[Anandtech] AMD SoC with 24 Vega CUs for Chinese Consoles (Where is Vega Mobile?)

PeterScott

Platinum Member
Jul 7, 2017
2,605
227
96
#1
https://www.anandtech.com/show/1315...zen-soc-with-24-vega-cus-for-chinese-consoles

"The announcement states that a custom chip has been created for Subor that is based on four Zen cores running at 3.0 GHz and 24 compute units of Vega running at 1.3 GHz. The chip is supported by 8GB of GDDR5 memory"

First a semi-custom GPU for Intel, and now a custom Zen/Vega SoC for China consoles.

So where is that Vega Mobile chip, that seemed to have been announced months ago?
 

Kenmitch

Diamond Member
Oct 10, 1999
7,167
110
126
#3
Google it. There's multiple mobile offerings with Vega inside already. I guess you don't like what's offered currently? Maybe you could pony up a couple million and have one custom designed that meets your criteria.
 

Kenmitch

Diamond Member
Oct 10, 1999
7,167
110
126
#5
OP is talking about the discrete GPU, not Raven Ridge.
So was I. There's more than one laptop that already has a discrete Vega GPU in it already. It doesn't take much google action to find them. Just search AMD vega mobile and you'll see them. Somehow the OP missed them I guess.
 

jpiniero

Diamond Member
Oct 1, 2010
6,430
273
126
#6
So was I. There's more than one laptop that already has a discrete Vega GPU in it already. It doesn't take much google action to find them. Just search AMD vega mobile and you'll see them. Somehow the OP missed them I guess.
Looks like there might be one of those crazy desktop replacement laptops with the desktop Vega 56. But that's not Vega Mobile. There's one with the desktop 580 too.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
227
96
#7
So was I. There's more than one laptop that already has a discrete Vega GPU in it already. It doesn't take much google action to find them. Just search AMD vega mobile and you'll see them. Somehow the OP missed them I guess.
I was talking about this:
https://www.anandtech.com/show/1223...led-with-ryzen-apus-zen-on-12nm-vega-on-7nm/9
As far as AMD were prepared to announce, a single Vega Mobile design is being prepared, which might be subject to binning based on compute units. The full amount of compute units in the chip has not been made publicly available, however it measures basically half of the full Vega 64 chip next to it. This would normally suggest a 32 CU chip, however there are a number of front-end elements required which have a fixed size – compared to the Intel 24 CU chip for the Intel with Radeon Graphics processor, this chip seems to be of similar size, if not bigger.

What we do know about the Vega Mobile processor is that it is paired with 4GB of HBM2, using an embedded interposer to give an overall z-height of 1.7mm. This is AMD’s second or third generation of interposer technology, and it has come a long way since Fiji: the interposer is tight along the dimensions of the GPU and HBM, but it is also embedded into the PCB. The embedding allows it to reduce the height, making it a better fit for ‘thin and light’ mobile platforms. The interesting correlation here is that at 1.7mm, it matches Intel’s EMIB implementation which is also 1.7mm.
Still MIA.
 

Mopetar

Diamond Member
Jan 31, 2011
4,486
421
126
#8
Is this actually Vega or just Polaris tech being branded as Vega? I’m not sure if it matters much either way though.

Edit: After reading a few articles it does appear to be actual Vega tech and not just the name as was the case with the Intel chip.

Assuming that the specs are true, this should be quite similar in performance to the PS4 Pro as the much higher clock speeds will compensate for the reduced number of cores.

That should make this a fairly capable machine.
 
Last edited:

Kenmitch

Diamond Member
Oct 10, 1999
7,167
110
126
#9
It's Vega and it's in a mobile device so I consider it as vega mobile.

AMD giving priority to products that are already sold makes good business sense. I doubt Subor called up AMD and said if you make it we might buy it. I'd imagine a contract was struck with a minimum amount of product at X amount of $'s was signed. This was most likely one of those future design wins that was somewhat mentioned in one of the past presentations.

The Chinese market is said to be huge and relatively untapped so I see this as a good move on AMD's part.

What would you do?
 
Last edited:

Mopetar

Diamond Member
Jan 31, 2011
4,486
421
126
#10
As much as AMD fans or tech enthusiasts want mobile Vega does the market actually want it?

We know that NVidia has much better power/performance characteristics right now so what value does a stand-alone Vega that can be used as a mobile part provide?

I don’t think we see one until Polaris gets replaced and there are midrange desktop parts that will double as notebook parts. Basically we get it with APUs or SoCs for now.
 
Mar 11, 2004
18,410
744
126
#11
It's Vega and it's in a mobile device so I consider it as vega mobile.

AMD giving priority to products that are already sold makes good business sense. I doubt Subor called up AMD and said if you make it we might buy it. I'd imagine a contract was struck with a minimum amount of product at X amount of $'s was signed. This was most likely one of those future design wins that was somewhat mentioned in one of the past presentations.

The Chinese market is said to be huge and relatively untapped so I see this as a good move on AMD's part.

What would you do?
But not even AMD considers that mobile Vega.

I agree, and I don't really get this "where's mobile vega" stuff in response to this. If anything the question should be, why didn't they release a GDDR5X version of Vega for $299-399 this year so they'd at least have something. But then maybe they assumed that mining winding down there'd be HBM Vega cards ditched for similar or lower prices and so it'd kill the market for it or something.

As much as AMD fans or tech enthusiasts want mobile Vega does the market actually want it?

We know that NVidia has much better power/performance characteristics right now so what value does a stand-alone Vega that can be used as a mobile part provide?

I don’t think we see one until Polaris gets replaced and there are midrange desktop parts that will double as notebook parts. Basically we get it with APUs or SoCs for now.
I kinda agree. I think the biggest issue is that it probably just isn't profitable so AMD can't price it aggressively enough to compete with Nvidia's offerings, so there's little point to releasing it. The Macbook Pro I don't think could handle it. Using a Polaris GPU, it has to throttle the CPU pretty hard when the GPU is active (think it was like down to 2.0GHz sustained CPU clock), and Vega would be even worse in that regard or offer not enough extra performance to make the cost worth it if they pushed it to a similar TDP. Maybe it'll be an option in the standard iMac or something? Maybe some svelte external GPU box that Apple will make? Those are about the only two options I'd think, because Apple is locked to AMD, and they can float the price to be high enough. Unless AMD is willing to take a profit hit and offer a package deal of mobile Ryzen paired with mobile Vega, but I doubt that. Possibly AMD is wanting companies to make a worthwhile package (meaning, make a good design/chassis, we give you mobile Ryzen and mobile Vega for little, but you're set so when we release Ryzen 2 and whatever our next GPU is, you can slot those in quicker and have a competitive design ready quicker), so they're making them put in more work, and we'll see some at next CES, and then the CES after that we see the versions with the next stuff.

I agree on that as well. I think AMD is choosing to focus on what's next. That and if they're doing these custom APUs for several customers that's probably taking some of their resources, but then they probably also make more money from that than from dGPU Vega (even if it was more competitive I don't know that it'd be a big money maker for them in consumer). It kinda sounds like they've shaken up things quite a bit after the Vega launch, and Raja's departure. Will be really interesting to see how they respond (and with Lisa Su putting focus on the GPU now that they've got Ryzen out and going well).
 

Mopetar

Diamond Member
Jan 31, 2011
4,486
421
126
#12
But not even AMD considers that mobile Vega.

I agree, and I don't really get this "where's mobile vega" stuff in response to this. If anything the question should be, why didn't they release a GDDR5X version of Vega for $299-399 this year so they'd at least have something.
I think that the obvious answer is that it wouldn’t be significantly better than Polaris in order to justify the cost. They may as well wait for a new node and faster memory so that they can release a small die part that can serve as a pipe cleaner of sorts and slot in as a true replacement to Polaris. Small Vega now would be trying to fight a battle that’s long over.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
227
96
#13
I agree, and I don't really get this "where's mobile vega" stuff in response to this. If anything the question should be, why didn't they release a GDDR5X version of Vega for $299-399 this year so they'd at least have something. But then maybe they assumed that mining winding down there'd be HBM Vega cards ditched for similar or lower prices and so it'd kill the market for it or something.
.
The "Wheres Vega Mobile" comes from AMD Showing Vega Mobile in February, and telling us it's coming. But it never showed up. They even held it up to get photographed. It's the small one shown here:





Intel got a GPU chip from AMD for Kaby Lake G of similar size., and now there is a console chip. Both of which have GPU sections similar size to Vega Mobile. But still no sign of Vega Mobile.

So I was wondering, where is that thing that AMD showed us at the beginning of the year and forgotten. So instead we get designs for third parties.

That is why I was openly wondering where Vega Mobile is. This APU was not specifically a mobile part, but it was reminder that AMD is building similar size GPU sections for other people, but doesn't have one for sale.

Since asking I have speculated, that it is simply too expensive to be competitive with current HBM/Interposer costs.
 
Dec 29, 2015
72
44
61
#14
The "Wheres Vega Mobile" ...
My "Total Speculation" with a rose colored glasses on:
AMD may be trying to ramp 7nm as fast as they can and is sacrificing current production capacity to prepare for a rapid 7nm production ramp next year. AMD is currently producing all their 14nm products at GF except for console chips. GF has only 1 Fab capable of producing 14nm products and is trying to bring up 7nm in the same Fab. Knowing that in 2019, AMD will have not only a superior architecture but also process advantage against Intel, it makes sense for them to prepare for a rapid ramp of 7nm production. If that comes at the cost of current production, I think it is a worthwhile trade-off. AMD would need to produce all the chips they promised to current customers. But, if they have to delay unreleased product to make room for equipment upgrade necessary for 7nm production next year? I'd choose to delay the unreleased product. The demand for new graphics chips will be low for a while anyway while the used gpu inventory from miners flood ebay.
 

Mopetar

Diamond Member
Jan 31, 2011
4,486
421
126
#15
That’s a possibility, but I think the easier explanation is that mobile Vega doesn’t really have a reason to exist. The performance per watt favors NVidia too much to justify a discreet mobile part. That it likely wouldn’t be cost competitive either means AMD is better off devoting manufacturing capacity to other chips that will sell in quantity and at good margins.
 

Qwertilot

Golden Member
Nov 28, 2013
1,414
37
106
#16
Maybe if it had launched at the same time as Pascal they'd have squeezed something out, but they'd only have had a few months before it got thoroughly obsoleted by whatever NV are calling their next generation.
 

crisium

Platinum Member
Aug 19, 2001
2,595
19
136
#17
I'm not sure there would be a point to Vega Mobile compared to a dedicated CPU + a 1050 Ti or a 1060 Max Q. Why would the open laptop market support this? And if the new generation offers even better performance per watt then forget it.

For a console it makes much more sense. Though I wonder how it compared to the Polaris APUs in the PS4 Pro and Xbone X in terms of manufacturing cost. Though even it has less compute power just having Zen instead of Jaguar makes it better obviously.
 
Mar 11, 2004
18,410
744
126
#18
The "Wheres Vega Mobile" comes from AMD Showing Vega Mobile in February, and telling us it's coming. But it never showed up. They even held it up to get photographed. It's the small one shown here:





Intel got a GPU chip from AMD for Kaby Lake G of similar size., and now there is a console chip. Both of which have GPU sections similar size to Vega Mobile. But still no sign of Vega Mobile.

So I was wondering, where is that thing that AMD showed us at the beginning of the year and forgotten. So instead we get designs for third parties.

That is why I was openly wondering where Vega Mobile is. This APU was not specifically a mobile part, but it was reminder that AMD is building similar size GPU sections for other people, but doesn't have one for sale.

Since asking I have speculated, that it is simply too expensive to be competitive with current HBM/Interposer costs.
No, I get that. I just don't get why that keeps getting brought up with this product. Its a completely different chip, for a different market with different parameters. Wondering where mobile Vega is has nothing to do with this chip.

Intel's chips isn't Vega though, its actually Polaris. Again a completely different product (and that seemingly was in the works for some time, there were rumors about it what, back around Polaris launch?).

I agree with that last bit. Its very probably too expensive and AMD isn't willing to take a loss on it to make it worthwhile to use, so no one is going to use it when they can use superior products from Nvidia in that market. AMD isn't going to announce that they have basically no market for it though. Its like wondering why current consumer Tegra chips (the Switch is using the older one that is what almost 3 years old at this point?) are practically nowhere to be seen. But in a way they can't not make those chips as that will definitely make sure there's no market for them. So they announce them, and then just let the market dictate if they actually get used. It certainly speaks that there's something up, that speaks to other issues (Nvidia seems to want high margins or something that puts them out of contention in mobile when there's an abundance of alternatives; dGPU consumer Vega is a dud as its tied to HBM2/interposer which makes it too costly for the performance, let alone perf/W to use - which is sad since a point of HBM is that it was supposed to help make the GPU package smaller and more efficient).

That’s a possibility, but I think the easier explanation is that mobile Vega doesn’t really have a reason to exist. The performance per watt favors NVidia too much to justify a discreet mobile part. That it likely wouldn’t be cost competitive either means AMD is better off devoting manufacturing capacity to other chips that will sell in quantity and at good margins.
Right. Even if AMD makes them, there probably just isn't a market for them. I have a hunch we'll see them update that Intel GPU with the Vega mobile chip at CES or something. But that's probably the only place that chip makes any sense to use, and that's only because the product stack already exists otherwise it wouldn't (actually I don't know that it makes sense then even, its just that people that really want one of those Skull Trail NUCs and maybe some laptops, but even then I don't know that it makes sense compared to other ones with dGPU from Nvidia).

I'm not sure there would be a point to Vega Mobile compared to a dedicated CPU + a 1050 Ti or a 1060 Max Q. Why would the open laptop market support this? And if the new generation offers even better performance per watt then forget it.

For a console it makes much more sense. Though I wonder how it compared to the Polaris APUs in the PS4 Pro and Xbone X in terms of manufacturing cost. Though even it has less compute power just having Zen instead of Jaguar makes it better obviously.
Yeah. The thing is, AMD has to show they're still making mobile dGPUs otherwise companies are going to stop even considering using them (more than they have).

It can't cost too much because the deal was only for like ~$60 million, and its an APU pairing Zen with Vega. I think it shows that HBM2 and the interposer just is not fiscally feasible for consumer stuff yet. That's why I can't figure out why AMD didn't do a dGPU of Vega with GDDR5X. That should have let them get to $349 or $399 and compete with the 1070 (clock it more conservatively so that its in the sweet spot of efficiency, so that the performance against the 1070 would be better, while keeping the efficiency better). I think they probably just had too much else going on and didn't have the resources (although I wonder how much it would have taken, but maybe it would have required a new chip with GDDR memory controllers - although I thought they said it could support either, but maybe they didn't include them on the Vega chips they made, or fused them off or something; I wonder if there weren't enough memory controllers so it would've been stuck at like 256bit so it would've been bandwidth starved too without redoing the chip for more memory controllers). Now, I think that is a better question, why didn't they do that instead of Vega mobile, since there would at least have been a market for it. It'd be a good pairing of that card with Ryzen 2xxx series for OEMs.

The reports that they've focused so much resources on Navi means it'll be really interesting to see how that turns out. Those reports also have been weirdly negative in calling it a lower end/consumer level chip, although, I believe that was always the point of it (with it starting AMD trying on card mGPU solutions where they appear/function similar to a similar size single monolithic die), and that's where AMD has made their money in dGPUs for years, so I think that's prudent anyway. But since it also is reported they've tailored it heavily to what Sony wants for the Playstation, it will be interesting to see how it turns out and if it does attempt the mGPU die strategy. That might have gone out the window if HBM was supposed to be a key part of that, and its cost/complexity makes it unfeasible for consumer stuff.
 

PeterScott

Platinum Member
Jul 7, 2017
2,605
227
96
#19
It can't cost too much because the deal was only for like ~$60 million, and its an APU pairing Zen with Vega. I think it shows that HBM2 and the interposer just is not fiscally feasible for consumer stuff yet. That's why I can't figure out why AMD didn't do a dGPU of Vega with GDDR5X. That should have let them get to $349 or $399 and compete with the 1070 (clock it more conservatively so that its in the sweet spot of efficiency, so that the performance against the 1070 would be better, while keeping the efficiency better).
IMO, HBM is too expensive, and without HBM, they really have no competitive edge. So the GPU division is in a rough spot. They really need a architecture uplift to improve efficiency (per watt and area). Maybe they are counting on leapfrogging NVidia to 7nm.
 

Dribble

Golden Member
Aug 9, 2005
1,693
118
126
#20
The people actually buying vega are the miners and they don't want a laptop version.
 


ASK THE COMMUNITY

TRENDING THREADS