Discussion Qualcomm Snapdragon Thread

[Joe NYC]

Theyve done quite fine without it thus far, I'd say.

AMD has 1, that is one in English, uno in Spanish, design win for Strix Halo.

 

[Joe NYC]

Of those 3, Intel has said it's a poor market fit for their customers (remember, you're the customer's customer). Apple is their own customer. And Qualcomm isn't all-in on it.

Last time I checked, there were 47 unique Lunar Lake models on the market vs. 1 Strix Halo model.

What would you rather take:
- 1 design win and OEM claiming to be happy
- 47 design wins and OEM claiming to be unhappy

I think it all BS, with OEMs raping customers on RAM and including garbage RAM on AMD laptops.

While @adroc_thurston mentioned that 8533 memory speed laptops are coming, including for Kraken, on the same day, I encountered a Kraken laptop with (garbage) 5600 LPDDR5 memory speed.

 

[gdansk]

Last time I checked, there were 47 unique Lunar Lake models on the market vs. 1 Strix Halo model.

And? What does that have to do with it? Intel sayeth

Intel stressed that Lunar Lake's MoP package-level memory design is a "one-off" solution. In the future Panther Lake, Nova Lake and its successors products, Intel will return to the traditional package.

They're not doing it again. Well, maybe in 2029. AMD and Intel are in the same boat here. They don't want inventory management and their customers do not want memory configuration tied to the SKU. Qualcomm is trying but they have only 1 (memory) configuration to avoid some of these problems.

 

[CouncilorIrissa]

While @adroc_thurston mentioned that 8533 memory speed laptops are coming, including for Kraken, on the same day, I encountered a Kraken laptop with (garbage) 5600 LPDDR5 memory speed.

That's DDR5 5600, not LPDDR5.
It's actually faster in CPU workloads.

 

[Joe NYC]

That's DDR5 5600, not LPDDR5.
It's actually faster in CPU workloads.

You may be right, I did not look at it in detail, just a glance.

But I don't think there are desktop AM5 based Kraken chips out yet, are there? And it would be quite strange to include it on a laptop.

 

[Joe NYC]

And? What does that have to do with it? Intel sayeth

Intel management is very scared of low margins, even if the dollar margins are the same. Intel is scared even of the perception.

AMD, OTOH, especially the new CFO Jean Hu, just loves margin dollars (whatever way you get to them), she does not have the same paranoia as Intel management.

Maybe she needs to have a talk with the product people.

They're not doing it again. Well, maybe in 2029. AMD and Intel are in the same boat here. They don't want inventory management and their customers do not want memory configuration tied to the SKU. Qualcomm is trying but they have only 1 (memory) configuration to avoid some of these problems.

The CPU vendors are scared to take the decisive (and correct) step that Apple took. Seems like a paralysis.

 

[gdansk]

Intel management is very scared of low margins

Consider it.
How would Strix Halo be any more attractive if AMD went the middle road like Qualcomm? There would be one single 64GB MoP STH SKU. A SKU even more irrelevant for gamers because it's more memory than they need and more irrelevant for MLturds because it's less than they want. With the on board approach, at least they can disappoint all three markets (gamers, imaginary, and ML) equally.

 

[Joe NYC]

Consider it.
How would Strix Halo be any more attractive if AMD went the middle road like Qualcomm? There would be one single 64GB MoP STH SKU. A SKU even more irrelevant for gamers because it's more memory than they need and more irrelevant for MLturds because it's less than they want.

The "middle of the road" approach is not completely crazy.

Suppose AMD can offer Strix Halo in FP10 socket with 2 of the 4 memory channels internal, other 2 external.

It would offer (even more than) full Strix Halo bandwidth and performance, and it would not require any extra work on part of OEMs than they do for any Strix Point or Kraken laptop.

The 2 internal channels can run at full speed of LPDDR5 that's on the market, and could be 32 and 64 GB memory sizes.

I bet there would be more than 1 Strix Halo design on the market with the "middle of the road" solution.

 

[gdansk]

I bet there would be more than 1 Strix Halo design on the market with the "middle of the road" solution.

Nope. No one at all wants to pay for that 300mm² RDNA3½ die. It isn't because of a memory configuration. MoP in no way saves turds

 

[adroc_thurston]

only AMD stubbornly shooting itself in the foot by refusing to adopt it.

They're doing the right thing. They always do the right thing.

They really should try to win the XBox contract to get more dev. feedback / support for their GPU drivers though...

You do understand that bidding for Xbox includes Actual Real graphics players (you know, NV/AMD).

Last time I checked, there were 47 unique Lunar Lake models on the market vs. 1 Strix Halo model.

What would you rather take:
- 1 design win and OEM claiming to be happy
- 47 design wins and OEM claiming to be unhappy

I think it all BS, with OEMs raping customers on RAM and including garbage RAM on AMD laptops.

how the hell are you comparing completely different market segments.
LNL is an ULT part, stxH is something closer to ye olde GT4e configs.

Suppose AMD can offer Strix Halo in FP10 socket with 2 of the 4 memory channels internal, other 2 external.

It would offer (even more than) full Strix Halo bandwidth and performance, and it would not require any extra work on part of OEMs than they do for any Strix Point or Kraken laptop.

The 2 internal channels can run at full speed of LPDDR5 that's on the market, and could be 32 and 64 GB memory sizes.

please stop doing this dementia it's bad for you.

Consider it.
How would Strix Halo be any more attractive if AMD went the middle road like Qualcomm? There would be one single 64GB MoP STH SKU. A SKU even more irrelevant for gamers because it's more memory than they need and more irrelevant for MLturds because it's less than they want. With the on board approach, at least they can disappoint all three markets (gamers, imaginary, and ML) equally.

are you still upset over stxH ASPs?
that's very sad my friend, very sad.

 

[Joe NYC]

Nope. No one at all wants to pay for that 300mm² RDNA3½ die. It isn't because of a memory configuration. MoP in no way saves turds

Maybe we should take it up in another thread and keep this one Qualcomm specific.

 

[Raqia]

You do understand that bidding for Xbox includes Actual Real graphics players (you know, NV/AMD).

Qualcomm being one as the biggest mobile GPU vendors whose custom GPUs are actually being sought after and used for playing mobile games as well as the dominant choice in VR.

 

[adroc_thurston]

Qualcomm being one as the biggest mobile GPU vendors

that's the problem.
Mobile IP sucks donkeyballs for playing Actual Real Videogames.

 

[gdansk]

Qualcomm being one as the biggest mobile GPU vendors whose custom GPUs are actually being sought after and used for playing mobile games as well as the dominant choice in VR.

When in comparison to the powerhouses of PowerVR and Mali, even Radeon has god-tier drivers.
I think a telling example is that even though Exynos 2400 is considered a garbage SoC on a weak process the Radeon GPU can emulate Switch games better than my 8 Elite S25U with or without turnip. The 8 Elite may have a lot more horsepower but it doesn't matter when it refuses to render as it should. Oh and you don't have to use custom drivers.

 

[Raqia]

I agree --horse power isn't the issue.

If anything they should have an edge in laptops with their experience implementing ultra efficient solutions because for anything smaller than a desktop, efficiency is performance. Lord knows my gaming laptop chokes on its own heat without a cooling stand and it's pointless to have a more powerful gpu than something like the 5060 in that form factor. (I happen to consider bulky gaming laptops a clunky, worst of both worlds compromise and would rather have both a desktop and a thin & light for less cost to boot...)

Drivers with shader replacements tailored to specific games and dev iteration \ feedback for the Windows platform is precisely what they are missing. They should offer up an SoC at close to cost for a console to get this going.

 

[adroc_thurston]

Horse power isn't an issue

Yeah it is, their caches alone are a joke.
It will never ever run any AAA title properly until they build an actual real microarch for it all.

 

[Hesperax]

https://twitter.com/x/status/1971317384096174166

Qualcomm seems to be in the Valve Deckard (Valve's next VR headset). Rather than Xbox, I would think the next logical step would be to try to convince Valve to make some flavor of an ARM Steam Deck.

They could make a lighter-weight device similar to the Playstation Portal. Designed to stream AAA Games from your Gaming PC (and maybe around 5W-8W for on-deivce usage). To differentiate from the Steam Deck 2.

 

[adroc_thurston]

I would think the next logical step would be to try to convince Valve to make some flavor of an ARM Steam Deck.

they tried.

 

[Raqia]

Yeah it is, their caches alone are a joke.
It will never ever run any AAA title properly until they build an actual real microarch for it all.

If you mean the pre 8xx series, yes. For low bandwidth phone dram, tiling and gmem are well suited for UIs with high overdraw. They lightly modified and shoehorned a mobile-first architecture onto a laptop PC for the first gen. X and it shows.

But they did do a uArch for more PC like loads and thermals: the 830 is fully capable of running AAA titles in its phone incarnation even and with emulation.

 

[adroc_thurston]

fully capable of running AAA titles in its phone incarnation even and with emulation.

stretching the definition of "fully capable" here.

 

[Doug S]

They really should try to win the XBox contract to get more dev. feedback / support for their GPU drivers though...

Are you sure there will be a next Xbox to win a contract for?

 

[adroc_thurston]

Are you sure there will be a next Xbox to win a contract for?

Yeah AMD already won it.

 

[Hesperax]

Snapdragon X2 Elite Extreme: Qualcomm mischt TSMC N3X und N3P für High-End-Chip

Qualcomm hat gegenüber ComputerBase bestätigt, beim Snapdragon X2 Elite Extreme die Nodes TSMC N3X und N3P innerhalb des Dies zu mischen.

www.computerbase.de

Elite Extreme is a mix of N3P and N3X

 

[adroc_thurston]

Snapdragon X2 Elite Extreme: Qualcomm mischt TSMC N3X und N3P für High-End-Chip

Qualcomm hat gegenüber ComputerBase bestätigt, beim Snapdragon X2 Elite Extreme die Nodes TSMC N3X und N3P innerhalb des Dies zu mischen.

www.computerbase.de

Elite Extreme is a mix of N3P and N3X

Hey, overdrive fins.
Finally someone not AMD used them.

 

[ToTTenTranz]

The 192b memory bus for the top SKU looks interesting. Good to see the continued march towards faster memory with both wider buses and high MT.

Isn't it more like the other way around?

For the past decade non-apple SoC designers kept pumping up CPU and GPU compute throughput like crazy for useless antutu memebench numbers, without a memory subsystem that could ever keep up in real life workloads.


This is just trying to make up for that mistake.