[SiSoft] AMD monster APU?

whm1974

Diamond Member
Jul 24, 2016
9,460
1,570
96
If it isn't, then I doubt that this chip will be anywhere near cheap or even cost effective to buy.
 

wahdangun

Golden Member
Feb 3, 2011
1,007
148
106
Hmm the total cu match Rx 570D, but only 2 GB and 188 Gb/s.

And what is that and fanghuang ? Is this China semi custom deal?
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
I doubt this is an APU. It could be Zen CPU + Vega 28 + 2 GB HBM2 in a single package like Kaby G. It will be interesting to see in what market segment AMD wants to sell this combo.
 

ElFenix

Elite Member
Super Moderator
Mar 20, 2000
102,414
8,356
126
I doubt this is an APU. It could be Zen CPU + Vega 28 + 2 GB HBM2 in a single package like Kaby G. It will be interesting to see in what market segment AMD wants to sell this combo.
psxbox one five niner
 
  • Like
Reactions: dark zero

Glo.

Diamond Member
Apr 25, 2015
5,661
4,419
136
It is 4C/8T design, according to leaks.

So no, it is not Console APU, nor it is MCM design. It is massive APU, with 28 CUs, and single CCX.

A direct response from AMD for Kaby Lake-G, but done better.

One more thing. Remember Ryzen Gaming in AMD Roadmap, for Mobile Premium?
 
Last edited:

ksec

Senior member
Mar 5, 2010
420
117
116
Put this in the next Mac Mini and my cash is ready.

The only problem / concern, 2GB HBM?
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
It is 4C/8T design, according to leaks.

So no, it is not Console APU, nor it is MCM design. It is massive APU, with 28 CUs, and single CCX.

A direct response from AMD for Kaby Lake-G, but done better.

One more thing. Remember Ryzen Gaming in AMD Roadmap, for Mobile Premium?

Or it could just be a really expensive proof of concept to try to steal away Apple from Intel basically trying to say anything mobile we can do better.

But yeah it would be a pretty interesting Mobile only, thin and light, gaming solution.
 

Glo.

Diamond Member
Apr 25, 2015
5,661
4,419
136
Hmm it's the 24CU's that bother me. RR is 11cu's with on disabled on the 2700 right? So Snowy Owl would only have 22CU's right?
RR has 11 CUs and that is full design, with 10 CUs available in 2700.

Snowy Owl APU is Raven Ridge.

This thing is something different, targeted at higher price margin.

Remember - Raven Ridge will be next gen. Ryzen 3 lineup. Ryzen 5 are going to be High Clocked 4C/8T, 6C/12T designs, with similar patter to this years.
 

Glo.

Diamond Member
Apr 25, 2015
5,661
4,419
136
If the clocks are conservative enough like 3.2 GHz for the CPU, and 1250 MHz for the GPU it may be possible to pack both of them into 95W TDP power envelope.
 

Topweasel

Diamond Member
Oct 19, 2000
5,436
1,654
136
RR has 11 CUs and that is full design, with 10 CUs available in 2700.

Snowy Owl APU is Raven Ridge.

This thing is something different, targeted at higher price margin.

Remember - Raven Ridge will be next gen. Ryzen 3 lineup. Ryzen 5 are going to be High Clocked 4C/8T, 6C/12T designs, with similar patter to this years.

Snowy Owl is and always has been a Zeppelin based part.
Two standard Zeppelin dies in MCM configuration, on SP4r2 BGA package.

https://cdn.wccftech.com/wp-content/uploads/2017/05/AMD-Snowy-Owl-Block-Diagram-1-740x411.jpg

I thought Snowy owl was always supposed to be an MCM APU for servers. We only thought that TR was snowy ridge for the first couple months of trying to figure out where TR slotted in (which it didn't because I don't think AMD was certain that they were going to release it until Ryzen got accepted as well as it did).

I get a forbidden when trying to click that link.

Edit: I did find the article and looks you are probably right if that slide is right. I think the use of snowy owl tricked people because they were using bird names for most of the APU related parts.
 

jpiniero

Lifer
Oct 1, 2010
14,510
5,159
136
Snowy Owl is and always has been a Zeppelin based part.
Two standard Zeppelin dies in MCM configuration, on SP4r2 BGA package.

https://cdn.wccftech.com/wp-content/uploads/2017/05/AMD-Snowy-Owl-Block-Diagram-1-740x411.jpg

IIRC there were rumors of a model with a Summit Ridge die + Vega 20 + HBM2. Vega 20 seems to be MIA, possibly cancelled so I don't know if that model will actually get released. But I suppose AMD could have reused the product and just thrown in Vega 12 or something.

The question of course is whether it's really worth it and what OEM will actually want this.
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
IIRC there were rumors of a model with a Summit Ridge die + Vega 20 + HBM2. Vega 20 seems to be MIA, possibly cancelled so I don't know if that model will actually get released. But I suppose AMD could have reused the product and just thrown in Vega 12 or something.

The question of course is whether it's really worth it and what OEM will actually want this.

If this thing is a real deal, then I'm quite certain that it is a semicustom design.

None of the Zen compatible infras (AM4, SP3, SP4r2 or FP5) are compatible with a large GPU. AM4 & FP5 packages which support GPU are too small and have too weak power delivery in terms of the package.
A real leak would require that a compatible infra would have existed for a while, since obviously they don't emerge over night. I'm not aware of any other infras, besides the ones listed earlier.
 
  • Like
Reactions: coercitiv

CluelessOne

Member
Jun 19, 2015
76
49
91
The benchmark stated that it is a Server/Workstation platform. So assuming current AMD sockets for Threadripper (TR4) or EPYC (SP3?) what does that mean?
Max TDP of 180 Watt like Threadripper?
8 channel DDR4? How does that work for a 1 (one) CCX CPU?
What kind of problems does this solve in Server or Workstation market?

Edit:
I just had a crazy thought. I am assuming it is a CPU with a GPU attached. What if it is the other way around? A GPU with a CPU attached? A test for accelerator card add in? How does that even work?
 
Last edited:

wahdangun

Golden Member
Feb 3, 2011
1,007
148
106
Imagine 2 of this in mcm, it will a killer platform, maybe named it TR Extreme Gamer ?

Or used to create mining farm?
 

itsmydamnation

Platinum Member
Feb 6, 2011
2,744
3,077
136
If this thing is a real deal, then I'm quite certain that it is a semicustom design.

None of the Zen compatible infras (AM4, SP3, SP4r2 or FP5) are compatible with a large GPU. AM4 & FP5 packages which support GPU are too small and have too weak power delivery in terms of the package.
A real leak would require that a compatible infra would have existed for a while, since obviously they don't emerge over night. I'm not aware of any other infras, besides the ones listed earlier.

If this thing is real

Given that a 2700 u is 10cu 15-25 watt tdp, they could do 24-26 Cu in 45-60watt. It would be pretty big but, 330-350mm sq.

Vega is quite efficient when you keep the clocks in the lowish 1GHz. With add hbm and the hbcc it would certainly be a threat to nv mobile descrete. It's gpu bom could actually be cheaper, small interposer small stack, vs 4-8 gb gddr5....

I might do some Vega 56 based tests see what clock I can get in around 100-120 watt mark
 

raghu78

Diamond Member
Aug 23, 2012
4,093
1,475
136
Videocardz confirming a desktop APU with 4c/8t and 28CU Vega with 2 GB HBM2 cache.

https://videocardz.com/74464/amd-preparing-mobile-ryzen-5-apu-with-vega-11-graphics

This proves a lot of people wrong who were claiming HBM2 could never make it to mainstream high volume APUs in 2018. Now we can understand the reason behind Samsung's massive USD 26bn capex for 2018 of which USD 7bn is DRAM. With both Intel and AMD producing HBM2 based products in high volume somebody needed to ramp HBM2 capacity in a big way and thats Samsung.

https://www.electronicsweekly.com/b...ungs-2017-capex-may-kill-competition-2017-11/

Sent from my SM-G935V using Tapatalk
 

Yotsugi

Golden Member
Oct 16, 2017
1,029
487
106
This proves a lot of people wrong who were claiming HBM2 could never make it to mainstream high volume APUs in 2018.
Yes.
Acktually, the only problem with HBM2 was ramp. Back then there was only P100, now there's many.
Now we can understand the reason behind Samsung's massive USD 26bn capex for 2018 of which USD 7bn is DRAM.
Well, they also need to ramp DDR4 production.
The prices for memory are insane right now.
With both Intel and AMD producing HBM2 based products in high volume somebody needed to ramp HBM2 capacity in a big way and thats Samsung.
Ye, but where's Hynix?