Hacker News Reader

Comment by Pet_Ant | Hacker News Reader

It is exciting to have more competition in the CPU space. While it is currently possible to build a non-x86-64 desktop class computer, all the current options are really expensive (looking at you Talos). It would be nice to have more architectures at this level eventually. Even if only to see what the real costs of the 8086-legacy support is to modern performance (if any).

View on HN

It'll be interesting to see the Snapdragon X Elite next year. It won't take the performance crown but it should finally give reasonably priced ARM outside of the Apple ecosystem something in the high performance class.

As for the real costs the upper bound seems to be no more than ~30% on average for the software layers to do it so probably less than that if the hardware is built from the ground up to run x86 code only. My guess is the real difference is <10%. Still appreciable, but not something that really drives the gaps.

Considering how most of the software we run is open source, I don't think legacy support is that important. I think lots of organizations would be happy to switch for 10% improvement in performance or power.

Open source could really facilitate a golden age in CPU design where backwards compatibility really doesn't matter.

Not all open source code is trivial to port to a new architecture.

> Considering how most of the software we run is open source, I don't think legacy support is that important

At the bottom of the open software pyramid are drivers, and they are often closed-source. Qualcomm is a major contributor to a lot of Android devices having short support periods. Even though Android is open source, Qualcomm's drivers are not, and Qualcomm would rather have consumers buy devices with the latest snapdragon than keep using an upgraded device from 5 years ago.

Device drivers aren't really relevant to servers and cloud instances which I had in mind where power savings and TCO rule.

But yes, for consumer electronics it's a whole other story.

What drivers are you talking about? Most of the divers required to run a Snapdragon SoC are in the upstream kernel.

For the typical desktop user, windows drivers. This includes things like anti cheat in games. For the typical HN desktop user it should be good though.

Previous Qualcomm laptops have been very expensive unfortunately.

8cx Gen 2 are ~$500 new. That's not exactly a bargain Chromebook but it's also not particularly expensive. Of course, we'll have to see what they do for pricing on this upcoming chip. Given the performance levels even just a desktop x86 CPU that competes may sell for that $500, not counting the entire rest of the computer.

Snapdragon X Elite is faster than all but the M3 Max.

From what I've seen it actually beats the M3 series in single-core performance.

That Nuvia acquisition has really paid off for Qualcomm.

> Snapdragon X Elite is faster than all but the M3 Max. > From what I've seen it actually beats the M3 series in single-core performance.

Citations? Because from what I have found [1], Qualcomm only claims it’s 7% faster than a non-Pro, non-Max M2 in single-core. It’s 50% faster than the M2 in multi-core but that’s using 50% more cores

And all of this on a system with a TDP almost 4 times that of the M2 system, meaning it’s consuming a lot more power to get essentially the same performance.

All of this on a chip that’s coming out in 2024 at the earliest, when Apple will be introducing the M4 series — two generations down the road from what the Elite X is performing at.

—

[1] https://www.tomshardware.com/news/snapdragon-x-elite-outperf...

Looks like the numbers I was looking at were for the high-power configuration, and differed a little between sources, nontheless the low-power configuration, which has a significantly lower TDP (23W) than the M3 Pro (~30W), while having the same core count, is still very close to M3 Pro performance even if M3 Pro is a bit higher.

https://cdn.mos.cms.futurecdn.net/zwxHUu7zdH4y5wgbzpwnD9-970..., from https://www.windowscentral.com/hardware/laptops/results-are-...

Your graph says the single core performance of the high tdp option is lower than any of the M3 models and the multi core performance is lower than both the pro and max. It is close though, but it's also comparing CPUs out now to CPUs to be released in the future.

Not to mention, Qualcomm has a pretty bad history of meeting its promised dates.

In 2021, it claimed to be on track to shipping M-series-level chips in 2023. [1]

In 2022, it claimed to be on track to shipping M-series-level chips in 2023. [2]

Now, in 2023, it claims to be on track to shipping M-series-level chips in 2024. At least that date is still in the future, but I think it should be looked at skeptically in light of their past unfulfilled promises.

—-

[1] https://www.theverge.com/2021/11/16/22785181/qualcomm-next-g...

[2] https://www.pcworld.com/article/1382740/qualcomm-dubs-nuvia-...

Be careful what you wish for. I don't want this kind of competition. You'd have to be crazy to use these given the sheer amount of tech theft and hybrid warfare coming out of China.

Reply on news.ycombinator.com