economics work.

I was talking about a dGPU that is smaller than 200 mm2 in die size, like 156 mm2 or something that has 192 bit memory bus .

Like: RX 5500 XT/GTX 1650 segment replacement .

 

[Glo.]

I would be very, very suprised if AMD went with HBM2, because they're not even planning on doing that with consumer facting dGPUs afaik. Not anymore anyway.

As for LPDDR5... no clue. I mean, it would explain some things, but I would have thought supply would be tight that even Intel dropped LPDDR5 support from Tiger Lake.

Well, LPDDR5 will have lower cost than HBM2, and lower production complexity, so...

... there is plenty of forces who would be VERY interested in such solution.

LPDDR5 could be the ONLY thing that fits here, after all. You can get even similar memory bandwidth to HBM2, but for 16 CU chip, that has RDNA2 CUs it might not be necessary. 128 GB/s might be enough.

 

[moinmoin]

Regarding launch timing AMD's annual report on Form 10-K for 2019 contains some statements by Su dated March:
"We will build on this momentum with the launch of our next-generation Ryzen CPUs powered by our “Zen 3” processor core late in the year."
"We are on-track to accelerate our data center business with the introduction of our third-generation EPYC processor family late in the year."

Certainly sounds more like Autumn than Summer.

 

[NTMBK]

I was talking about a dGPU that is smaller than 200 mm2 in die size, like 156 mm2 or something that has 192 bit memory bus .

Like: RX 5500 XT/GTX 1650 segment replacement .

But that's my point- can you fit enough pins for a 192-bit bus on a chip that small? Smallest 192-bit chip I know of is the GP106 (Geforce 1060), and that's a 200mm^2 chip.

 

[Glo.]

But that's my point- can you fit enough pins for a 192-bit bus on a chip that small? Smallest 192-bit chip I know of is the GP106 (Geforce 1060), and that's a 200mm^2 chip.

RDNA2 CUs should be 20% more dense than RDNA1.

P.S. I believe that yes, you can make sub 200mm2 GPU die with 192 bit memory bus. But it will also be larger than 156 mm2 .

 

[DisEnchantment]

But that's my point- can you fit enough pins for a 192-bit bus on a chip that small? Smallest 192-bit chip I know of is the GP106 (Geforce 1060), and that's a 200mm^2 chip.

The die size is not an issue, because during Wafer assembly, there will be traces running on the relatively much bigger package substrate from the tiny logic die to the pin out/or ball grid or whatever. So you can have small 80mm2 cellphone chip connecting to multiple channels of LPDDR for example and it is normal.
Packaging is much cheaper than making the die itself.

EDIT:
Added pic to show the tiny dies getting processed , the final assembled chip has much bigger contacts and pins which can be routed on the PCB to DDR chips.
This is for simple chips, there are different ways to do wafer assembly.

 

[eek2121]

None of these benchmarks are even explicitly targeting the specific CPU microarchitecture, just the target ISA. If your background is compiler work then I hope you can differentiate between a compiler's front-end and back-end. Your previous big post is just a huge pile of nonsense, full of technical inaccuracies. I wish people would stop chiming in from an authoritative position when in reality they live in lala-land.

Regarding launch timing AMD's annual report on Form 10-K for 2019 contains some statements by Su dated March:
"We will build on this momentum with the launch of our next-generation Ryzen CPUs powered by our “Zen 3” processor core late in the year."
"We are on-track to accelerate our data center business with the introduction of our third-generation EPYC processor family late in the year."

Certainly sounds more like Autumn than Summer.

With the virus in full swing, AMD themselves likely don’t know the launch date.

 

[GoodRevrnd]

With the virus in full swing, AMD themselves likely don’t know the launch date.

Before Cyberpunk drops pls.

 

[Glo.]

But that's my point- can you fit enough pins for a 192-bit bus on a chip that small? Smallest 192-bit chip I know of is the GP106 (Geforce 1060), and that's a 200mm^2 chip.

After thinking about it: RX 480: 36 CU, 14 nm chip, 256 Bit GDDR5, 232 mm2.

Its a no issue.

 

[RetroZombie]

I think it makes sense for a premium laptop APU, because it's replacing a 128-bit CPU and 128-bit GPU with a 192-bit APU- an overall reduction. But for low-end GPUs, it's going to be hard to make the economics work.

Actually you solved the problem.

Since it would only be for mobile and soldered no socket, the high end parts would be binned/segmented with three memory channels, mid range parts dual channel, low end parts dual/single.

With this the oems would be forced to put more ram in the laptop models with high end processors which actually would be a good thing.

 

[Gideon]

Meanwhile Apple keeps adding Ice-Lake CPUs o Macos beta. I'm still not convinced the AMD apu will actually be used this soon:

https://twitter.com/x/status/1247934672648253441

 

[eek2121]

Meanwhile Apple keeps adding Ice-Lake CPUs o Macos beta. I'm still not convinced the AMD apu will actually be used this soon:

https://twitter.com/x/status/1247934672648253441

While I don't want to turn this into another Apple thread, this is somewhat relevant. Apple and Intel struck an exclusive contract a few years ago. Even if that contract is expired, I expect Intel to bend over backwards for Apple. Apple is currently the biggest PC OEM out there. Intel is likely selling chips close to cost just to keep Apple happy. Anyone that thinks Apple is moving to ARM/AMD is dreaming.

Apple will continue releasing iPads with ARM, while enhancing functionality. I do not expect them to replace the processors used in macOS based systems with ARM CPUs. While it would give them end-to-end control, they don't need end to end control of the Mac. However, what I do expect them to do is work on making the Mac obsolete via the iPad and other upcoming devices.

All that being said, AMD certainly is throwing an interesting wrinkle into things with their Ryzen Mobile products. The 4800H, 4800HS , and 4900HS chips are significantly faster than anything available in a Macbook currently. The 4700U is very competitive with top end as well. Shoot, the 4800U looks like it is dead even with the best CPU available for the Mac.

 

[Thunder 57]

While I don't want to turn this into another Apple thread, this is somewhat relevant. Apple and Intel struck an exclusive contract a few years ago. Even if that contract is expired, I expect Intel to bend over backwards for Apple. Apple is currently the biggest PC OEM out there. Intel is likely selling chips close to cost just to keep Apple happy. Anyone that thinks Apple is moving to ARM/AMD is dreaming.

Apple will continue releasing iPads with ARM, while enhancing functionality. I do not expect them to replace the processors used in macOS based systems with ARM CPUs. While it would give them end-to-end control, they don't need end to end control of the Mac. However, what I do expect them to do is work on making the Mac obsolete via the iPad and other upcoming devices.

All that being said, AMD certainly is throwing an interesting wrinkle into things with their Ryzen Mobile products. The 4800H, 4800HS , and 4900HS chips are significantly faster than anything available in a Macbook currently. The 4700U is very competitive with top end as well. Shoot, the 4800U looks like it is dead even with the best CPU available for the Mac.

Don't forget the newly leaked 4900U

 

[uzzi38]

Meanwhile Apple keeps adding Ice-Lake CPUs o Macos beta. I'm still not convinced the AMD apu will actually be used this soon:

https://twitter.com/x/status/1247934672648253441

Honestly neither am I. Not when the prospect of an ARM Macbook next year is extremely real, and the chip in question could very well extend to MBPs.

 

[uzzi38]

Don't forget the newly leaked 4900U

Doesn't exist apparently. That long Chinese forum post about Renoir I linked in the G14 thread talked about how the 4900U got a name change and became the 4800U. The 4600U was a last minute addition before CES.

 

[darkswordsman17]

While I don't want to turn this into another Apple thread, this is somewhat relevant. Apple and Intel struck an exclusive contract a few years ago. Even if that contract is expired, I expect Intel to bend over backwards for Apple. Apple is currently the biggest PC OEM out there. Intel is likely selling chips close to cost just to keep Apple happy. Anyone that thinks Apple is moving to ARM/AMD is dreaming.

Apple will continue releasing iPads with ARM, while enhancing functionality. I do not expect them to replace the processors used in macOS based systems with ARM CPUs. While it would give them end-to-end control, they don't need end to end control of the Mac. However, what I do expect them to do is work on making the Mac obsolete via the iPad and other upcoming devices.

All that being said, AMD certainly is throwing an interesting wrinkle into things with their Ryzen Mobile products. The 4800H, 4800HS , and 4900HS chips are significantly faster than anything available in a Macbook currently. The 4700U is very competitive with top end as well. Shoot, the 4800U looks like it is dead even with the best CPU available for the Mac.

I think Intel will try but I'm not sure that's enough, as its not simply a matter of cost/money (well it does but there are other costs that matter). Development for a single GPU family might have value. I also think Intel's behavior on cellular modems (dropping 5G) might possibly sour Apple as they definitely want to have some leverage over Qualcomm. Heck, they've supposedly been toying with AMD CPUs for years (think they even had dev systems testing out Excavator based systems), partially to keep Intel from getting too comfortable. Now obviously they didn't make the change, but I could see Intel's recent issues and AMD's success to be possibly appealing enough for Apple. EPYC/Threadripper would have made the new Mac Pro much more interesting. And it might become important to have Infinity Fabric between CPU and GPU in the future.

Yeah, I think the iPad line starts becoming a 2 in 1 type of device situation (namely the Pro line), and I'd guess that's how they transition. I've thought the Air and iPad Pro line will converge, and then possibly the standard Mac, maybe they bring back iBook branding?

Yeah, that's got to have Apple interested.

Which, random speculation, any possibility that Apple might make a gaming console? It'd probably be marketed more as a general media device, think something between the AppleTV and Mac Mini. But it'd maybe run multiple OSes or something, like a base MacOS layer, with a Metal based gaming environment for max performance, with a GUI overlay. Kinda like what Microsoft does on the Xbox. But maybe this would be something like Nvidia did with their BFG TVs which had Shields. Apple was developing a smart TV, maybe they changed things up (or maybe are still working on something like that)?

 

[uzzi38]

With the virus in full swing, AMD themselves likely don’t know the launch date.

Nah, they got a pretty good clue as far as I'm aware. In fact, there should still be another launch prior to Zen3 by a while. Rumour has it B550 and desktop Renoir are landing in June.

I have good reason to believe it's the case.

 

[Shivansps]

Nah, they got a pretty good clue as far as I'm aware. In fact, there should still be another launch prior to Zen3 by a while. Rumour has it B550 and desktop Renoir are landing in June.

I have good reason to believe it's the case.

They discontinued the 3400G, Desktop Renoir has to be closer than "end of the year".

 

[DisEnchantment]

Nah, they got a pretty good clue as far as I'm aware. In fact, there should still be another launch prior to Zen3 by a while. Rumour has it B550 and desktop Renoir are landing in June.

I have good reason to believe it's the case.

I believe also the consoles also have a lot to do with how AMD's plans got shaped.
AMD will have to ship like 8-9 million chips (being conservative) to Sony and MS for initial launch by Sep (there has to be enough lead time for MS and Sony to assemble, test, package and ship etc.), which will consume more than two months of their wafer allocations of 30K wpm (from H2) and then maintain a steady output thereafter.
I think they already thought about this like several quarters ago, so I don't think they could have launched Zen3 earlier even if they wanted to or could.
So most of the Zen2 stuff they plan to sell would need to be stockpiled within these months.

 

[Glo.]

Honestly neither am I. Not when the prospect of an ARM Macbook next year is extremely real, and the chip in question could very well extend to MBPs.

ARM based stuff is only for Apple services access, just like Chromebooks. There is no software for ARM MacBooks to extend it to x86 laptops.

Nah, they got a pretty good clue as far as I'm aware. In fact, there should still be another launch prior to Zen3 by a while. Rumour has it B550 and desktop Renoir are landing in June.

I have good reason to believe it's the case.

Pretty much this.

 

[amrnuke]

They discontinued the 3400G, Desktop Renoir has to be closer than "end of the year".

If they have discontinued the 3400G, it's somewhat odd since there are tons of 3400Gs in stock at my local MicroCenter, and when the number dips below 10+ in stock, it gets replenished to 10+ fairly quickly over the past month or two. I think it's possible MC are sitting on a lot of warehouse stock, or it's possible AMD didn't actually discontinue it.

Since they are 12nm parts, they are GloFo, right?

 

[DrMrLordX]

Since they are 12nm parts, they are GloFo, right?

Yup.

 

[uzzi38]

ARM based stuff is only for Apple services access, just like Chromebooks. There is no software for ARM MacBooks to extend it to x86 laptops.

It is a very real prospect.

This was my line of thought until recently as well, but I would not be surprised in the least if ARM Macbooks got the go-ahead as early as next year. Not in the slightest.

 

[Glo.]

It is a very real prospect.

This was my line of thought until recently as well, but I would not be surprised in the least if ARM Macbooks got the go-ahead as early as next year. Not in the slightest.

The moment Apple switches their laptops to ARM, they kill their computer entirely.

There is no software for Apple ARM based computers. The program that is allowing launching iPad Apps on Mac OS is something different from porting whole software stack from x86 to ARM on desktop and laptop computers.

ARM based laptop and Desktop will be limited only to running software ports from iPad to the OS it will use.

Apple ARM CPUs if they want to replace x86 HAVE TO be better that x86_64 in a wide variety of solutions. If Apple cannot scale IPC of their CPUs for clock speeds, and power, what is the benefit?

We have to rememeber, AMD has just released 7 nm APUs that fit in 15W thermal envelopes. Dual Core, dual thread CPU from Apple fits in 3W envelope with 2.6 GHz. If we scale it by 4 times, to get 8 core, 8 thread CPU, we get 12W, 2.6 GHz CPU. But what about 35W TDP's? What about 3.5 GHz clock speeds?

There is way too much of burdens to overcome to even consider ARM based Macs that head is starting to hurt if you think about it.

For now, ARM based Mac can be ONLY Apple Services Access machine Only. For now. But will it really be better solution than x86_64 in the long run? Im VERY sceptical.

 

[jpiniero]

There is no software for Apple ARM based computers. The program that is allowing launching iPad Apps on Mac OS is something different from porting whole software stack from x86 to ARM on desktop and laptop computers.

ARM based laptop and Desktop will be limited only to running software ports from iPad to the OS it will use.

Apple is pushing their dev stack to be arch-neutral.

BTW They don't necessarily need to abandon x86 entirely.