News Qualcomm’s Snapdragon X Elite for PCs Has 12 Oryon Cores, Tops Out at 4.3 GHz

[Admin]

Qualcomm unveiled its Snapdragon X Elite processor for Windows laptops using its Oryon CPU cores for the first time.

Qualcomm’s Snapdragon X Elite for PCs Has 12 Oryon Cores, Tops Out at 4.3 GHz : Read more

 

[thestryker]

It seems like these are aiming square at AMD/Intel laptops in the business and productivity area. I do wish they scaled up the memory bandwidth a bit, similarly to Apple, so that the graphics had more to work with. That being said these still ought to really put pressure on AMD/Intel to improve efficiency and hopefully that will have impact on the entire product stacks.

 

[JamesJones44]

The fireworks between ARM and Qualcomm might be getting a lot brighter after this announcement. I'll be curious if these processors actually ship if they can't resolve their differences by the ship date.

 

[ivan_vy]

year by year we keep hearing ARM new revolution in PC space yet never comes. Need to convince big OEM players first to make things happen.

 

[thestryker]

The fireworks between ARM and Qualcomm might be getting a lot brighter after this announcement. I'll be curious if these processors actually ship if they can't resolve their differences by the ship date.

They would need an injunction to stop them, and I don't believe they have one.

 

[JamesJones44]

They would need an injunction to stop them, and I don't believe they have one.

I think at this point ARM has been negotiating in "good faith", but ARM has threatened to cancel Qualcomm's license if Qualcomm continues to develop Nuvia based processors.

Arm replies to Qualcomm's claims of biz model changes, more

Softbank's chip designer warns US giant may lose all of its CPU licenses at this rate

www.theregister.com

 

[ThomasKinsley]

I'm not too sure about Qualcomm, but this is great news for ARM. I heard MediaTek has announced that they are getting into the desktop market as well. x86 has had a great run, but with all of the recent exploits (Spectre, Meltdown, Downfall Inception, Zenbleed, etc.) and scummy business practices, I am ready to move on, even if it's just to voice my frustration with the current trend.

 

[thestryker]

I think at this point ARM has been negotiating in "good faith", but ARM has threatened to cancel Qualcomm's license if Qualcomm continues to develop Nuvia based processors.

Arm replies to Qualcomm's claims of biz model changes, more

Softbank's chip designer warns US giant may lose all of its CPU licenses at this rate

www.theregister.com

It doesn't matter, because that would just start a massive new lawsuit and have no impact on Qualcomm's ability to produce these chips. The only way for Arm to stop this is by getting an injunction from the court which is very much unlikely and the AvQ trial date (last information I saw) isn't until Sept 2024 which will be well after these have shipped. Should Arm actually win the case Qualcomm might be in trouble to some degree, but how much would depend on the ruling.

 

[rluker5]

I'm not too sure about Qualcomm, but this is great news for ARM. I heard MediaTek has announced that they are getting into the desktop market as well. x86 has had a great run, but with all of the recent exploits (Spectre, Meltdown, Downfall Inception, Zenbleed, etc.) and scummy business practices, I am ready to move on, even if it's just to voice my frustration with the current trend.

I thought that ARM was also vulnerable to a lot of those as well. They just didn't get patches.

 

[MartianM]

It seems like these are aiming square at AMD/Intel laptops in the business and productivity area. I do wish they scaled up the memory bandwidth a bit, similarly to Apple, so that the graphics had more to work with. That being said these still ought to really put pressure on AMD/Intel to improve efficiency and hopefully that will have impact on the entire product stacks.

AMD have a strong roadmap for improving efficiency with Zen 4, 5 and beyond. Intel Meteor Lake (15th gen?) should bump both efficiency and iGPU performance & performance/watt.

What I find interesting is that Nuvia claimed they could design a better core than Apple, and with all the resources of Qualcomm, a 12x perf core chip can only do 50% better than an M2 4+4 chip. So 8 Nuvia cores matches 5-6 equivalent M2 perf cores, the M3 bumps this to 4-5 equivalent cores next week, and it isn’t out until mid next year. They are still multiple years behind Apple, just like in their phone performance cores.

 

[thestryker]

AMD have a strong roadmap for improving efficiency with Zen 4, 5 and beyond. Intel Meteor Lake (15th gen?) should bump both efficiency and iGPU performance & performance/watt.

While Qualcomm has said nothing about single core (I bet AMD is probably almost as good, or maybe even better here) the multicore comparison is very good. A lot of work will need to be done to close that gap in laptop processors.

Barring a miracle they won't be anywhere near the performance level Qualcomm is claiming here graphics wise. They're comparing with AMD's current top IGP, though with slow DRAM, and have a significant lead. Based on what I've seen DRAM scaling wise the AMD part stands to gain 10-20% performance which means Qualcomm would still be looking at 40-50% more performance. That isn't a gap Intel or AMD will be closing any time soon.

What I find interesting is that Nuvia claimed they could design a better core than Apple, and with all the resources of Qualcomm, a 12x perf core chip can only do 50% better than an M2 4+4 chip.

What we're seeing here is Nuvia's server core adapted to a laptop format. The next one ought to be showing a lot more of Qualcomm's design influence. I think it'll also depend on whether or not they continue to do high end Snapdragon chips based on Arm's reference design as well.

So 8 Nuvia cores matches 5-6 equivalent M2 perf cores, the M3 bumps this to 4-5 equivalent cores next week, and it isn’t out until mid next year. They are still multiple years behind Apple, just like in their phone performance cores.

This is pure guess work as we don't actually know how good Apple's performance cores are in isolation. Not to mention nobody knows what sort of performance increase lies with M3. That's not to say you're wrong just that we have no actual information yet.

Qualcomm also doesn't have to compete with Apple in this space just AMD and Intel. They need to have a product that offers real advantages without a price premium which may have impacted their design.

 

[ThomasKinsley]

I thought that ARM was also vulnerable to a lot of those as well. They just didn't get patches.

Perhaps someone else can provide additional detail, but from what I understand, ARM is affected by Spectre and, to a lesser degree, Meltdown. Downfall is Intel-only. Zenbleed and Inception are AMD-only.

 

[KraakBal]

That's funny, my laptop can do multiple weeks of battery life if i open it once a week for a minute

 

[Order 66]

In terms of TFLOP comparisons the graphics capabilities are a little worse than an RX 470, which I feel would be pretty good for a mobile chip, but for a desktop chip, it isn't that impressive. if this kind of performance was in something like a steam deck, that would be impressive. Power draw would probably be too much for something like a steam deck though.

 

[bit_user]

It claims the Snapdragon X Elite's GPU offers approximately twice the performance of the competition in 3D Mark WildLife Extreme at one-fourth the power, while it claimed 80% improvement at one-fifth the power against a different integrated GPU.

No, @AndrewFreedman - the correct way to read this chart:

​

...is twice the performance or one fourth the power! It's a big difference! In the first case, they're holding power consumption equal. In the second, they're holding performance equal. That's what the horizontal and vertical lines are all about. You do not get twice the performance at one fourth the power!!

Obviously, the same goes for the other such chart.

 

[bit_user]

I do wish they scaled up the memory bandwidth a bit, similarly to Apple, so that the graphics had more to work with.

Huh? They have like 1.5x the bandwidth of the base M2 and 2/3rds as much as the M2 Pro (source).

​

Considering the size of its GPU, I think that's probably enough. Note how they didn't compare their graphics to Apple's - only Intel's and AMD's.

That being said these still ought to really put pressure on AMD/Intel to improve efficiency and hopefully that will have impact on the entire product stacks.

Yes, more focus on efficiency is good. Perhaps it even has something to do with AMD's rumored work on their own ARM cores.

Nvidia and AMD to Develop Arm CPUs for Client PCs: Report

A major challenge for Intel?

www.tomshardware.com

They would need an injunction to stop them, and I don't believe they have one.

Yeah, but when the facts change (i.e. Qualcomm actually starts shipping), that might enable them to (re-) apply for one. I think.

Then again, I think injunctions tend to be issued only when irreparable harm is being done. In this case, the only harm is that Qualcomm is shipping chips without (potentially) having obtained a license. So, I think that wouldn't count as irreparable, since Qualcomm could simply pay ARM for the license (+ punitive damages) if they lose the case.

 

[bit_user]

year by year we keep hearing ARM new revolution in PC space yet never comes. Need to convince big OEM players first to make things happen.

Well, it's been hard for ARM to catch x86, because even though ARM has been improving faster, x86 has remained a moving target.

If this is anywhere close to as good as they're saying (and remember, these are the main CPU designers from Apple), it should be a game changer.

 

[bit_user]

I think at this point ARM has been negotiating in "good faith", but ARM has threatened to cancel Qualcomm's license if Qualcomm continues to develop Nuvia based processors.

What does it even matter, if ARM is correct that Qualcomm's license doesn't apply to Nuvia's IP? Qualcomm disbanded their prior core design team years ago. Nuvia is literally the only reason Qualcomm needs an architectural license.

 

[bit_user]

x86 has had a great run, but with all of the recent exploits (Spectre, Meltdown, Downfall Inception, Zenbleed, etc.) and scummy business practices, I am ready to move on, even if it's just to voice my frustration with the current trend.

Some of those hit ARM cores, too. Not as many, in part because most ARM CPUs don't have SMT. Also, there might be something to be said for the simplicity of ARM's ISA making it easier to avoid security bugs. However, don't get complacent!

I thought that ARM was also vulnerable to a lot of those as well. They just didn't get patches.

Where'd you hear they're unpatched?

 

[bit_user]

AMD have a strong roadmap for improving efficiency with Zen 4, 5 and beyond. Intel Meteor Lake (15th gen?) should bump both efficiency

To the extent those depend on process node improvements, the same would apply to ARM. In other words, as long as you hold parameters like the process node & core size constant, x86 will never match ARM on efficiency.

What I find interesting is that Nuvia claimed they could design a better core than Apple, and with all the resources of Qualcomm, a 12x perf core chip can only do 50% better than an M2 4+4 chip. So 8 Nuvia cores matches 5-6 equivalent M2 perf cores

Don't assume either that all of the benchmarks are fully multithreaded or that Qualcomm can scale linearly.

However, working against Qualcomm is the fact that Apple makes the M2 on TSMC N5, whereas the Snapdragon X Elite is set to be made on TSMC N4. So, that 50% number already factors in a slight process advantage.

 

[JamesJones44]

What does it even matter, if ARM is correct that Qualcomm's license doesn't apply to Nuvia's IP? Qualcomm disbanded their prior core design team years ago. Nuvia is literally the only reason Qualcomm needs an architectural license.

They would need it for their mobile processors wouldn't they? I don't think they are shipping bare bones ARM designs for Snapdragon are they? (honest question I haven't looked that closely at them).

 

[thestryker]

They would need it for their mobile processors wouldn't they? I don't think they are shipping bare bones ARM designs for Snapdragon are they? (honest question I haven't looked that closely at them).

They are Arm designs they haven't been in house for years (somewhere early in the Snapdragon 800 series was the last one).

 

[thestryker]

Huh? They have like 1.5x the bandwidth of the base M2 and 2/3rds as much as the M2 Pro (source).

Yes, but they're not scaling it at all which is what I mean. This is obviously just the first chip and doesn't mean anything for future ones, but I have a hard time believing a GPU with the level of absolute performance they're claiming wouldn't benefit from more bandwidth.

This SoC seems designed to have broad market appeal though so it makes sense to minimize die area. They certainly don't need more performance (assuming their claims are accurate) to soundly beat their competition.

Yeah, but when the facts change (i.e. Qualcomm actually starts shipping), that might enable them to (re-) apply for one. I think.

Then again, I think injunctions tend to be issued only when irreparable harm is being done. In this case, the only harm is that Qualcomm is shipping chips without (potentially) having obtained a license. So, I think that wouldn't count as irreparable, since Qualcomm could simply pay ARM for the license (+ punitive damages) if they lose the case.

Exactly why I don't think Arm can do anything about these chips unless they win the case. Even then I'm not sure a judge would grant them an injunction (assuming Qualcomm wanted to appeal rather than pay up). Of course I'd be surprised if Qualcomm didn't have new chips coming out before the case was over period given how long these things seem to go.

 

[bit_user]

They would need it for their mobile processors wouldn't they?

Not an architecture license. That's only needed if you want to design your own cores. Qualcomm hasn't done that, in their phone SoCs, since the Snapdragon 820 (2015) & 835 (2017). They finished (and then immediately killed) Centriq, their server CPU, in 2018. I've seen the demise of Centriq get blamed on the Broadcom hostile takeover bid. I'm not sure if it's true, but nobody seems to contradict it.

I don't think they are shipping bare bones ARM designs for Snapdragon are they? (honest question I haven't looked that closely at them).

They make it confusing by slapping their own names on the CPU cores in Snapdragons, but they're indeed just ARM-designed. The only partial-exception is when the Cortex-X1 launched, as that was something of a customized job ARM did for Qualcomm. I guess it turned out so well that ARM decided to make the X-tier a normal thing.

Here's a mapping from their Kryo numbers -> ARM's Cortex names:

Kryo - Qualcomm - WikiChip

Kryo is a series of semi-custom ARM cores from Arm integrated by Qualcomm in their Snapdragon SoCs.

en.wikichip.org

Unfortunately, it needs to be updated. This article covers some of the newer models, as well:

How to understand Kryo CPU numbering in Qualcomm Snapdragon processors

Qualcomm's Snapdragon processors use Kryo CPUs. These CPUs have their own number system, separate to the Snapdragon numbering. Confused? Don't be.

www.androidauthority.com

 

[bit_user]

Yes, but they're not scaling it at all which is what I mean. This is obviously just the first chip and doesn't mean anything for future ones, but I have a hard time believing a GPU with the level of absolute performance they're claiming wouldn't benefit from more bandwidth.

I trust Nuvia to know what they're doing. I believe they only left Apple after finishing the M1 range. So, let's wait and see what comes along after this first SoC.