News Snapdragon X Elite Outperforms Intel, AMD, Apple CPUs (In Vendor Benchmarks)

[Admin]

Qualcomm shares Snapdragon X Elite benchmarks, comparing it to Intel, AMD, and Apple processors.

Snapdragon X Elite Outperforms Intel, AMD, Apple CPUs (In Vendor Benchmarks) : Read more

 

[kaalus]

Why did they not compare against M2 Pro, which has similar TDP to Snapdragon? The vanilla M2 they tested against has half of the performance cores and half of the GPU cores of M2 Pro, so you can pretty much double Apple scores in all categories in the table. Suddenly, Snapdragon does not look so good.

And M2 is over a year old. In fact, M3 is likely coming out later today during the Apple event.

 

[bit_user]

Hey, where are the graphs?

From Anandtech:

Source: https://www.anandtech.com/show/2111...ormance-preview-a-first-look-at-whats-to-come
​

The Anandtech article has more, including iGPU performance.

Since the Snapdragon X Elite is made on TSMC N4, the comparison vs. Ryzen 9 7940HS is on the same node. The M2 is at a slight disadvantage, being made on TSMC N5. The M3 will use TSMC N3 (or N3E), which should make that matchup a lot more interesting.

 

[bit_user]

Why did they not compare against M2 Pro, which has similar TDP to Snapdragon?

The cynical interpretation would be that they wanted to make themselves look better. However, their decision would be defensible if they're planning on pricing closer to the M2 than M2 Pro (which I doubt - this is Qualcomm, after all).

 

[suryasans]

Hey, where are the graphs?

From Anandtech:

​

​

​

​

​

Source: https://www.anandtech.com/show/2111...ormance-preview-a-first-look-at-whats-to-come​

​

The Anandtech article has more, including iGPU performance.

Since the Snapdragon X Elite is made on TSMC N4, the comparison vs. Ryzen 9 7940HS is on the same node. The M2 is at a slight disadvantage, being made on TSMC N5. The M3 will use TSMC N3 (or N3E), which should make that matchup a lot more interesting.

This article is just a resume from Anandtech's article which is the first hand.

 

[bit_user]

This article is just a resume from Anandtech's article which is the first hand.

Anandtech didn't make the graphs, though. Those are slides from Qualcomm. No reason not to include them here.

 

[JamesJones44]

Why did they not compare against M2 Pro, which has similar TDP to Snapdragon? The vanilla M2 they tested against has half of the performance cores and half of the GPU cores of M2 Pro, so you can pretty much double Apple scores in all categories in the table. Suddenly, Snapdragon does not look so good.

And M2 is over a year old. In fact, M3 is likely coming out later today during the Apple event.

The Config B setup is basically identical to an M2 Max performance wise. 2780 SC (SXE) vs 2736 SC (M2 Max) and 14000 MC (SXE) vs 14493 MC (M2 Max).

The M2 Max manufactured on the N5P while the Snapdragon X Elite is on the N4. The N4 node is an enhanced 5 NM node so there isn't much difference there. I'm not surprised the numbers basically line up with an M2 Max (I guess I am a little because I don't believe the Snapdragon X Elite has the memory on package, but it does have 2 extra cores) .

Remember Apple has exclusivity rights to the TSMC's 3NM node until the end of the year ish (don't remember the exact date 3NM opens for other customers).

 

[Alvar "Miles" Udell]

That's all well and good, but the real question is how (and IF) it will perform with x86, being an ARM chip. Part of Apple's success with the M silicon is it has blanket support and the performance loss for x86 translation is minimal. If Snapdragon X Elite either doesn't work with x86 or has a decent performance hit, then it's not going to matter to anyone who doesn't use programs which have an ARM variant.

 

[Kamen Rider Blade]

Talk is cheap, I'll wait until independent reviews come out.

 

[ezst036]

Have any motherboards been announced?

A chip by itself is not usable.

 

[bit_user]

The M2 Max manufactured on the N5P while the Snapdragon X Elite is on the N4. The N4 node is an enhanced 5 NM node so there isn't much difference there.

TSMC N4 is supposed to have 6% higher density and approximately 4% better performance (at the same power), than N5.

Source: https://fuse.wikichip.org/news/6439...ily-with-a-new-enhanced-performance-n4p-node/
​

There's also N4P, which ups the performance advantage to 11%.

I'm not surprised the numbers basically line up with an M2 Max (I guess I am a little because I don't believe the Snapdragon X Elite has the memory on package, but it does have 2 extra cores) .

I'm not sure if they've said how it's connected, but they did say they're using LPDDR5X-8533, which I'm not sure you can do off-package. That's certainly faster than the LPDDR5-6400 the M2 uses. That's enough to net them 136 GB/s of bandwidth, at only 128-bit. It seems the main benefit of putting it on-package is really power savings, which you can turn around and spend on frequency, if you want the additional performance.

Unlike desktop PCs, phones have long been able to adjust their memory frequency on-demand. I'm sure their new laptop SoC is doing this with memory, as well. So, even though it might burn a lot of power at 8533 MT/s, it's only running that fast when under heavy load. The rest of the time, they can scale back and enjoy some power savings.

Remember Apple has exclusivity rights to the TSMC's 3NM node until the end of the year ish (don't remember the exact date 3NM opens for other customers).

That's irrelevant now, because Qualcomm said devices wouldn't be shipping until mid 2024. However, it probably factored into their decision to use N4, since I'm sure they expected to have it in shipping devices this year.

 

[bit_user]

That's all well and good, but the real question is how (and IF) it will perform with x86,

Why? The blockbuster games will probably have native ARM ports.

Mostly, I'll bet they're targeting the corporate market, where MS Office is already available for ARM. The corporate users are attractive because they value thin & light, long battery life, and good connectivity - all things which play into Qualcomm's strengths. Furthermore, they use a fairly limited set of software and any legacy x86 programs they use probably aren't very performance-intensive. Last, but not least, they have deep pockets and enough volume to sustain Qualcomm for a few generations.

 

[bit_user]

Have any motherboards been announced?

A chip by itself is not usable.

Microsoft has some NUC-like mini-PC for Qualcomm's SoCs. I think it's on its second refresh. They will doubtlessly refresh it for the new Snapdragon X Elite.

Windows Dev Kit 2023 (Project Volterra)

Powered by Arm64 and running Windows 11, this desktop device enables you to develop Windows apps for Arm, on Arm. Find device specifications, set up instructions, Arm-native developer tools, Support, and FAQs.

learn.microsoft.com

If you want this thing on an ATX board, that would be pretty pointless. It probably has no more than half a dozen PCIe lanes, if that. It really belongs either in a laptop or a mini-PC.

 

[rtoaht]

Meteor lake is around the corner. And when the snapdragon finally arrives sometime in 2024, there will be Arrow Lake. So the comparison with Raptor lake i7 is meaningless. It’s telling that they still had to choose an i7, not i9.

The other interesting part is, according to Qualcomm, i7 is faster in all the benchmarks compared to Ryzen 9. That’s a weird but interesting 3rd party data comparing Intel and AMD. I always prefer 3rd party reviews comparing AMD and Intel chips. Never thought that 3rd party would be Qualcomm. lol.

 

[bit_user]

Meteor lake is around the corner. And when the snapdragon finally arrives sometime in 2024, there will be Arrow Lake.

It's running late - not like Intel knows anything about that. Otherwise, Qualcomm would've obviously gone for TSMC N3 or N3E. I think it'll beat Arrow Lake (mobile) to market, but yes it'll launch against Meteor Lake and then have to match up against Arrow Lake. However, Qualcomm is no stranger to doing frequent refreshes, so a N3E version could launch maybe only 6-9 moths after.

So the comparison with Raptor lake i7 is meaningless. It’s telling that they still had to choose an i7, not i9.

They were going for a comparable power level & core count. It's not a mobile workstation or gaming CPU. My work laptop is a i7-12850HX. It's bulky, weighs a ton, has loud fans, and has awful battery life. I think it wouldn't be a very fair matchup to benchmark it against an i9, considering what different machines they are.

The other interesting part is, according to Qualcomm, i7 is faster in all the benchmarks compared to Ryzen 9.

Understand what you're comparing. The Ryzen is only 8 cores, 16 threads. The i7-13800H is 14 cores, 20 threads. So, it has a natural advantage on multithreaded benchmarks. Also, it's hardly news that Raptor Lake has better single-threaded performance than Zen 4. However, where AMD's Phoenix stands out is on efficiency.

 

[NinoPino]

That's all well and good, but the real question is how (and IF) it will perform with x86...

x86 means nothing on the goodness of the CPU design, and apparently Qualcomm have done a good job.

 

[bit_user]

apparently Qualcomm have done a good job.

Eh, not exactly Qualcomm. This CPU came from a company called Nuvia. Worth reading about, if you don't know them.

​

Qualcomm to Acquire CPU Designer Nuvia: Focuses on High-Performance Processors

Qualcomm to rival AMD, Apple, and Intel

www.tomshardware.com

​

Years ago, Qualcomm used to design its own CPU cores, but had trouble keeping pace with the standard designs that ARM licenses out.

 

[Colif]

Anandtech didn't make the graphs, though. Those are slides from Qualcomm. No reason not to include them here.

the people who own Anandtech also own Tom's so I am sure they don't mind us using it too much

 

[bit_user]

the people who own Anandtech also own Tom's so I am sure they don't mind us using it too much

True. I try to cite and source from Toms, as much as possible. If I have trouble finding it (this site's search function is pretty bad), or something just wasn't covered here, Anandtech is my next stop.

Sadly, Anandtech isn't what it used to be.
: (

BTW, they mentioned one their excellent former phone SoC reviewer was again involved in the Qualcomm benchmarks, but from the other side, this time. He's working for Qualcomm, now!

 

[JamesJones44]

TSMC N4 is supposed to have 6% higher density and approximately 4% better performance (at the same power), than N5.

That's N4 vs N5. N5P vs N5 claims 7% higher performance.

TSMC’s 3nm Conundrum, Does It Even Make Sense? – N3 & N3E Process Technology & Cost Detailed

Shrinking finally costs more, Moore's Law is now dead in economic terms

www.semianalysis.com

Based on those numbers the two nodes look almost identical in claimed performance.