Post your thoughts and theories on what a 4200G and a 4400G might be like (construction, architecture, GPU etc). More cores? What clock speeds? What TDP?
Post your thoughts and theories on what a 4200G and a 4400G might be like (construction, architecture, GPU etc)...
Owen Hernandez
Other urls found in this thread:
anandtech.com
twitter.com
Jonathan Thompson
The 3000 series is not even out yet.
Jayden Reed
We already know what it is though. I'm interested in some educated guesses on what Zen 2 + Navi APU's would be like.
Liam Mitchell
4200g and 4400g?
Total flop. I got disillusioned with 3400g release and the fact that it has no pcie 4.0. I'm going intel.
Xavier Perry
Fuck off Adored. We won't be your unnamed sources.
Christopher Thomas
Monolithic die, 4 cores, maybe 16CU Navi IGP.
IGPs aren't going to grow and waste silicon space when memory bandwidth is still so monstrously limited. These are first and foremost low power mobile chips, and AMD will not alienate their core market segment by radically increasing costs with complex packaging and HBM.
OEMs have had 3000 series chips for a few months.
>expecting PCI-E 4.0 from a Raven Ridge refresh
Dummy
Angel Hall
Why would a budget chip have pcie 4? Which Intel chip at that range has it?
Charles Foster
What a stupid thread. Just compare equivalent Zen+ and Zen2 chips and add the difference on top of whatever the performance of the 3000 series APUs is. Everything else will follow the same formula, it's one generation behind.
Luke Clark
>OEMs have had 3000 series chips for a few months.
No. Nobody had desktop APUs, which is what the thread is about. Don't lie.
Julian Wilson
>Why would a budget chip have pcie 4?
Because we'll have to pay premium for x570 boards. Also nobody is going to pay "premium" for AMD. PCIE and APUs are the only gimmick people cared about. I was shilling AMD for the past year but I had enough now.
Elijah Davis
>16CU Navi IGP
Not happening. 12 at higher clocks at best because of memory bandwidth limitations.
Hunter James
The mobile chips launched at CES, retard. The desktop chips are just repacked mobile chips.
Dylan Miller
>4 cores
No please!
I long for a 6 core APU.
Gavin Murphy
3400G was leaked for real way before the desktop series.
And the mobile APUs were shipped to OEM integrators probably half a year ago. We'll have products shortly. Not like OEM or retail CPUs, I mean entire laptops. We're talking several months to integrate and a few months to build stock for availability. They've had them for a while.
It's going to be Zen 2 if the current trend follows. 2000 APUs were Zen/Ryzen. 3000 is confirmed Zen/Ryzen+. So 4000 will probably be Zen 2. That means 4 cores, up to 4.5 or 4.6 singel core turbo probably. I'm going to bet on Navi since it'll have to be monolithic and hence 7nm, and there's no point in sticking with the less efficient GCN gen 5/Vega if you're on 7nm anyway. Although they did produce some APUs in the past on different process than the desktop counterpart. IIRC Kaveri or something was like that.
It should be pretty good IMO. Although Intel is now bringing heat with 10nm and 2.5x larger GPUs for mobile by end of the year. They're skipping desktop sure, but moving that focus onto laptops instead.
Nathaniel Gonzalez
>expecting PCI-E 4.0 from a Raven Ridge refresh
Wait... didn't AMD initially promise pci-e on 400 series motherboards? Doesn't that mean it depends on motherboards rather CPUs, and then backed out? Something doesn't add up here. And if it depends on CPU couldn't they add that to the "refresh" which is a smaller production process than 2400g anyways?
Smells of designated street from a mile, here.
Nathaniel Hill
there wont be zen2 APU
Mason Jones
it's probably a year away, after all. I wonder if they'll sell a cut down version of whatever they're designing for MS/Sony
Kayden Ross
>Although they did produce some APUs in the past on different process than the desktop counterpart. IIRC Kaveri or something was like that.
AMD originally planned to move the whole Construction core family over to bulk after 32nm PD-SOI, but they canceled all desktop parts after Piledriver. Kaveri was fabbed on a customized SHP variant, later Carrizo and Bristol Ridge were a cheaper generic bulk process.
Its not like they decided to use a different process just for the APUs, its that they canceled all future desktop parts, and were using cheaper silicon than expensive SOI wafers.
Aiden Sanchez
4C4T and 4C8T again.
There'll probably be another model with at least 6C12T, maybe Ryzen 5 3700G or something.
Brayden Harris
They aren't adding cores to the CCX, and they aren't increasing die area by a full CCX for something that'll end up in 15w laptops. You're not getting a 6 core Zen2 APU.
Jack Turner
8c 16t @ 5 Ghz
Grayson Gonzalez
I'm hoping for less power consumption, yet more unf. Something I can put in my Antec ISK 110 that can run something semi-new at 1080 medium settings and won't blow the 90W PSU.
Why would the Zen+ APU's suddenly have pcie4.0 when the other Zen+ CPU's don't?
Lincoln Nelson
The question is why would you want PCI-E 4.0 on an APU at all. They're not high end workstations, no sane person is going to put $1000 worth of high end NVMe SSDs in a budget system, no one is going to put two high end GPUs in one either.
Its just concern trolling.
Cameron Diaz
>The question is why would you want PCI-E 4.0 on an APU at all.
Future proofing.
>high end GPU
Soon there will be PCI-E 5.0 GPUs. Nobody buys a computer for a year or two and so far GPUs age much faster than CPUs. And we don't know if NVMe SSD won't be the standard soon thanks to China.
Adrian Johnson
>I need to future proof my $99 desktop!
Its funny that you thought this was a legitimate argument. The GTX 2080ti does just fine on a PCI-E 3.0 X8, it only has a 2-3% marginal performance reduction vs a full X16 slot.
PCI-E 4.0 does literally nothing for GPUs that aren't utilizing the bandwidth. That goes doubly so for 5.0, you fucking moron.
You aren't buying a budget system and putting two $1000+ GPUs in it. You aren't buying super high end SSDs. You don't have a single legitimate complain. You are concern trolling because you're a nigger.
Kayden Bell
Makes sense
I'm too tired from work to pick up on the trolling
Chase Flores
And because the performance reduction is so small from halving the effective bandwidth, we can readily see that it would take a GPU roughly 2X the performance of the 2080ti to actually make use of PCI-E 4.0. To fully saturate a 3.0 X16 connection would require a GPU that powerful. Thats not happening any time soon.
Kevin Smith
Possibly a different I/O die from Matisse to handle the GPU chiplet as well. We'd be getting 8 cores + 12-16 Navi CUs at a 65W TDP. Very exciting stuff, but what we really want is for AMD to implement vertical stacking. They're already going wide with chiplets, it's only logical to go tall this time around. Think of an APU with stacked DRAM for both the CPU and GPU, it's going to be insanely powerful. The first one of a kind is probably going to eliminate the mid-range market of GPUs; think of RX 5700XT on an APU in 2 years.
David Barnes
What amd need is something to confront nvidia in the server market. IBM servers with nvlink are insanely powerful.
Carson Barnes
7nm APUs aren't MCM.
Zachary Richardson
How do you replace the hard drive if it’s embedded in the mcm?
Xavier Butler
You wouldn't.
Jacob Roberts
it will be 8 cores.
it will be chiplets like matisse.
Jason Edwards
No, Adored, it won't.
APUs are low cost solutions, not premium performance solutions.
Its a monolithic die, fed by system RAM, no HBM, no massive 20CU IGP, and it will not feature two CCXs. The number one market AMD targets with these chips is entry level 15-25w laptops. The second market is SFF OEM systems like thin clients and POS machines.
You don't drive up base costs with premium hardware in these market segments.
William Carter
This is current mobile Ryzen. This package is socket FP5.
The next generation of APUs will fit this same BGA socket.
AMD is not putting a full MCM desktop package inside of thin cheap laptops. The 7nm APU will be a single monolithic die fitting in this same BGA FP5 package.
Reminder that if you watch adoredtv and regurgitate any of his stupid lies you should kill yourself promptly.
Brody Adams
The last rumours hinted towards having stacked memory on the IO die. An IO die with ~20cu navi and 2/4/8GB of stacked memory at hbm speeds isn't that farfetched an idea if they can fix stacking and use 7nm. Die size is an issue I see though.
Jonathan Taylor
Didn't the 28nm carrizo clock pretty well on 28nm? It would have been nice to see an 8+core version of that with a cache that wasn't crippled.
Logan Jackson
They clocked moderately well inside of their power envelope. In terms of ultimate clocks they never got never Bulldozer or Piledriver. Still some people get 4.7-4.8ghz with questionable stability. Anandtech included it in their article on Bristol Ridge way back when.
anandtech.com
Luis Turner
>APUs are low cost solutions, not premium performance solutions.
Exactly. These chiplets by then will have insurmountable yields and the I/O die is already on a very mature process. Why the fuck would they design another monolithic die when you can tweak your I/O die to handle a GPU chiplet as well?
Tyler King
Another retard Adored fanboy.
These chiplets aren't going to have super high yields, and the packaging itself if far more expensive. Using an MCM right off the bat incurs more losses due to issues with assembly.
The 7nm APUs are still on FP5. One single monolithic die, moderate in size, low power, low cost.
Xavier Morris