Discussion AMD's Future APU Gone ARM?

[Doug S]

Ironically Mini ITX was created by VIA, a relatively minor CPU manufacturer vs Intel and AMD.

It's not impossible that an OEM like Dell could create a new standard for mobos much as they did for RAM, or Intel could just decide to have another crack at it themselves.

Mini ITX was picked up because that was all there was for SFFs, but the selection is tiny compared to ATX/mATX. A serious SFF standard would drop ATX and have a single 24v DC coax input and rely on VRMs to translate voltages as needed on the board. Then you do away with the power supply in the case and use a cheap 24v brick to power it. It would drop all the legacy crap with those fiddly little headers for connecting front ports and LEDs, and stick an RJ45 port or two on the board to handle those connections to the case. That sort of thing.

But there's no one with authority to say "hey that's a good idea" - a board maker and a case maker might both recognize that dropping headers for RJ45 makes sense from an ease of use and cost perspective, but they can't implement that because unless all or at least most boards and cases support it you're just selling this weirdo product with an RJ45 that requires you make/buy some Frankenstein cable to connect it to the legacy headers on the other end.

What incentive does Dell have to create a mobo standard with DIYers in mind? Dell doesn't sell motherboards, they sell PCs.

 

[Shivansps]

If you are so confident with your Mali theory, then the upcoming Exynos SoC is not related to AMD ARM APU. Why not you start a thread dedicated to Samsung ARM SoC with Mali GPU? I am sure based on non-existence agreement and non-mentioning Mali should be great help. Go ahead, I am waiting...

We just need ONE ARM thread and be done with it, just like with RISC.

 

[SarahKerrigan]

We just need ONE ARM thread and be done with it, just like with RISC.

Considering the large spectrum of ARM SoC's available and in development, that sounds like a lot. Would you advocate for only one x86 thread?

(Also, can we have a permanent Itanium thread? Please? For me?)

 

[NTMBK]

Considering the large spectrum of ARM SoC's available and in development, that sounds like a lot. Would you advocate for only one x86 thread?

(Also, can we have a permanent Itanium thread? Please? For me?)

We did have an Itanic thread, but it hit an iceberg.

 

[SarahKerrigan]

We did have an Itanic thread, but it hit an iceberg.

Oi. Too soon. Kittson was canceled eleven years, two months, and seventeen days ago and I'm still in mourning.

 

[igor_kavinski]

Oi. Too soon. Kittson was canceled eleven years, two months, and seventeen days ago and I'm still in mourning.

Do you have an Itanium server or system board to play with?

What's the highest core CPU you own and which one is it?

 

[SarahKerrigan]

Do you have an Itanium server or system board to play with?

What's the highest core CPU you own and which one is it?

I have "only" one Itanium system at this point, an rx2800 i4 (final generation HP rack system.)

Highest core count CPU is a TileGX that's been gathering dust for a few years.

 

[Shivansps]

Considering the large spectrum of ARM SoC's available and in development, that sounds like a lot. Would you advocate for only one x86 thread?

(Also, can we have a permanent Itanium thread? Please? For me?)

X86 is just AMD and Intel... You really want to have one thread per ARM SoC or company?

 

[SarahKerrigan]

X86 is just AMD and Intel... You really want to have one thread per ARM SoC or company?

Also VIA/Centaur/Zhaoxin, who have a thread here too! Do not forget them! They are pleasant and have a rich history!

 

[FlameTail]

Is nobody talking about this AMD ARM SoC that was announced recently?

AMD Announces Versal Gen 2 Adaptive SoCs - AI Focused & Newer Arm Cores - Phoronix

www.phoronix.com

The new AMD Versal Gen 2 SoCs aim to provide up to 3x higher TOPs per Watt than their first-generation devices. On the CPU side, they tout the "new high-performance integrated Arm CPUs" as up to 10x more scalar compute than earlier Versal SoCs. With Versal coming from their Xilinx acquisition, they continue making use of Arm SoCs and with the Gen 2 SoCs are now using Cortex-A78AE processor cores.

 

[SarahKerrigan]

Is nobody talking about this AMD ARM SoC that was announced recently?

AMD Announces Versal Gen 2 Adaptive SoCs - AI Focused & Newer Arm Cores - Phoronix

www.phoronix.com

Xilinx gonna Xilinx.

At least AMD doesn't seem to be running it into the ground like Intel did with Altera (but all things in time...)

 

[gdansk]

When 10nm slipped and broke its own back Altera became a pointless acquisition.
AMD never intended to make Xilinx into a captive customer by acquisition.
AMD also remembers that it survived two disasters (K5 and Bulldozer) by being somewhat diversified.

 

[Shivansps]

Is nobody talking about this AMD ARM SoC that was announced recently?

AMD Announces Versal Gen 2 Adaptive SoCs - AI Focused & Newer Arm Cores - Phoronix

www.phoronix.com

It is an AI SoC, this is not going to be anywhere near the consummer market.

That said its interesting of what im seeing here, 8xA78AE cores and... Mali G78AE? If they already playing around with that they could launch a consumer soc based on the consumer version of that.

 

[moinmoin]

At least AMD doesn't seem to be running it into the ground like Intel did with Altera (but all things in time...)

Xilinx leadership is essentially taking over many parts of AMD so I wouldn't worry about Xilinx.

 

[FlameTail]

Xilinx leadership is essentially taking over many parts of AMD so I wouldn't worry about Xilinx.

So should we worry about AMD then?

 

[moinmoin]

So should we worry about AMD then?

I wouldn't think so. Last I checked Xilinx is essentially taking over embedded (duh), R&D and software development (traditionally AMD's big weakness relative to their products).

 

[Nothingness]

I think in the past AMD has been successful to integrate its acquisitions (think ATI). Intel buys companies and just destroys them (I know people in two companies/divisions that were bought by Intel, that wasn't pretty at all and everything they did just got trashed).

 

[igor_kavinski]

Intel buys companies and just destroys them

Because the influential executives there are arrogant and they suffer from "Not Invented Here" syndrome. Any success from a non-Intel team could jeopardize them politically inside the company.

 

[NTMBK]

Is nobody talking about this AMD ARM SoC that was announced recently?

AMD Announces Versal Gen 2 Adaptive SoCs - AI Focused & Newer Arm Cores - Phoronix

www.phoronix.com

Xilinx made ARM SoCs before acquisition, and they continue to do it after acquisition. They've been doing it for over a decade.

 

[SarahKerrigan]

Because the influential executives there are arrogant and they suffer from "Not Invented Here" syndrome. Any success from a non-Intel team could jeopardize them politically inside the company.

Combine that with internal political feuds on a scale that makes IBM look mild. (I don't know if this is necessarily still the case, but it certainly used to be, and I've seen nothing to suggest that it's changed.)

 

[marees]

Mendocino would be a good example of something you can replace with an first attempt at a ARM SOC. They kinda need to for low power devices, the only thing stopping something that is already old now, the RK3588, from entering the low end notebook and pc market at the low end is the lack of drivers, otherwise it would be there already.
Next gen ARM SoC may have gpus that may rival at least some of the RMB RDNA2 SoCs (G710/G720).

So it could be possible if they think they cant make a x86 SoC cheap and low power enoght.

Sonoma Valley (4x zen 5c 4? CU RDNA 3.5? on samsung 4nm) is the replacement for mendocino right

 

[Ghostsonplanets]

Sonoma Valley (4x zen 5c 4? CU RDNA 3.5? on samsung 4nm) is the replacement for mendocino right

Yes. Same platform, new SoC. So it's a drop-in replacement that can be slotted into current Mendoccino designs.

 

[Ghostsonplanets]

Huh...This came out of nowhere, but...

https://twitter.com/x/status/1786553716268556727

I'm not sure how reliable Roland Quant still is nowadays, but @Tigerick @SpudLobby maybe this answer how will future Exynos designs be able to release on Mobile and PC?

Umm... so there's S5E9945 right? (Exynos 2400)
And then there's S5E9955, which follows after that.
Exynos 2400 has an AMD GPU. The follow-up will have AMD GPU, too, it seems.

After that, in 2026, Samsung seems to aim for the release of a next-gen flagship SoC feat it's own GPU.

Still, I'm leaning towards false for this one as it doesn't make sense to me. Samsung tried to create their own GPU in the past, but failed to do so and that was the catalyst for them to go after AMD for the IP licensing agreement.

 

[igor_kavinski]

failed to do so

That doesn't stop determined people/companies from trying again and again. Look at Intel and their obsession with wanting to have a performant GPU. Went from using Pentium cores to do raytracing with Larrabee to successively improved HD graphics to finally taking it very seriously and culminating with ARC. It's possible that Samsung is internally working on successively better prototype GPUs and their internal projections may point to them being ready by 2026.

 

[Nothingness]

That doesn't stop determined people/companies from trying again and again. Look at Intel and their obsession with wanting to have a performant GPU. Went from using Pentium cores to do raytracing with Larrabee to successively improved HD graphics to finally taking it very seriously and culminating with ARC. It's possible that Samsung is internally working on successively better prototype GPUs and their internal projections may point to them being ready by 2026.

Indeed. Intel's original sin was to try to use a CPU/ISA architecture to make a GPU. That's inefficient. There are many reasons why there are dedicated blocks to do various tasks (DSP, NPU, ISP, etc.) a CPU could in theory do. Intel lost years trying with Larrabee. But that doesn't mean they won't finally succeed in designing a good GPU (I didn't follow, but isn't ARC acceptable as a mid-range GPU?).