Tuesday, June 18th 2024
First Reviews are Live and Snapdragon X Elite Doesn't Quite Deliver on Promised Performance
The first reviews of a notebook with Qualcomm's Snapdragon X Elite SoC have appeared today, and it looks like the promised performance isn't quite there. And yes, all the reviews that went live today are all based on Asus' Vivobook S 15 OLED, so it might be a bit too early to state that Qualcomm isn't delivering on its claimed performance, as other manufacturers might deliver better performance. Let's start with the battery life. The Vivobook S 15 OLED comes with a 70 Wh battery pack which enables it to deliver better battery life than many AMD or Intel notebooks, but Apple's MacBook Air 15 M3 delivers on average a 40 percent better battery life, with a smaller 66.5 Wh battery pack. Browsing the web or watching movies aren't really too taxing for the Snapdragon X Elite, but under heavier loads the battery life drops off a cliff.
When it comes to application performance, the Snapdragon X Elite offers good multicore performance in benchmarks like Cinebench 2024 and PCMark 10, but it falls way behind in most other tests, ranging from video encoding to file extraction and document conversion, with Intel Core Ultra 7 155H based notebooks often pulling ahead by 50 percent or more. Despite being equipped with LPDDR5X-8448 memory, the Snapdragon X Elite falls behind in both the memory copy and write tests in AIDA64 compared to the Intel powered laptops. However, it's not all doom and gloom, as the Qualcomm chip delivers an impressive memory latency of a mere 8.1 ns, compared to 100+ for the Intel based laptops. It also outclasses the Intel laptops when it comes to memory read performance.
Asus went with a fairly basic Micron 2400 SSD which is a DRAM-less Phison based drive and this might be part of the reason for some of the less flattering results in some tests. However, this shouldn't affect the gaming tests and this is another area where the Snapdragon X Elite doesn't deliver, and most games are unplayable at 1080p resolution. Many games don't run on the Qualcomm chip for obvious reasons, but many that do, suffer from texture and graphics glitches at times. Most games don't even manage 30 FPS at reduced graphics settings, let alone 60 FPS, but then again, this is hardly expected from an integrated GPU. Considering that the Vivobook S 15 OLED comes in at US$1300 with 16 GB of RAM and 1 TB SSD, you would expect it to deliver in terms of performance, but it seems like Qualcomm and Microsoft have a lot of work to do to optimize the platform as a whole.
Sources:
Windows Central, Notebook Check (in German)
When it comes to application performance, the Snapdragon X Elite offers good multicore performance in benchmarks like Cinebench 2024 and PCMark 10, but it falls way behind in most other tests, ranging from video encoding to file extraction and document conversion, with Intel Core Ultra 7 155H based notebooks often pulling ahead by 50 percent or more. Despite being equipped with LPDDR5X-8448 memory, the Snapdragon X Elite falls behind in both the memory copy and write tests in AIDA64 compared to the Intel powered laptops. However, it's not all doom and gloom, as the Qualcomm chip delivers an impressive memory latency of a mere 8.1 ns, compared to 100+ for the Intel based laptops. It also outclasses the Intel laptops when it comes to memory read performance.
124 Comments on First Reviews are Live and Snapdragon X Elite Doesn't Quite Deliver on Promised Performance
It's always good to have options.For now you can run WSL2 with aarch64 ubuntu and you can get native ARM pgsql/docker/etc. A proper linux ARM laptop would definitely get me to buy one, though, but I have some faith in WoA and application support has been expanding nicely.
They just need to add some more RAM at the entry level prices.This is a good read about it.
chipsandcheese.com/2024/03/27/why-x86-doesnt-need-to-die/
I can confirm the battery life is far superior to my x86 notebooks. I am test driving this to see if it's a viable competitor to the 13th Gen Intels I have been buying for work. Overall, my impressions are that the chipset is very impressive, comparable to 13th-14th Gen Intel while sipping power; however, the software needs another 6-12 months. Would I buy this for my workforce moving forward? I need another couple of weeks to decide, but from a purely general business usage perspective, I am genuinely impressed.
My distaste for these toy CPUs being sold at desktop CPU prices aside, I'm sure this will part fools from their money and thus cut into Apple's marketshare. And anything that hurts Apple is good for consumers.
I don't see any advantage in buying qualcomm's buggy product.
I have one question though: will the development of Prism help those with older 8cx devices?
Oh, and Pentium / i686 was dual-issue / dual-pipelined. Intel didn't experiment with microcodes until Pentium4 IIRC, which is where people started talking bullshit about "RISC Core" even though the microcodes are "Perform an entire AES encryption step", which is hardly "RISC". By the time we're talking about 486 and Pentium btw, we're at "16-bit original FPU modified x87, 32-bit extended MMX, SSE". The later stuff hasn't happened yet, but it seems very difficult to call this a "Reduced Instruction Set Computer", especially as Pentium still supports the "loop" assembly instruction, push, pop, divide, and other "complex" instructions that perform multiple tasks in one code.
Unless you're talking load/store architectures btw, which is all the microcode system converts Intel Assembly into. Load/store is just one component of RISC, nothing else Intel / AMD do with x86_64 is anything close to the RISC they talked about in the 80s or 90s.
-----------
Oh, btw. both ARM and RISC-V are microcode engines today. Because their instructions are so complex they need multiple microcodes to implement them.
Everyone's Divide / Modulo instruction is microcoded. Because that's what you do, you destroy RISC-mindset because divide is so common that it makes sense to accelerate it at the machine level. But it also doesn't make sense to implement divide in its entirety all in hardware, because division is a very complex set of operations. So you compromise with microcode.
---------
My point is that CISC vs RISC has been stupid for decades. The entire debate is just a bunch of people misunderstanding microprocessor implementations and circlejerking over it.
PowerPC, ARM, and the legion of lost 80s/90s processors (SPARC, Alpha, MIPS, PA-RISC) all claimed to be RISC. Furthermore, not a single company claims their ISA to be CISC. CISC is what "other" companies call Intel's design, almost a derogatory term that's pushed upon the competitor to make these other designs feel better about themselves.
If I were to push a company responsible for the RISC-mania, it'd be IBM and its POWER architecture. IBM did many CPU designs and seemed to push RISC as a marketing term the most (even as RISC was used all over the place, IBM was probably the most powerful and widespread mouthpiece to the pseudo-concept).
That being said: deep discussions about pipelines, processor efficiency, throughput etc. etc. allowed for these companies to leapfrog and advance CPUs. Alas, the "CISC" x86 instruction set by Intel (and later AMD) turned out to also take all those innovations anyway, and become the fastest processor of the 2000s, 2010s.
-----------
RISC-V continues the tradition of claiming advanced ISA design, much like its SPARC / ALPHA / MIPS bretherin before it. After all, every new CPU design needs to say why its better than the "CISC" machine over there, without necessarily using the trademarked term (Intel) in their marketing.
-------
I think with all the processor / ISA wars of the last decades, I've come to the conclusion that it doesn't matter. AMD came out with the dual ARM+x86 "Zen" design back in 2016, proving that all of these ISAs can convert between each other on a modern core anyway. And that's when I began to realize how similar ARM vs x86 was at the instruction level.
ARM stands for "Advanced RISC Machine" by the way, and was one of the other major marketers of the RISC term. (ARM wasn't very big in the 90s, but is a big deal now). With AMD making ARM + x86 compatible processor (even if it was just for its own internal tests), it proves how bullshit this whole discussion was. Decoders are not the critical element of modern processors, and they can be swapped out without much hassle. (At least, not a hassle for these multi-$$Billion megacorps).
- Jetbrains IDEs (IntelliJ, Rider) work perfectly and have ARM64 versions, as so C# and Java developer tools.
- Many databases can be ran using WSL2 natively on aarch64 ubuntu, and honestly it's also how it's done usually on x86 windows (ex. redis, pgsql usually tell you its better to run them on WSL2 as they're linux-first, obviously)
- Web browsers are fine (Edge, Firefox, Chrome have all WoA versions)
- foobar2000, 7-zip and misc utilities all had ARM downloads, I was also able to find a qbittorrent buildbot.
- Most benchmarking software I have installed have an ARM64 version now
- PowerToys, of course
- HWInfo64 has been contacted by qualcomm and are on ongoing efforts to port it, which also means probably PerfMon will be available
- Syncthing has a WoA version, aswell.
- Office suite also has an ARM version.
With that I wouldn't be missing any programs I use, including developer and work tools.
Outliers:
- Discord (which you can run in your browser)
- Telegram (with plans)