Question AMD Phoenix/Zen 4 APU Speculation and Discussion

[uzzi38]

I can finally make this thread.

https://twitter.com/x/status/1519669375283957760

Phoenix is indeed RDNA3. My advice to everyone: treat the old APU rumours as being out of date.

 

[dr1337]

You pay in the case of 7800X3D 34% more than for 7700X, in return you gain 11% higher FPS. For that you also need a $1599 GPU and 1080p, because at higher resolutions the difference gets smaller. At least with 7950X3D you pay only 19.5% more for comparable increase in FPS.

For this performance increase, I don't consider It a good deal or fair price when It cost only a fraction of that extra price.

Not every game is CPU bound and beyond that I doubt I've played most games compiled in that average. Have you?

I noticed the difference between a 3600 and 5800x3d on an rx580 before finally finding an acceptable deal on a 6900xt. Even when playing at 4k, because in CPU limited games where resolution or GPU rarely matters, I wouldn't factor in the cost of the GPU at all. I already ran stalker anomaly at max settings and max AA 4k but even after upgrading the graphics card I still get huge frame drops when lots of NPCs are fighting. To me seeing an average like this only exemplifies the performance of the vcache. 11% faster with cache in normal games seems huge considering the thermal&clock hit.


I can appreciate how the product isn't for everyone, don't get me wrong. Its not essential to PC gaming even for enthusiasts. But as someone using this technology for over a year now, I find it to be very nice. 10 years ago I expected processors from today would be so fast that old CPU limited games would just get blown away. Sadly we aren't even close to that and this premium technology is the best any company has done thus far.

 

[TESKATLIPOKA]

Not every game is CPU bound and beyond that I doubt I've played most games compiled in that average. Have you?

With my GTX 1650M? I am happy that It can play Warhammer III.

I noticed the difference between a 3600 and 5800x3d on an rx580 before finally finding an acceptable deal on a 6900xt. Even when playing at 4k, because in CPU limited games where resolution or GPU rarely matters, I wouldn't factor in the cost of the GPU at all. I already ran stalker anomaly at max settings and max AA 4k but even after upgrading the graphics card I still get huge frame drops when lots of NPCs are fighting. To me seeing an average like this only exemplifies the performance of the vcache. 11% faster with cache in normal games seems huge considering the thermal&clock hit.


I can appreciate how the product isn't for everyone, don't get me wrong. Its not essential to PC gaming even for enthusiasts. But as someone using this technology for over a year now, I find it to be very nice. 10 years ago I expected processors from today would be so fast that old CPU limited games would just get blown away. Sadly we aren't even close to that and this premium technology is the best any company has done thus far.

Don't take me wrong, 7800X3D is one of the best CPUs for gaming and It has fantastic efficiency, but It's really not cheap.

This debate wasn't really about a concrete CPU with 3D cache, although It ended up like that, but about the 3D cache Itself and Its extra price.
AMD is not asking a small amount for this cache, and this could be a problem for future APUs.
APU laptops are already pretty costly for what they offer, CPU is ok, but IGP performance is weak, you need to pay only a $200 for a totally different level of gaming performance.
This cache would make a big difference for APUs, but It would also make them even more expensive.
So I have to wonder when we will see them in APUs.

 

[dr1337]

APU laptops are already pretty costly for what they offer, CPU is ok, but IGP performance is weak, you need to pay only a $200 for a totally different level of gaming performance.

The mobile market has some serious pricing and availability issues, at least where I live. Right now I could buy a 5600h+gtx 1660 laptop for the same price as the cheapest 5700u ultrabook. Theres no way its cheaper to have a dGPU than it is to built a thin and light, yet the pricing suggests otherwise. The prices are all made up and performance teirs don't matter.

This cache would make a big difference for APUs, but It would also make them even more expensive.
So I have to wonder when we will see them in APUs.

I'd think they're working on getting it done as soon as possible, it just seems like the biggest no brainer. But maybe the market is the problem. The most expensive APU laptop I could buy today is $1300 and has a 6800u inside it, cheapest with the same chip I could find was $899. AMD if anything desperately needs a halo APU for the OEMs cause IMO it'd be a hard sell to even ship Phoenix at these macbook air prices. And I can't imagine too many people are buying the overpriced ultrabooks over better specc'd random other laptops on the market, or a mac...

I do wonder if they have some manufacturing issues, or maybe vcache in a mobile product is harder to validate because of more vibrations and direct die cooling? But maybe they have a much wiser plan in store for enhanced APU performance; If they could glue an rdna3 MCD to the side of an APU, even that 16mb cache would help let alone giving the APU a gig of its own GDDR6. If we don't see cache come to the APUs within the next two generations then I'd chalk that up to a loss for AMD.

 

[Joe NYC]

With my GTX 1650M? I am happy that It can play Warhammer III.


Don't take me wrong, 7800X3D is one of the best CPUs for gaming and It has fantastic efficiency, but It's really not cheap.

This debate wasn't really about a concrete CPU with 3D cache, although It ended up like that, but about the 3D cache Itself and Its extra price.
AMD is not asking a small amount for this cache, and this could be a problem for future APUs.
APU laptops are already pretty costly for what they offer, CPU is ok, but IGP performance is weak, you need to pay only a $200 for a totally different level of gaming performance.
This cache would make a big difference for APUs, but It would also make them even more expensive.
So I have to wonder when we will see them in APUs.

I understood that your point was cost and price of 3D cache itself.

On cost, I think you are in the right ballpark with $5 range for silicon, which could be even lower for SRAM only die.

On packaging, it starts high and drops as volume grows and automation of the process improves.

The goal for the cost N6/N7 SRAM die + packaging to be less than the price of SRAM die area on N4/N5. And then on N3, N2. So, I think the cost of packaging is going down to single digits.

Where the cost of packaging is now is unknown. TSMC is making investments in this type of packaging, so it is likely just a negotiation of how AMD and TSMC split the cost of this ramp.

As far as how AMD is pricing 3D cache, I think it is wrong to look at a premium product such as 7800x3d. 5800x3d and upcoming 5600x3d will be a better guide.

https://twitter.com/x/status/1674774166799544321

 

[TESKATLIPOKA]

I'd think they're working on getting it done as soon as possible, it just seems like the biggest no brainer. But maybe the market is the problem. The most expensive APU laptop I could buy today is $1300 and has a 6800u inside it, cheapest with the same chip I could find was $899. AMD if anything desperately needs a halo APU for the OEMs cause IMO it'd be a hard sell to even ship Phoenix at these macbook air prices. And I can't imagine too many people are buying the overpriced ultrabooks over better specc'd random other laptops on the market, or a mac...

I do wonder if they have some manufacturing issues, or maybe vcache in a mobile product is harder to validate because of more vibrations and direct die cooling? But maybe they have a much wiser plan in store for enhanced APU performance; If they could glue an rdna3 MCD to the side of an APU, even that 16mb cache would help let alone giving the APU a gig of its own GDDR6. If we don't see cache come to the APUs within the next two generations then I'd chalk that up to a loss for AMD.

It's also possible OEMs don't really want such a product -> halo APU.
There already existed a CPU+GPU+ 4GB HBM2 combo on a single package -> Kaby Lake G.
It ended up in only a few products and didn't see a successor.
Yes, It's not an APU, but still pretty similar concept.

 

[TESKATLIPOKA]

As far as how AMD is pricing 3D cache, I think it is wrong to look at a premium product such as 7800x3d. 5800x3d and upcoming 5600x3d will be a better guide.

I just looked in my local store and 5800X3D is sold for €305.90, which is a lot less than 7800X3D, but It also costs 59% more than 5800X, which can be bought for only €192.90.
Difference is >€100.
5600X3D looks like will be released only in the USA at MicroCenter for $229.

 

[Joe NYC]

I just looked in my local store and 5800X3D is sold for €305.90, which is a lot less than 7800X3D, but It also costs 59% more than 5800X, which can be bought for only €192.90.
Difference is >€100.
5600X3D looks like will be released only in the USA at MicroCenter for $229.

$229 is right in the range I predicted a couple of days ago ($199-$239). It will not be long before we see it discounted to $199.

BTW, there were some rumors that AMD bought out entire TSMC's entire SoIC capacity for a couple of years, so we will likely see further proliferation of V-Cache. 7600x3d will likely be the next candidate. And, later this year, Mi300 is going into production.

The way SoIC capacity is ramping, I think it is probable we will see some versions of Strix (Halo?) with some sort of 3D cache, either x3d V-Cache style or Mi300 style.

 

[Exist50]

AMD if anything desperately needs a halo APU for the OEMs cause IMO it'd be a hard sell to even ship Phoenix at these macbook air prices.

I think there's a big market for a halo APU. All of those x50 and x60 series GPUs in a laptop could be replaced with a sufficiently big APU, and the integrated cost and power advantage would really help compete with Nvidia. The biggest problem is probably branding. Nvidia's brand is so strong in the gaming market that AMD (and Intel, for that matter) need to produce something very compelling to convince OEMs and consumers to bite.

I do wonder if they have some manufacturing issues, or maybe vcache in a mobile product is harder to validate because of more vibrations and direct die cooling?

I don't think vcache makes as much sense for APUs right now. Would it provide an advantage vs applying the same incremental cost to other components? Also, laptops run hotter, so that lower thermal limit might be a problem.

 

[GodisanAtheist]

Haven't paid attention for 59 pages but how about 16MB of IC before going straight for the 3d VCache for APUs?

 

[jpiniero]

I think there's a big market for a halo APU. All of those x50 and x60 series GPUs in a laptop could be replaced with a sufficiently big APU, and the integrated cost and power advantage would really help compete with Nvidia. The biggest problem is probably branding. Nvidia's brand is so strong in the gaming market that AMD (and Intel, for that matter) need to produce something very compelling to convince OEMs and consumers to bite.

I'm just not sure if the cost savings would be there versus keeping it separate. And the power savings might not be there because you'd probably only have the big IGP tile which draws more.

What I'm imagining is something like if Meteor Lake had it's IGP tile replaced with a real discrete GPU and had the packaging to wire in external GDDRX.

 

[Exist50]

I'm just not sure if the cost savings would be there versus keeping it separate. And the power savings might not be there because you'd probably only have the big IGP tile which draws more.

What I'm imagining is something like if Meteor Lake had it's IGP tile replaced with a real discrete GPU and had the packaging to wire in external GDDRX.

I'm thinking something more like what Apple does with additional LPDDR channels. Or just have a very large cache to compensate for poor memory bandwidth. Then you can save vs two different memory pools, and hopefully also board complexity/area. Extra memory bandwidth might help a handful of CPU workloads as well.

That does assume, of course, that the GPU IP is competitive enough to bother making such a product.

 

[insertcarehere]

I think there's a big market for a halo APU. All of those x50 and x60 series GPUs in a laptop could be replaced with a sufficiently big APU, and the integrated cost and power advantage would really help compete with Nvidia. The biggest problem is probably branding. Nvidia's brand is so strong in the gaming market that AMD (and Intel, for that matter) need to produce something very compelling to convince OEMs and consumers to bite.

I don't think vcache makes as much sense for APUs right now. Would it provide an advantage vs applying the same incremental cost to other components? Also, laptops run hotter, so that lower thermal limit might be a problem.

We've had this debate before in this thread , if AMD believed there is significantly untapped market potential for a big APU, they would have pushed it out already. Nothing stops AMD from copying much of the work they did for the PS5/XSX chips and putting the result into market. The fact that nothing of this sort exists 3 years after the launch of the consoles suggest that to them, the demand just isn't there.

 

[Exist50]

We've had this debate before in this thread , if AMD believed there is significantly untapped market potential for a big APU, they would have pushed it out already. Nothing stops AMD from copying much of the work they did for the PS5/XSX chips and putting the result into market. The fact that nothing of this sort exists 3 years after the launch of the consoles suggest that to them, the demand just isn't there.

Just this generation, AMD added a second die to their mainstream APU lineup. They are clearly being very conservative with how they expand their product line. I think it would be a mistake to interpret that as a lack of demand or even will.

Also, AMD is likely to move graphics to a separate chiplet within the next few generations. Combine that with advanced packaging, and you have a much more cost effective solution than a dedicated, monolithic die like the consoles use.

 

[SpudLobby]

What's kind of interesting is, Intel has the better chiplet setup for an Apple-style APU at least relative to anything we know about from AMD - though they'll figure something out with TSMC and one of their suppliers.

I do hope Strix Halo doesn't end up with some disgusting IO that drains power and has the CPU still split up like they do on the desktop. At some point it's just a slightly cheaper gamer laptop setup which is fine but partially misses the mark IMO.

With Apple they've managed a 256-Bit Bus with some low power agility - the M2 Pro chips do lose some idle and battery life relative to the M2 base models, but it's nothing egregious, I don't think it's a dealbreaker. If AMD takes a sloppier route with packaging such that Halo looks more like Dragon Range with a GPU, lol. In principle nothing will beat monolithic power or performance wise but would be great if they can aim more towards "scaled-up mobile APU with a cost focus (via chiplet tiles)" rather than "dGPU with cost savings on RAM and a CPU attached" which is what Halo kind of sounds like to me - though the TDPs might suggest otherwise and no I don't think they'll literally use the same desktop packaging but still, matter of degrees.

Really, it's just a shame MTL isn't testing this. Maybe Arrow Lake will but rumors have trended bearish on this.

You don't want a literal fat dGPU on the tile or GDDR6. Just a 1.5-3x normal iGPU tile and large bus. 6+8 CPU tile on 20A + a 192EU/320EU N3 tile with a 256-bit LPDDR5X bus would be huge - permitting the GPU architecture and drivers are even faintly comparable to AMD by that point which isn't unreasonable.

 

[Shivansps]

Just this generation, AMD added a second die to their mainstream APU lineup. They are clearly being very conservative with how they expand their product line. I think it would be a mistake to interpret that as a lack of demand or even will.

Also, AMD is likely to move graphics to a separate chiplet within the next few generations. Combine that with advanced packaging, and you have a much more cost effective solution than a dedicated, monolithic die like the consoles use.

AMD is not going to start eating into their entry level GPU territory unless the IGP perf is threatened by Intel.

 

[Exist50]

I do hope Strix Halo doesn't end up with some disgusting IO that drains power and has the CPU still split up like they do on the desktop.

I think it's interesting to speculate where they'd draw the cutlines for such a product. I know there are rumors claiming a big GPU+AI+IO die, with reused compute dies, and I guess that sounds fine for a really high end product, but I agree it would be interesting to see something from the other direction with a mobile-derived CPU+IO complex and another die for GPU and maybe AI. Or something like Intel's SoC cores might be interesting, but that'll probably come after they have hybrid better established.

AMD is not going to start eating into their entry level GPU territory unless the IGP perf is threatened by Intel.

Why not? For the same level of performance, it should be cheaper for AMD to offer it as an iGPU, so that's a win for everyone.

 

[Exist50]

Oh, and just caught this. Phoenix 2 pictured. https://www.tomshardware.com/news/amd-phoenix-2-die-shot

 

[burninatortech4]

Pretty cool product. Is PX2 the "small" APU die product for 2023 or is this its own thing?

Stoney (Excav) > Dali/Pollock (Zen1) > Mendocino (Zen2) > PHX2 (Zen4)

or will we still see a separate Mendocino successor with an even smaller die?

 

[Thunder 57]

Oh, and just caught this. Phoenix 2 pictured. https://www.tomshardware.com/news/amd-phoenix-2-die-shot

View attachment 82587

Oh yay another hentai girl.

 

[soresu]

Oh, and just caught this. Phoenix 2 pictured. https://www.tomshardware.com/news/amd-phoenix-2-die-shot

View attachment 82587

PHX2 sounds like an ultra thin/light or embedded chip?

 

[soresu]

Pretty cool product. Is PX2 the "small" APU die product for 2023 or is this its own thing?

Stoney (Excav) > Dali/Pollock (Zen1) > Mendocino (Zen2) > PHX2 (Zen4)

or will we still see a separate Mendocino successor with an even smaller die?

Given the tme between the respective architecture releases and those chips I'd wager on PHX2 coming out more like next year if it matches the trend.

These smaller chips end up in lots of products due to their cost.

 

[NostaSeronx]

Pretty cool product. Is PX2 the "small" APU die product for 2023 or is this its own thing?

Stoney (Excav) > Dali/Pollock (Zen1) > Mendocino (Zen2) > PHX2 (Zen4)

or will we still see a separate Mendocino successor with an even smaller die?

Stoney became its own thing. They haven't actually produced a marketed replacement for it.

This the progession AMD generally advertises:
Premium FPx :: Raven -> Picasso -> Renoir -> Cezanne -> Rembrandt -> Phoenix
Mainstream FPx :: Bristol -> Raven2/Dali -> Phoenix2
Value Mainstream FTx :: Pollock -> Mendocino -> next part: Sonoma Valley
Low-cost Entry FTx :: Beema -> Stoney -> next part: name not announced (yet)

Beema/Stoney => BGA market cost: $20-$35, in 2019 <== No successor to this
Raven2/Dali => BGA market cost: $70-$100, in 2019 <== Mendocino dual-core should be in this range, while Mendocino Quad should be $120-$160

Stoney Refresh 2(A9-9435, Never Released) is being sold as a Chinese A9-9400, for the August2022~November2022 MiniPCs. Which China OEMs only do new prior generations in preparation for a new generation. China [GloFo,AMD] have in closed circles already leaked usage of 12FDX in September 2022 in successor parts for A9-9400(China-variant).

//Stoney and Refresh points, current:
Stoney => A9-9400 (US-variant)
Stoney Refresh => A9-9420 (US)
Stoney Refresh+ => A9-9425 (US)
Stoney Refresh2 => A9-9435(US,Never Released)/A9-9400(China)

 

[jpiniero]

Given the tme between the respective architecture releases and those chips I'd wager on PHX2 coming out more like next year if it matches the trend.

These smaller chips end up in lots of products due to their cost.

If anything Little Phoenix is being rushed out because Big Phoenix is very unpopular. Presumably it's the price that's the issue.

 

[Thunder 57]

Stoney became its own thing. They haven't actually produced a marketed replacement for it.

This the progession AMD generally advertises:
Premium FPx :: Raven -> Picasso -> Renoir -> Cezanne -> Rembrandt -> Phoenix
Mainstream FPx :: Bristol -> Raven2/Dali -> Phoenix2
Value Mainstream FTx :: Pollock -> Mendocino -> next part: Sonoma Valley
Low-cost Entry FTx :: Beema -> Stoney -> next part: name not announced (yet)

Beema/Stoney => BGA market cost: $20-$35, in 2019 <== No successor to this
Raven2/Dali => BGA market cost: $70-$100, in 2019 <== Mendocino dual-core should be in this range, while Mendocino Quad should be $120-$160

Stoney Refresh 2(A9-9435, Never Released) is being sold as a Chinese A9-9400, for the August2022~November2022 MiniPCs. Which China OEMs only do new prior generations in preparation for a new generation. China [GloFo,AMD] have in closed circles already leaked usage of 12FDX in September 2022 in successor parts for A9-9400(China-variant).

//Stoney and Refresh points, current:
Stoney => A9-9400 (US-variant)
Stoney Refresh => A9-9420 (US)
Stoney Refresh+ => A9-9425 (US)
Stoney Refresh2 => A9-9435(US,Never Released)/A9-9400(China)

Haha, whenever I hear "stoney" or "tunnelborder" or FDX anything I already know who posted it.

 

[Exist50]

Pretty cool product. Is PX2 the "small" APU die product for 2023 or is this its own thing?

Stoney (Excav) > Dali/Pollock (Zen1) > Mendocino (Zen2) > PHX2 (Zen4)

or will we still see a separate Mendocino successor with an even smaller die?

It's presumably supposed to compete with Intel's 2+8 U series chips around 15W. ADL 2+8 is in a ton of devices, and it makes sense for AMD to put out more dies to better match the price point and power envelope.

Mendocino is more of a cost-focused product, and doesn't have a clear successor yet. I think eventually we'll see something with 4x Zen 4c on N4, but that's just a guess.