Discussion RDNA4 + CDNA3 Architectures Thread

Page 113 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

DisEnchantment

Golden Member
Mar 3, 2017
1,773
6,750
136





With the GFX940 patches in full swing since first week of March, it is looking like MI300 is not far in the distant future!
Usually AMD takes around 3Qs to get the support in LLVM and amdgpu. Lately, since RDNA2 the window they push to add support for new devices is much reduced to prevent leaks.
But looking at the flurry of code in LLVM, it is a lot of commits. Maybe because US Govt is starting to prepare the SW environment for El Capitan (Maybe to avoid slow bring up situation like Frontier for example)

See here for the GFX940 specific commits
Or Phoronix

There is a lot more if you know whom to follow in LLVM review chains (before getting merged to github), but I am not going to link AMD employees.

I am starting to think MI300 will launch around the same time like Hopper probably only a couple of months later!
Although I believe Hopper had problems not having a host CPU capable of doing PCIe 5 in the very near future therefore it might have gotten pushed back a bit until SPR and Genoa arrives later in 2022.
If PVC slips again I believe MI300 could launch before it

This is nuts, MI100/200/300 cadence is impressive.



Previous thread on CDNA2 and RDNA3 here

 
Last edited:

marees

Senior member
Apr 28, 2024
956
1,272
96
Very interesting, but your calculation focuses on area, and I'm curious about three elements:
- costs of packaging of Halo
- R&D, which I expect has been particularly extensive (LP core, advanced packaging, big APU in a laptop, etc)
- volume assessment

As always I'm a little skeptical of the AMBeliebers that scream that "chipletz are de wey" since they focus on the gains and ignore the extra costs. Which aren't big for cheaper stuff (fan-outs) but get higher pretty fast.
Halo is chiplet-based, Strix is not. To me, that sounds like an extra layer of problems to deal with in a supposed low power laptop. I suspect it may be their (first?) implementation of a silicon bridge, which is less money than MTL's Silicon interposer, but is still more than traces in PCB, as Halo is definitely the first part to leave AMD's Zen 2 era "just do the cheapest" approach. It's a testbed for Zen 6 and later, so no point in going cheap.

I expect both packaging costs to be relatively high, and R&D to have been heck of a lot.
Ofc volume is supposed to quash that, but I doubt that it'll sell as much as people believe. NV sticker is still worth a lot, Halo completely negates the value of putting a dGPU in a laptop. So I expect that Halo laptops will have to be more price competitive, I.E cheaper than full laptops with dGPUs. I also can easily expect NV to come out blasting about how "real laptops have dGPUs" and all the obvious jazz they'll do to impede Halo.

If Halo sells for cheaper perf to perf than Point + 4050/60/70, and its situation is "less area, less complexity to integrate, more packaging costs, has to sell for less of a price", I really really am doubtful about any margins or large scale sales.
Of course I'm willingly ignoring Halo's ability to be a Windows M1, a really low power stable chip, but that one is still out there, we don't know a thing about Z5 LP except that it's meant to exist. And we have to be doubtful about the market size of people who want a really low power full AMD laptop, but will be told "by the way, if you want it, you gotta pay for the midrange gaming sized GPU". Market of windows low power users? Huge. Market of those willing to pay another $100/200 for a gaming GPU? Tiny.

Halo will be a hard sell, it'll have cost a lot to develop, and its advantage is not that high in area, it's there for sure but we're talking 100m² if we're being ultra generous.
So I'm going to keep my original stance that adroc is full of copium on this and that Halo will be a net loss for AMD. Which is fine because it's an experimental, high risk product, that's meant to open the way to Z6 and beyond, and that's really all I want out of AMD, to take risks. But they're not making their money back on this one. I think we should expect Halo to open a lot of doors, but to be a bust financially speaking.
And I'm glad about it. AMD's been playing the poor man's game for too long, Halo is a real hard leap forward, so do it Lisa, lose money and come back with most of the hard lifting done for Zen 6.
As much as I am excited for AMD's innovation in the fusion space, I am still very skeptical that they will make any progress with entrenched laptop makers

Microsoft sells arm but not AMD in its surface lineup, ffs.

One ray of hope for AMD is that devices like RoG Ally & Legion Go (along with a future Xbox OS) will mainstream AMD APUs & give it the much needed leg up in this space.
 

branch_suggestion

Senior member
Aug 4, 2023
640
1,357
96
I expect they'll go for something less power consuming and more advanced/expensive.
InFO made sense for N31, but for a low power laptop, I'm fully expecting a silicon bridge.
BW demands for the InFO are far lower than N31, as is power which for N31 was already <20W. Si bridges are hella expensive and hard to scale.
Sure, it's just a manner of speaking. It's a "future investment oriented product". Lose now, gain later.
Again this is all I've been asking out of AMD/Radeon for years.
Bit by bit.
Ponte Vecchio Redux Reloaded
That is what happens when they ask product teams to subsidize packaging development.
Yet again the Invisible Hand of Cerny virtuously pushes us towards good engineering.
Gaben and Cerny better be given good champagne when they visit AMD.
They wish, consoles are a means to an end and a nice cushy revenue stream. Everyone has good ideas about what to build, but you need to overcome the incumbents.
The enterprise folks are extremely stubborn for example, and the existing supply chains don't like the possibility of being obsoleted.
 

Mahboi

Golden Member
Apr 4, 2024
1,057
1,969
96
BW demands for the InFO are far lower than N31, as is power which for N31 was already <20W. Si bridges are hella expensive and hard to scale.
Yeah but if you're going to go for a seminal product that is meant to lead development of high margin stuff in the next ~4-5 years, might as well go all out.
Also with InFO, even if the power reqs are 1/5th N31, that's 4W, that's a fair bit for a laptop.
I'm keeping my bet there, we'll see when Kepler or Adroc will feel like throwing infos.
That is what happens when they ask product teams to subsidize packaging development.
Pardon???
They wish, consoles are a means to an end and a nice cushy revenue stream. Everyone has good ideas about what to build, but you need to overcome the incumbents.
The enterprise folks are extremely stubborn for example, and the existing supply chains don't like the possibility of being obsoleted.
Give my men champagne and stop being so stingy.
 

Joe NYC

Platinum Member
Jun 26, 2021
2,951
4,329
106
Very interesting, but your calculation focuses on area, and I'm curious about three elements:

Since the number I arrived at is quite high - 80 mm2 - it's the most significant

- costs of packaging of Halo

This type of packaging is quite low cost. It has been used for a long time in low cost smart phones, so the cost should be in low single digits.

Ther is also savings on die area by elimination of SerDes and power savings. So Strix Halo packaging is an asset, not a liability

- R&D, which I expect has been particularly extensive (LP core, advanced packaging, big APU in a laptop, etc)

That is true, but these are technologies that are going to carry AMD into the future. LP cores will probably be in all future notebook CPUs in Zen 6

- volume assessment

That's hard to assess. In the past generations, AMD was typically 10-50% better in GPU performance, but it did not particularly outweigh other features in Intel CPUs.

But ~3x GPU performance, or dropping dGPU is going differentiate Strix Halo from the rest of the field more significantly.


As always I'm a little skeptical of the AMBeliebers that scream that "chipletz are de wey" since they focus on the gains and ignore the extra costs. Which aren't big for cheaper stuff (fan-outs) but get higher pretty fast.
Halo is chiplet-based, Strix is not. To me, that sounds like an extra layer of problems to deal with in a supposed low power laptop. I suspect it may be their (first?) implementation of a silicon bridge, which is less money than MTL's Silicon interposer, but is still more than traces in PCB, as Halo is definitely the first part to leave AMD's Zen 2 era "just do the cheapest" approach. It's a testbed for Zen 6 and later, so no point in going cheap.

I don't think the cost of packaging is going to be a particular problem.

The problem with Strix Halo is that it does not play to the strength of chiplets fully, in being able to have the most appropriate node for a certain functionality, limiting the usage of the most expensive silicon - in case of Strix Halo, likely N3E, of which there may be well over 200 mm2. Could be over 250 mm2.

This big SoC is goig to be single purpose, no re-use.

But considering adding another 2x 100mm2 CPU CCD chiplets on top of ~250mm2 SoC still offers significant advantage in yields, and still some cost savings with N4 vs. N3.

Another advantage chiplets used to have was Intel's low yields, high cost of extra large dies, and inability of make more than 800 mm2 die.

These advantages do not apply as much with smaller chips and on high yielding TSMC nodes.

I expect both packaging costs to be relatively high, and R&D to have been heck of a lot.
Ofc volume is supposed to quash that, but I doubt that it'll sell as much as people believe. NV sticker is still worth a lot, Halo completely negates the value of putting a dGPU in a laptop. So I expect that Halo laptops will have to be more price competitive, I.E cheaper than full laptops with dGPUs. I also can easily expect NV to come out blasting about how "real laptops have dGPUs" and all the obvious jazz they'll do to impede Halo.

If Halo sells for cheaper perf to perf than Point + 4050/60/70, and its situation is "less area, less complexity to integrate, more packaging costs, has to sell for less of a price", I really really am doubtful about any margins or large scale sales.

One other variable to keep in mind is that low end dGPU such as x5x, x6x have 4 GB, 6 GB or 8 GB of memory, and the user is stuck with it for the rest of the life of the laptop.

Strix Halo will likely have 32 GB as a base configuration, leaving pretty much unlimited memory for the GPU in any game.

So, a good reply to dGPU marketing would be: is your game running dog slow with 6GB of memory?

Of course I'm willingly ignoring Halo's ability to be a Windows M1, a really low power stable chip, but that one is still out there, we don't know a thing about Z5 LP except that it's meant to exist. And we have to be doubtful about the market size of people who want a really low power full AMD laptop, but will be told "by the way, if you want it, you gotta pay for the midrange gaming sized GPU". Market of windows low power users? Huge. Market of those willing to pay another $100/200 for a gaming GPU? Tiny.

People still buy notebooks with dGPU, regardless of whether they need dGPUs, whether you think they need dGPUs.

Strix Halo offers a better choice to people who buy dGPUs. That's the target market.

For people who currently don't buy notebooks with dGPUs - Strix Halo is not the best fit.

Halo will be a hard sell, it'll have cost a lot to develop, and its advantage is not that high in area, it's there for sure but we're talking 100m² if we're being ultra generous.
So I'm going to keep my original stance that adroc is full of copium on this and that Halo will be a net loss for AMD.

I think the biggest challenge is that AMD is not NVidia. AMD does not have the same brand name, brand power in laptops.

If NVidia came up with Strix Halo, there would be zero doubt that this is the most brilliant idea. People would automatically buy it. Reviewers would be falling over each other recommending it.

That's the challenge for AMD to overcome. Not technical challenges of Strix Halo.

Which is fine because it's an experimental, high risk product, that's meant to open the way to Z6 and beyond, and that's really all I want out of AMD, to take risks. But they're not making their money back on this one. I think we should expect Halo to open a lot of doors, but to be a bust financially speaking.
And I'm glad about it. AMD's been playing the poor man's game for too long, Halo is a real hard leap forward, so do it Lisa, lose money and come back with most of the hard lifting done for Zen 6.

Surely it is high risk, but having brand recognition and brand power has high rewards. Which is what AMD is trying to do, after things fell apart.

The precise moment when things started falling apart for AMD was Q4 2021, when AMD failed to launch 5800x3d and unquestionably lost (handed over to Intel) the Halo effect. Followed by Rembrandt.

Rebuilding of the client brand started with launch of 7800x3d and Phoenix. And it is proceeding well.
 

Joe NYC

Platinum Member
Jun 26, 2021
2,951
4,329
106
I expect they'll go for something less power consuming and more advanced/expensive.
InFO made sense for N31, but for a low power laptop, I'm fully expecting a silicon bridge.

InFO has power and area savings over Infinity On Package AMD is using in desktop and Dragon Range.

CPU doesn't light up the link to memory as much as the GPU, so InFO is good enough for now even for mobile.
 

ToTTenTranz

Senior member
Feb 4, 2021
387
755
136
Microsoft sells arm but not AMD in its surface lineup, ffs.
From last-minute change to Nvidia GPU in OG Xbox announced on stage with AMD folks sitting on the front row, to last-minute cancellation of whatever mobile/handheld Van Gogh was built for, to Microsoft deciding not to include AMD's hardware tessellation in DX10 (after supporting and older version of it on X360 for years).. perhaps AMD is better off not depending on Microsoft for absolutely anything.



Strix Halo will likely have 32 GB as a base configuration, leaving pretty much unlimited memory for the GPU in any game.
And LLMs with >13B parameters and good enough tokens/s performance, which until recently was a task reserved to >$1500 discrete GPUs.
If AMD fails to mention and properly support this, they're just setting money on fire.


Rebuilding of the client brand started with launch of 7800x3d and Phoenix. And it is proceeding well.
Marketing has been godawful since Robert Hallock left, though.
AMD is now drastically changing their CPU/APU naming scheme every generation / 18 months.


AKA Satya Nadella
So... does this mean there will be handhelds with cut-down Halo or the statement was ironic and there won't be any?
 
Reactions: Tlh97

Mahboi

Golden Member
Apr 4, 2024
1,057
1,969
96
So... does this mean there will be handhelds with cut-down Halo or the statement was ironic and there won't be any?
Handhelds are in the 15-25W range if they don't wanna succ.
Frankly, just stick with a cut down Strix Point at that range.

From last-minute change to Nvidia GPU in OG Xbox announced on stage with AMD folks sitting on the front row, to last-minute cancellation of whatever mobile/handheld Van Gogh was built for, to Microsoft deciding not to include AMD's hardware tessellation in DX10 (after supporting and older version of it on X360 for years).. perhaps AMD is better off not depending on Microsoft for absolutely anything.
Hey, if you could get rid of our problems by just not working with them, it'd be easy.
Really, just kill all the problems. Literally.
But things don't work so good after you killed them.
And LLMs with >13B parameters and good enough tokens/s performance, which until recently was a task reserved to >$1500 discrete GPUs.
If AMD fails to mention and properly support this, they're just setting money on fire.
AMD has been failing to sell a lot of things.
Marketing has been godawful since Robert Hallock left, though.
AMD is now drastically changing their CPU/APU naming scheme every generation / 18 months.
He left for Intel as they were entering Raptor Lake and Alchemist...I think the man just enjoys pain.
That may be a prerequisite for working marketing at AMD.
That and being bald.
 
Reactions: Tlh97

Joe NYC

Platinum Member
Jun 26, 2021
2,951
4,329
106
And LLMs with >13B parameters and good enough tokens/s performance, which until recently was a task reserved to >$1500 discrete GPUs.
If AMD fails to mention and properly support this, they're just setting money on fire.

It would be interesting if some vendor out there offers Strix Halo with 2x LPCAMM2. There are already 64 GB modules, so 2x 64GB = 128 GB. Theoretical max is 256 GB.

256 GB happens to be more memory than any datacenter GPU. The highest is Mi300x with 192 GB.

Marketing has been godawful since Robert Hallock left, though.
AMD is now drastically changing their CPU/APU naming scheme every generation / 18 months.

I am not a big fan. But there was a fan in the picture, when a big brown object hit it, in Q4 2021, for AMD brand, Hallock "celebrated" AMD meltdown in client with this god-awful and tone-deaf video:

Curiously, things started to turn around at AMD (for the better) after his departure, but that may just be a coincidence

 
Reactions: Tlh97

Mahboi

Golden Member
Apr 4, 2024
1,057
1,969
96
This type of packaging is quite low cost. It has been used for a long time in low cost smart phones, so the cost should be in low single digits.

Ther is also savings on die area by elimination of SerDes and power savings. So Strix Halo packaging is an asset, not a liability

I don't think the cost of packaging is going to be a particular problem.
Disagree, I heard that many times and yet the advantage of packaging in price has been seriously offset except in very expensive I.E enterprise products.
Also everyone assumes that the "type of packaging" will be InFO apparently, I don't believe so.
One other variable to keep in mind is that low end dGPU such as x5x, x6x have 4 GB, 6 GB or 8 GB of memory, and the user is stuck with it for the rest of the life of the laptop.

Strix Halo will likely have 32 GB as a base configuration, leaving pretty much unlimited memory for the GPU in any game.

So, a good reply to dGPU marketing would be: is your game running dog slow with 6GB of memory?
32? I thought it was 16, but 32 unified LPDDR5 makes more sense.
Still, it's a bit of a one trick poney to market, since that's all that AMD has been having an advantage with GPU wise.
For something like Halo you want to make something new and special shine out.
People still buy notebooks with dGPU, regardless of whether they need dGPUs, whether you think they need dGPUs.

Strix Halo offers a better choice to people who buy dGPUs. That's the target market.

For people who currently don't buy notebooks with dGPUs - Strix Halo is not the best fit.
My point was that you're not going to have a very large market if you push the cost of a fairly large GPU on people, or you're not going to have a large margin if you don't.
I think the biggest challenge is that AMD is not NVidia. AMD does not have the same brand name, brand power in laptops.

If NVidia came up with Strix Halo, there would be zero doubt that this is the most brilliant idea. People would automatically buy it. Reviewers would be falling over each other recommending it.

That's the challenge for AMD to overcome. Not technical challenges of Strix Halo.
And we're back to the product being hard to market correctly, and its price making it somewhat prohibitive for good margins.
Surely it is high risk, but having brand recognition and brand power has high rewards. Which is what AMD is trying to do, after things fell apart.
Agreed, this is what they lack the most. Intel is a dying brand and it's going to cost them everything. AMD has a poor but rising brand in Ryzen. Radeon though is still seen as the sick child of the company, and that's something that needs to come with Halo: a big old image fix. A Halo product helps a lot, but they need to brand it and themselves properly, something which they've failed at since I've been paying attention to them.
 
Reactions: Joe NYC

ToTTenTranz

Senior member
Feb 4, 2021
387
755
136
It would be interesting if some vendor out there offers Strix Halo with 2x LPCAMM2. There are already 64 GB modules, so 2x 64GB = 128 GB. Theoretical max is 256 GB.

256 GB happens to be more memory than any datacenter GPU. The highest is Mi300x with 192 GB.

Just 128GB would be enough to run inference on the Falcon 180B parameter model which is close to GTP-4.

Having a Falcon 180B running on a reasonably sized laptop is mindboggling.


And that's better than starving on bandwidth and power.

If full Halo with 256bit + 2x 8c CCD + 20WGP starts at 40W, then cut-down Halo 128bit + 1x 8c CCD + 10WGP can probably start at 25W. At the same time, those 32MB would do wonders at 720p / 1080p base resolution.

And if you're not willing t to feed 25-30W to your handheld then you should just get the Steam Deck OLED.
 
Reactions: Tlh97

Aapje

Golden Member
Mar 21, 2022
1,530
2,106
106
So simply being male means one is childish? That's the stupidest thing I've read lately. Maybe you've been watching too much TV where all the commercials depict dads as being buffoons.
No, I was referring to how many women seem to see male gaming to explain their buying decisions. My point is that it's exactly their 'I don't want to be like that' opinions, which is also visible in those commercials where men are shown as buffoons and women as the responsible ones, that make women less likely to buy dedicated or high end gaming hardware.
 

soresu

Diamond Member
Dec 19, 2014
3,698
3,034
136
Radeon though is still seen as the sick child of the company
Perhaps it needs a rebrand after all this time.

Just as with Athlon -> Phenom -> FX -> Ryzen.

Radeon has a very long lineage at this point, which is both in its favor and against it.

You have the established name, but also all the baggage that comes with failed products, and times that the drivers were worse than just subpar on OGL and DX11.
 

soresu

Diamond Member
Dec 19, 2014
3,698
3,034
136
No, I was referring to how many women seem to see male gaming to explain their buying decisions
IMHO it's definitely a cultural and environmental thing, and one that I would argue is definitely shifting with the times - slower than with boys/men, but certainly happening.

I've met plenty of gamer girls during my recent uni years who spent far more time at a PC or in front of a games console than doing more stereotypically female pursuits.
 

soresu

Diamond Member
Dec 19, 2014
3,698
3,034
136
FX only died because of an absolute catastrophe.
But Phenom (Agena/Phenom1's initial problems aside) was not - so why not just reboot Phenom, which IMHO was a decent brand name.
Nothing the likes of which GPU portion of the company has ever seen.
Ehhhh.....

Not to the extreme extent of FX, but arguable at the very least that it hasn't experienced serious problems in the past.

HD 2xxx was an unmitigated disaster and 3xxx only recovered a little bit from that - 4xxx was the first truly embraced product of the Terascale line, and then due to node problems 6xxx barely had any SKUs at all.

Then we have the initially promising GCN based line with 7xxx that rapidly seemed to circle the drain with each passing generation with Vega HBM seeming like more of a distracting novelty when the actual performance was displayed.
 

adroc_thurston

Diamond Member
Jul 2, 2023
5,413
7,588
96
But Phenom (Agena/Phenom1's initial problems aside) was not - so why not just reboot Phenom, which IMHO was a decent brand name.
They basically jettisoned their CPU biz before so new name had to come in.
Then we have the initially promising GCN based line with 7xxx that rapidly seemed to circle the drain with each passing generation with Vega HBM seeming like more of a distracting novelty when the actual performance was displayed.
Pre-2014 stuff is ancient history neither zoomers nor your average shill foundry talking head is interested in.
All AMD needs is the chainsaw and some fat bribes to make the shills sing.
You gotta astroturf your brand into existence. Worked for Ryzen, worked for geforce rtx™.
 

Aapje

Golden Member
Mar 21, 2022
1,530
2,106
106
IMHO it's definitely a cultural and environmental thing, and one that I would argue is definitely shifting with the times - slower than with boys/men, but certainly happening.

I've met plenty of gamer girls during my recent uni years who spent far more time at a PC or in front of a games console than doing more stereotypically female pursuits.

Yes, sure, mostly for younger women. But on the whole, I think that there are a decent number of women who would see a dedicated gaming desktop or console as a childish waste of money, while they can justify a light gaming laptop to themselves, much more than men who reason the same way. Men who buy such a laptop are more likely to have money limitations.
 
sale-70-410-exam    | Exam-200-125-pdf    | we-sale-70-410-exam    | hot-sale-70-410-exam    | Latest-exam-700-603-Dumps    | Dumps-98-363-exams-date    | Certs-200-125-date    | Dumps-300-075-exams-date    | hot-sale-book-C8010-726-book    | Hot-Sale-200-310-Exam    | Exam-Description-200-310-dumps?    | hot-sale-book-200-125-book    | Latest-Updated-300-209-Exam    | Dumps-210-260-exams-date    | Download-200-125-Exam-PDF    | Exam-Description-300-101-dumps    | Certs-300-101-date    | Hot-Sale-300-075-Exam    | Latest-exam-200-125-Dumps    | Exam-Description-200-125-dumps    | Latest-Updated-300-075-Exam    | hot-sale-book-210-260-book    | Dumps-200-901-exams-date    | Certs-200-901-date    | Latest-exam-1Z0-062-Dumps    | Hot-Sale-1Z0-062-Exam    | Certs-CSSLP-date    | 100%-Pass-70-383-Exams    | Latest-JN0-360-real-exam-questions    | 100%-Pass-4A0-100-Real-Exam-Questions    | Dumps-300-135-exams-date    | Passed-200-105-Tech-Exams    | Latest-Updated-200-310-Exam    | Download-300-070-Exam-PDF    | Hot-Sale-JN0-360-Exam    | 100%-Pass-JN0-360-Exams    | 100%-Pass-JN0-360-Real-Exam-Questions    | Dumps-JN0-360-exams-date    | Exam-Description-1Z0-876-dumps    | Latest-exam-1Z0-876-Dumps    | Dumps-HPE0-Y53-exams-date    | 2017-Latest-HPE0-Y53-Exam    | 100%-Pass-HPE0-Y53-Real-Exam-Questions    | Pass-4A0-100-Exam    | Latest-4A0-100-Questions    | Dumps-98-365-exams-date    | 2017-Latest-98-365-Exam    | 100%-Pass-VCS-254-Exams    | 2017-Latest-VCS-273-Exam    | Dumps-200-355-exams-date    | 2017-Latest-300-320-Exam    | Pass-300-101-Exam    | 100%-Pass-300-115-Exams    |
http://www.portvapes.co.uk/    | http://www.portvapes.co.uk/    |