Discussion All Custom APUs Discussion Threads

[Tigerick]

Qualcomm's ARM SoC Lineup

ARM/x86 SoC	Date	Node	Die Size (mm2)	Default TDP	CPU	GPU	FP32 (FLOPS)		Model	RAM + SSD	LCD	Battery	Weight	SRP
G3x Gen 2	Q4 2023	N4		15 W	X3 3.36GHz 8-core	A32 1GHz	3072		AYANEO Pocket S	12 + 128	1440p 6" IPS LCD	6000mAh	350 g	$589
8 Gen 2	Q4 2022	N4	118.4	8.5 W	X3 3.2GHz 8-core	A740 680MHz	2089		Retroid Pocket 6	12 + 256	1080p@120Hz 5.5" AMOLED	6000mAh	320 g	$259
G2 Gen 2	Q1 2025	N4P			X4 2.8GHz 8-core	A22			Retroid Pocket G2	8 + 128	1080p@60Hz 5.5" AMOLED	5000mAh	280 g	$219
7+ Gen 3	Q1 2024	N4P	89.54	5.5 W	X4 2.8GHz 8-core	A732 950MHz	1459
G3 Gen 3	Q1 2025	N4P			X4 3.3GHz 8-core	A33 1.05GHz	3226 ?		KONKR Pocket Fit	12 + 256	1080p@144Hz 6" LCD	8400mAh	386 g	$329
8 Gen 3	Q4 2023	N4P	137.3	8 W	X4 3.3GHz 8-core	A750 903MHz	2774		AYANEO Pocket S2	12 + 256	1440p 6.3" IPS	8000mAh	428 g	$499
									Steam Frame	16 + 256	Dual 2160 x 2160 LCD	21.6WHr	440 g	?
8 Elite Gen 4	Q4 2024	N3E	124.1	8 W	Oryon 4.32GHz 8-core	A830 1.1GHz	3379		KONKR Pocket Fit Elite	12 + 256	1080p@144Hz 6" LCD	8400mAh	386 g
									AYN Odin 3 Pro	12 + 256	1080p@120Hz 6" AMOLED	8000mAh	390 g	$399
									AYN Odin 3 Max	16 + 512	1080p@120Hz 6" AMOLED	8000mAh	390 g	$449
8 Elite Gen 5	Q3 2025	N3P	126.2		Oryon 4.61GHz 8-core	A840 1.2GHz	3686
8 Elite Gen 6 Pro?	Q4 2026	N2P					~5000		Steam Deck 2 ?
AMD Aerith+	Q4 2023	N6	131	~15 W	Zen 2 3.5GHz 4-core	RDNA2 8CU 1.6GHz	1600		Steam Deck OLED	16 + 512	720p@90Hz 7.4" OLED	6470mAh	640 g	$549
										128-bit LPDDR5
AMD Z2 A	Q1 2025	N6	131	15 W	Zen 2 3.5GHz 4-core	RDNA2 8CU 1.8GHz	1800		ROG Xbox Ally	16 + 512	1080p@120Hz 7" LCD	60WHr	670 g	$599
AMD Z2 Extreme AI	Q1 2025	N4P	233	28 W	Zen 5 5GHz 8-core	RDNA3.5 16CU 2.9GHz	5900 SI		ROG Xbox Ally X	24 + 1024	1080p@120Hz 7" LCD	80WHr	715 g	$999

Next Gen XBox Consoles and Handheld Preliminary Specs

	ROG Xbox Ally / SteamDeck OLED	PS6 Handheld ?	ROG Xbox Ally X		PS5	Steam Machine	Xbox Series S		Xbox Series X	X2 ?		PS5 Pro	PS6 ?
Date	2025	Fall 2027 ?	2025		2020		2020		2020			2024	Fall 2027?
Codename	Aerith Plus	Canis	Strix Point							Magnus		Custom	Orion
Model	Z2A		Z2 Extreme AI
	N6	3 nm	N4P		N6	N4 + N6	N7		N7P	N3P + N3P ?		N4P	3nm Single Die
		135 mm2 ?	233 mm2		260 mm2		197 mm2		360.45 mm2	144 + 264 = 408 mm2		279 mm2	280 mm2
CPU	4 x Zen 2	4 x Zen 6c + 2 x Zen 6 LP	4 x Zen 5 + 4 x Zen 5c		8 x Zen 2	6 x Zen 4	8 x Zen 2		8 x Zen 2	3 x Zen 6 + 8 Zen 6c 12MB L3		8 x Zen 2	7 x Zen 6c + 2 x Zen 6 LP 8-12 MB L3?
						16GB DDR5
GPU	RDNA2 8CU	RDNA5 16 CU	RDNA3.5 16CU		RDNA2 36CU	RDNA3 28CU SI?	RDNA2 20CU		RDNA2 52CU	RDNA5 68CU 12MB L2 ?		RDNA2 60CU	RDNA5 52CU 10MB L2
Memory	16GB 128-bit LPDDR5-6400	18GB 192-bit LPDDR5x-8533	24GB 128-bit LPDDR5x-8000		16GB 256-bit GDDR6	8GB GDDR6	8GB 128-bit+ 2GB 32-bit GDDR6		10GB 320-bit GDDR6 + 6GB 192-bit GDDR6	30GB 192-bit 36Gbps GDDR7 ?		16GB 256-bit GDDR6	20GB 160-bit 32Gbps GDDR7 ?
Memory Bandwidth	102 GB/s	204 GB/s	120 GB/s		448 GB/s		224 GB/s		560 GB/s	864 GB/s		576 GB/s	640 GB/s

 

[marees]

AMD not winning steak deck 2 is a real mistake imo, no where near as big as intel turning down Apple but it's of the same kind. Steam deck 2 having Qualcomm will encourage valve to spend more time making games run on ARM and supporting Qualcomm with their GPU drivers. which weakens the x86 hegemony in gaming and brings more capability to future laptop Qualcomm hardware. Serious strategic faux pas imo.

Qualcomm is in the vr headset

Steam deck 2 is probably AMD because unlike AMD, valve supports (older) radeon hardwares

 

[adroc_thurston]

Steam deck 2 is probably AMD because unlike AMD, valve supports (older) radeon hardwares

It is QCOM.
AMD told Valve to go pound sand.

 

[inquiss]

It is QCOM.
AMD told Valve to go pound sand.

Which is a strategic mistake imo

 

[DAPUNISHER]

Qualcomm is in the vr headset

Steam deck 2 is probably AMD because unlike AMD, valve supports (older) radeon hardwares

It'll be a few years before we see the Steam Deck successor https://www.techpowerup.com/342876/valve-steam-deck-2-not-until-50-performance-uplift

 

[adroc_thurston]

Which is a strategic mistake imo

Sorry pal 3m lifetime units is a joke.

 

[Tigerick]

It'll be a few years before we see the Steam Deck successor https://www.techpowerup.com/342876/valve-steam-deck-2-not-until-50-performance-uplift

Thanks for the link. And I quoted:

We're not interested in getting to a point where it's 20 or 30 or even 50% more performance at the same battery life

Clearly Valve is aiming for much bigger performance uplifts at the same or better battery life.

 

[marees]

Thanks for the link. And I quoted:


Clearly Valve is aiming for much bigger performance uplifts at the same or better battery life.

My guesstimate is zen 7 & rdna 5+ monolithic apu that replaces krackan. 2029 at the earliest

 

[inquiss]

Sorry pal 3m lifetime units is a joke.

It's not about the units it's about opening the door for ARM optimisation across the entire PC games ecosystem

 

[inquiss]

My guesstimate is zen 7 & rdna 5+ monolithic apu that replaces krackan. 2029 at the earliest

It's Qualcomm, Qualcomm don't make zen

 

[marees]

It's Qualcomm, Qualcomm don't make zen

How about qualcomm GPUs ??


Are they upto snuff ?

 

[adroc_thurston]

It's not about the units it's about opening the door for ARM optimisation across the entire PC games ecosystem

PC gaming ecosystem is Windows.

 

[inquiss]

PC gaming ecosystem is Windows.

And one of the advantages it has is full compatibility. It's not *the* reason arm on windows is failing but lack of game support and buggy drivers beyond Nvidia and AMD is a factor. This partnership allows valve to sharpen the alternative to windows

 

[ToTTenTranz]

One member said the APU is having different floorplan to remove iGPU and NPU, thus considered semi-custom.

Who said this?
I wonder if the iGPU on Phoenix2 / Hawk Point is really disabled.
Its ~2.5TFLOPs could be used to e.g. offload FSR3 framegen like we can already do with the Lossless Scaling app.

Odds are Valve simply paid for the Z1 SKU which has the iGPU intact.

Valve claims 6 times performance of SD which is having 1.6TF FP32. RDNA3 28CU @ 2.45GHz with DI should have 17.6TF; 11 times faster. My suspect is Valve/AMD is removing DI support, thus RDNA3 GPU @ 2.45GHz with SI should have 8.8TF: 5.5 times faster. Then it might considered semi-custom as well.

Valve could simply be using real performance numbers through benchmarks instead of theoretical throughput. In most games, RDNA3 has little to no IPC advantage over RDNA2.

It is QCOM.
AMD told Valve to go pound sand.

Nah.
Steam Deck 2 is AMD, probably a version of Medusa Premium.

AMD might have refused to do a semicustom for Valve at the moment, but this doesn't mean Valve can't use an off-the-shelf part.

Sorry pal 3m lifetime units is a joke.

So much of a joke that Valve just announced not one but two new products on the same ecosystem.

Besides, weren't you super apologetic about Strix Halo failing to get design wins because it was "a new swimlane"? Why is this not a valid excuse for the Steam Deck?

 

[adroc_thurston]

Steam Deck 2 is AMD, probably a version of Medusa Premium.

It's Qualcomm.
Deck doesn't have the budget for a part of thah calibre.
Which is why Valve e-begged for a semi-custom part.

So much of a joke that Valve just announced not one but two new products on the same ecosystem.

A Quest 3 bootleg and a really bad NUC? uhhhhh.

Besides, weren't you super apologetic about Strix Halo failing to get design wins because it was "a new swimlane"? Why is this not a valid excuse for the Steam Deck?

Halo is margin accretive.
Cheapo low BOM handhelds aren't.

 

[ToTTenTranz]

I It's Qualcomm.

Nah. It's AMD.

Deck doesn't have the budget for a part of thah calibre.

Medusa Premium is a 128bit LPDDR5X / 192bit LPDDR6 part. It's not a massive departure from other 128bit APUs which is the most probable reason why that new chip exists.

Sure, MDS Premium will be more expensive than Van Gogh, but Valve's top selling part has always been their most expensive one so that's where they're more likely to scale.

A Quest 3 bootleg and a really bad NUC? uhhhhh.

Someone's jealous.

Cheapo low BOM handhelds aren't.

And no matter how many times people try to explain Valve's main KPI for the Deck isn't the handheld itself but the increased software attachment ratio it brings, you'll still pretend to not know this.

 

[adroc_thurston]

Nah. It's AMD.

It's not and hasn't been for a while.

Medusa Premium is a 128bit LPDDR5X / 192bit LPDDR6 part. It's not a massive departure from other 128bit APUs which is the most probable reason why that new chip exists.

Memory is only a fraction of the cost here.
It's just a pricey part.

Someone's jealous.

Projection is unhealthy.

 

[ToTTenTranz]

It's not and hasn't been for a while.

Nah. I 100% bet you have zero actual insight on this, and you're simply assuming/wishing AMD became too good for Valve despite having 10% of the gaming PC market (lol).

Qualcomm doesn't have anything remotely capable of doing 200-300% of Steam Deck's performance in AAA games, and x86 on ARM emulation is still crap on Qualcomm.
Valve knows this, which is why they're not even trying to sell the Steam Frame as a good standalone for SteamVR titles.

Memory is only a fraction of the cost here.
It's just a pricey part.

It's not "a new swimlane".
And AMD won't be able to make as much per-APU when Nova Lake and N1/N1X are out there. They sure as hell won't be able to push the humongous margins they got from the Strix family in 2024-25 while they had little to no competition.

Projection is unhealthy.

Exactly. You should look out for yourself.

 

[adroc_thurston]

and you're simply assuming/wishing AMD became too good for Valve despite having 10% of the gaming PC market (lol).

They told them to piss off so yeah they're too good for that.

And AMD won't be able to make as much per-APU when Nova Lake and N1/N1X are out there. They sure as hell won't be able to push the humongous margins they got from the Strix family in 2024-25 while they had little to no competition.

I suggest you watch their FAD and listen to their CFO instead of saying dumbest stuff ever. Might help.

 

[DavidC1]

SD2 being ARM would be a mistake. GPU and x86 support on Linux isn't going to be solved in just few years.

Nah. I 100% bet you have zero actual insight on this, and you're simply assuming/wishing AMD became too good for Valve despite having 10% of the gaming PC market (lol).

I have to agree with this. The way Gaben talks is very realistic. For example when asked about whether it's realistic in gaming, he said "who cares about realism in games"? and that fun should be the priority. 6x performance is probably real gaming performance as well.

Whatever advances in marketshare Steam Deck had will plummet on the ARM version.

It's not about the units it's about opening the door for ARM optimisation across the entire PC games ecosystem

It won't happen this way.

 

[DavidC1]

Valve's real goal is moving away from Windows.

Remember Gabe Newell used to work for Microsoft. Gaming is actually very important for Windows. He said that they realized that gaming is what allowed Windows to have high marketshare. The average joes using their computer for word processing or youtube watching do not need more than the cheapest, most boring hardware. Most of the passion towards PC started from gaming. The gamers who then gain knowledge becomes a force multiplier for sales. That comment about Windows was back in the days when he was developing Steam. He later says it was with Windows 8 how bad Windows was becoming. So up until that point Steam was basically to propagate Windows gaming. Now, it serves a different purpose.

So if Valve were suddenly going to shift to ARM based cores, then they would stop the gain they had with Steam Deck. Steam Deck has good compatibility because it's x86. And gaming and gamers is one area that is finicky as heck. Due to the reason they want to push SteamOS, I wouldn't be surprised if we don't ever see a Steam Deck 2. If selling hardware was their primary goal, we'd have had $1500 Steam Deck 4 now. Also despite the hardware being few years old, many people still recommend the Steam Deck. Because of the purpose build hardware and price makes it reachable for many versus, a $1500 competing device on Windows that bring all the warts.

They would reach their goals faster by working with 3rd party vendors for SteamOS and eventually for desktop/laptop PCs.

 

[inquiss]

Valve's real goal is moving away from Windows.

Remember Gabe Newell used to work for Microsoft. Gaming is actually very important for Windows. He said that they realized that gaming is what allowed Windows to have high marketshare. The average joes using their computer for word processing or youtube watching do not need more than the cheapest, most boring hardware. Most of the passion towards PC started from gaming. The gamers who then gain knowledge becomes a force multiplier for sales. That comment about Windows was back in the days when he was developing Steam. He later says it was with Windows 8 how bad Windows was becoming. So up until that point Steam was basically to propagate Windows gaming. Now, it serves a different purpose.

So if Valve were suddenly going to shift to ARM based cores, then they would stop the gain they had with Steam Deck. Steam Deck has good compatibility because it's x86. And gaming and gamers is one area that is finicky as heck. Due to the reason they want to push SteamOS, I wouldn't be surprised if we don't ever see a Steam Deck 2. If selling hardware was their primary goal, we'd have had $1500 Steam Deck 4 now. Also despite the hardware being few years old, many people still recommend the Steam Deck. Because of the purpose build hardware and price makes it reachable for many versus, a $1500 competing device on Windows that bring all the warts.

They would reach their goals faster by working with 3rd party vendors for SteamOS and eventually for desktop/laptop PCs.

Their goal is to move off windows, I agree. Having the headset be arm allows them to, slowly over time, improve ARM performance and compatibility in a non-core product ready for SD2. It's clear what's happening.

 

[adroc_thurston]

Their goal is to move off windows, I agree. Having the headset be arm allows them to, slowly over time, improve ARM performance and compatibility in a non-core product ready for SD2. It's clear what's happening.

There ain't no sikrit plan or whatever.
QC is just the only vendor making 5-10W XR parts (and generally this kinda low power stuff).

 

[LightningZ71]

Do people really think that the steam deck 1 would exist if someone else hadn't already paid for the development of the Zen2/RDNA2 chip that it uses?

 

[branch_suggestion]

And no matter how many times people try to explain Valve's main KPI for the Deck isn't the handheld itself but the increased software attachment ratio it brings, you'll still pretend to not know this.

AMD can still go after Gabecube 2 if it is a thing.

And AMD won't be able to make as much per-APU when Nova Lake and N1/N1X are out there. They sure as hell won't be able to push the humongous margins they got from the Strix family in 2024-25 while they had little to no competition.

Quite literally the opposite, NVL is pricey and losing in the metrics that matter.
GB10 is a lolcow SoC, a trve Tegra classic that is not just buggy and underperforming, but also has very dumb design decisions that make it bad at even the niche stuff you'd think NV would want right.
MDS1 should be a GM bump over KRK1, MDS1-hi a GM bump over STXP, MDSP a GM bump and simpler socket than STXH whilst being in a similar perf tier, MDSH is a big GM bump as it should compete against higher class green sticker machines.

 

[DavidC1]

Their goal is to move off windows, I agree. Having the headset be arm allows them to, slowly over time, improve ARM performance and compatibility in a non-core product ready for SD2. It's clear what's happening.

A VR headset is not going to move the needle in any direction of doing that. The VR headset allows you to basically wireless stream what's on your computer to the headset, which just feeds Steam, which feeds no-Windows idea. If you think the handheld market is small, a VR headset market is a fraction of this. The Frame doesn't even look that cheap either, and $400 is considered cheap for VR headsets.