Ryzen 5 2400G & Ryzen 3 2200G APU (with Vega GPUs) review kits are being sent out

[Taf the Ghost]

4 minutes ago, Master Disaster said:

So it's a product that would make very little sense I know but is there a possibility of a Threadripper APU?

 

So 2 2400Gs glued together, maybe even with XFire on the GPUs?

 

Yeah it's dumb as all hell but theoretically could it be done?

It's drivers, actually, more than physical issues. No one has ever really spent the time getting "opaque" mGPU drivers working. That has a lot to do with, basically, needing a layer between the mGPUs to operate, which pretty much means you're doing a lot of cross-talk bandwidth and latency. Need actual SLI to return, haha.

[AlTech]

Can we start acting maturely and stop comparing products to the size of a banana? Thank you .

[Trixanity]

2 minutes ago, Master Disaster said:

So it's a product that would make very little sense I know but is there a possibility of a Threadripper APU?

 

So 2 2400Gs glued together, maybe even with XFire on the GPUs?

 

Yeah it's dumb as all hell but theoretically could it be done?

I suspect there's a bigger chance of an EPYC APU of some kind for some sort of HPC workload. I think TR is mainly just massive amount of cores for users who'll use discrete graphics anyway. I'm not even sure the EPYC APU would make much sense since a discrete card would ultimately be able to achieve much higher performance. Even if it had HBM2 onboard (as previous leaked slides hinted at), an EPYC CPU paired with something like MI25 (Vega for enterprise) would still make a lot more sense I think. Some of the guys who work in the business might be able to say more about whether a HPC APU makes any sense. I know we have some on the forum.

[Taf the Ghost]

5 minutes ago, Trixanity said:

Yeah, the unboxing I've seen were just that. They said they already begun testing it but won't even say if it's good (even though that'd be in AMD's favor to do so). So Linus might say he has it on the bench and even show the box but he won't be able to talk about benchmarks at all.

You really think they could fit 8 croes and 20 CUs without significantly increasing die size even on 7nm?

Yes. 7nm is about 1.5 Node Shrink. ~56% area decrease at the same transistor count. Which would be enough to fit a Zepplin die vertically and most of a RX580 die right next to it.

 

However, I expect the Zen2 Core to increase in size, relative to the shrink for further AVX & other units, and we don't know how well Vega will shrink. That's why 8c & 20 CUs sounds about as big as I think they could do at a solid size. Current RX560 with 16 CUs is ~123 mm². Node shrink to 7nm should get that under 60 mm², but whatever "Vega on 7nm" looks like will dictate how much of an actual shrink it is.

 

Which is why I could see AMD approving a small mobile, big mobile or both designs for 7nm. It depends how far much of a shrink they can achieve with each part and what they think will happen in the market. I could see AMD doing another 4c w/ ~13 CU design for Zen2 APUs to fill in behind them just like they've done for Zen1 generation. They could come in with a ~140 mm² that's 4c and something silly like a 10w Thermal Design.

[Sunshine1868]

InB4 @LinusTech compares it to an incomparable (and overpriced) intel chip and then recommends you buy the intel part.

 

Really though, these will be awesome for people who want a budget gaming build - the specs people are listing out really put this thing as a one-stop-shop for games that arent hyper-demanding and even have the makings of a great basic workplace CPU (APU)

 

Post sponsored by Intel

[ItsMitch]

4 hours ago, Misanthrope said:

Well seeing how now even the 1050 might get affected (Probably not due to mining but so many people buying the one and only option) it would be just about the best time for AMD to take such venture.

 

Their prices have been fucked for a few weeks now, it's sad. 

[Mangerinne]

8 hours ago, Agost said:

DDR4-2933 as an official ram clock? Is it legit?

I think it resists up to 3600mhz

[Okjoek]

I want a 2400G

[Sunshine1868]

18 minutes ago, Okjoek said:

I want a 2400G

I am absolutely going to build myself a work machine with a 2400G. itll be a HUGE upgrade from my i7-4790 + Intel HD Graphics 4600

[Dabombinable]

59 minutes ago, Mangerinne said:

I think it resists up to 3600mhz

Ok, that's GTX 650ti territory. And I have found that my GTX 650ti plays pretty much any game at the correct settings (it does have 2GB GDDR5 6GHz to help it though).

[Citadelen]

9 hours ago, Noyu said:

 

MY BAD UPON MORE GOOGLE-FU YOU WERE RIGHT!!

 

Zen (14nm) = Ryzen 1 and the Ryzen APUs  (2200G and 2400G)

Zen+ (12nm) = Ryzen 2

Zen 2 (7nm)

 

 

While talking to an AMD rep (someone who I can't remember the name or channel of), asked about the architecture of the APUs considering it was using the 2XXX naming scheme; he said that the architecture was the same. This means the only difference between the APUs and Ryzen 2XXX is the process node.

[Energycore]

5 hours ago, wcreek said:

So like I wonder if a 460 or 550/560 would be able to run in that sort of mode.

Not crossfire per se as RX 460/550/560 are Polaris and not Vega, but in games that support Explicit Multi-Adapter like AOTS

[cj09beira]

@Taf the Ghost so you think they will stick with 4c ccxs?

 

 

[TVwazhere]

50 minutes ago, Sunshine1868 said:

I am absolutely going to build myself a work machine with a 2400G. itll be a HUGE upgrade from my i7-4790 + Intel HD Graphics 4600

You could just add a discrete GPU you know... Unless you're trying to build a super SFF pc, you're not actually going to see that big of a difference, as a lot of the Ryzen lineup is compared to the Haswell lineup (also jsut saw it's a non K CPU.... might be a decent upgrade but dont expect huge improvements)

[mr moose]

It's no wonder you guys a messed up over there, not only are you clinging to the outdated imperial system but you are now trying to introduce fruit as a metric?

 

How big is the New APU:

 

Australia - 70mm across

US - half a furlong and a banana.  So just under a one 10th of a link or part of an orange.

 

 

[Blademaster91]

42 minutes ago, mr moose said:

It's no wonder you guys a messed up over there, not only are you clinging to the outdated imperial system but you are now trying to introduce fruit as a metric?

 

How big is the New APU:

 

Australia - 70mm across

US - half a furlong and a banana.  So just under a one 10th of a link or part of an orange.

 

 

The picture looks like it was taken by BitWit aka Kyle,the banana for scale is a meme.

But of course someone has to rant about those "messed up" americans and their imperial system that actually use both and more of metric when using any tools.

[RotoCoreOne]

If you performance is on par with it's non Apu counterpart I'd definitely be interested in building an itx system with this. Just wish they came with the wraith spire rgb cooler

[cj09beira]

3 minutes ago, RotoCoreOne said:

If you performance is on par with it's non Apu counterpart I'd definitely be interested in building an itx system with this. Just wish they came with the wraith spire rgb cooler

perf should be better and also cheaper than non apu versions, as its zen+ on 14nm with higher clocks than the non apus

[mr moose]

1 hour ago, Blademaster91 said:

 

But of course someone has to rant about those "messed up" americans and their imperial system that actually use both and more of metric when using any tools.

 

If I was American I wouldn't get to defensive about it. You might actually invite real criticism rather than just tongue in cheek humor.

[Mangerinne]

4 hours ago, Dabombinable said:

Ok, that's GTX 650ti territory. And I have found that my GTX 650ti plays pretty much any game at the correct settings (it does have 2GB GDDR5 6GHz to help it though).

I would really like to see a gameplay of the Ryzen 5 2400G with a ram of this speed: 3600mhz-3200mhz to compare the performance and the difference of FPS.

Edited February 9, 2018 by Mangerinne

[ItsMitch]

2 minutes ago, Mangerinne said:

I would really like to see a gameplay of the Ryzen 5 2400G with a ram of this speed: 3600mhz-3200mhz to compare the performance and the difference of FPS.

Quote

 

WHY IS YOUR TEXT SO BIG DO YOU LIKE TO SHOUT A LOT? ALSO I CAN'T DELETE THE BLANK QUOTE THANKS LTT

[Sunshine1868]

4 hours ago, TVwazhere said:

You could just add a discrete GPU you know... Unless you're trying to build a super SFF pc, you're not actually going to see that big of a difference, as a lot of the Ryzen lineup is compared to the Haswell lineup (also jsut saw it's a non K CPU.... might be a decent upgrade but dont expect huge improvements)

Work Machine does not equal "budget gaming rig" - I'd do it for the NVMe boot drive that I could put in it as well as the 16GB of fast RAM. Neither this future machine i'm talking about nor the Dell OptiPlex that currently sits under my desk will ever have a game on them.

 

...but I forgot this is the LTT forum and all we care about are frame rates and K-SKUs

[Mangerinne]

8 minutes ago, SC2Mitch said:

WHY IS YOUR TEXT SO BIG DO YOU LIKE TO SHOUT A LOT? ALSO I CAN'T DELETE THE BLANK QUOTE THANKS LTT

Idk, i copy the text from google traslate maybe was that`s it

[Taf the Ghost]

7 hours ago, cj09beira said:

@Taf the Ghost so you think they will stick with 4c ccxs?

 

 

For Zen2, almost assuredly. Zen3 might see a change there, but the impression I get is that the CCX is more tightly designed together than people realize. The funny thing about the CCXs is that it's actually an evolution of the Bulldozer design. How? The current Zeppelin die (ryzen, TR & epyc) is actually a 2 core design. The CCXs should be thought of as a single core, from an overall design perspective. The Dozer Disaster happened because the CMT had some nasty interior-to-the-CMT bottlenecks. Removing those is what bumped Zen up by 52% in IPC.

 

From that perspective, it's better to keep the CCX as tightly wound together as possible and get the IF bridge latency faster each generation. As Nodes get smaller, they can add more CCXs, since the Zen Design is really more like a boosted 8-way SMT per CCX. x86 just can't be that Wide, so the IBM approach isn't needed. This is also why putting more CCX on a die is actually the better approach for now. We'll see what happens when they get to 4 CCX on a die. The server parts of Zen scale by the Binary sequence, which is the reason it's 4 dies on Epyc and you can only currently associate 2 Epycs together, as there is connections to each of the Zen packages. But those packages can fit more CCXs, but can they go more than 4? No clue.

[Carclis]

This might be the chip that saves gamers from the miners. It's reliance on system memory should make it a poor performer in mining whilst still having very decent performance for an iGPU.