INCONSISTENT Benchmarks, WHAT'S going on?!
ฝัง
- เผยแพร่เมื่อ 13 มิ.ย. 2024
- Snapdragon X Elite benchmarks on the Samsung Galaxy Book 4 Edge are... inconsistent.
🛒 Gear Links 🛒
* 🍏💥 New MacBook Air M1 Deal: amzn.to/3S59ID8
* 💻🔄 Refurb MacBook Air M1 Deal: amzn.to/45K1Gmk
* 🎧⚡ Great 40Gbps T4 enclosure: amzn.to/3JNwBGW
* 🛠️🚀 My nvme ssd: amzn.to/3YLEySo
* 📦🎮 My gear: www.amazon.com/shop/alexziskind
🎥 Related Videos 🎥
* 🤖 The end of Apple Silicon’s reign - • The end of Apple Silic...
* 🌗 Volterra vs M1 Mac Mini | Visual Studio - • Volterra vs M1 Mac Min...
* 👨💻 15" MacBook Air | developer's dream - • 15" MacBook Air | deve...
* 🤖 INSANE Machine Learning on Neural Engine - • INSANE Machine Learnin...
* 👨💻 Fireship's CPU in 100 Seconds - • How a CPU Works in 100...
* 🛠️ Developer productivity Playlist - • Developer Productivity
Reddit post: / samsung_galaxy_book_ed...
- - - - - - - - -
❤️ SUBSCRIBE TO MY TH-cam CHANNEL 📺
Click here to subscribe: / @azisk
Join this channel to get access to perks:
/ @azisk
- - - - - - - - -
📱LET'S CONNECT ON SOCIAL MEDIA
ALEX ON TWITTER: / digitalix
- - - - - - - - -
#Windows #SnapdragonXElite #ARMProcessors - วิทยาศาสตร์และเทคโนโลยี
![[Watch Party] RoV APL 2024 Bacon VS TDT ( AlmonP Wanoiz )](http://i.ytimg.com/vi/LvZvKzu3VSk/mqdefault.jpg)
Never trust a reddit user who likely just put windows in power saving mode before running geekbench.
as long as it gets views, everything is good to make a fuss over a simple problem, even if you have to treat people as idiots
nah, power saving mode must not cut 40% of single core performance
Yeah. A reddit user wouldn't be as thorough as GN or some reputable reviewer. There seems to be an anti-Windows bias on that platform as well. You're either pulled to the ArchLinux cult or the Apple crap. Pick wisely.
@@anglchnn yeah, wait until it caps the max frequency at 5%
Definitely this. Reddit users aren't exactly known for being the brightest people
This is so inconsistent. Btw good investigation, Please Stay determined!
Good investigation my ass, if only he'd looked at the same reddit thread he would know that its most likely uncooked driver. The CPU waa not running at Qualcomm spec frequency. Then again, maybe he did see that but choose not to bring it up.
@@lasue7244 Computers also turn in much better benchmark scores if you uninstall Copilot, OneDrive, the Xbox related apps, and other useless background apps included with Windows. With Copilot installed I was seeing occasional startup times for Excel of up to 20 seconds, which went back to sub-second after uninstalling it.
I read the entire thread and determined that we'll see the truth for ourselves next week. I wanted to raise awareness of the different specs available that are not necessarily advertised as different.
@@AZiskyou could've just added a single line that this could be an early driver issue. This info was there in the same TomsHardware AND the reddit thread where you got rest of the details for this video. Doesn't makes sense to leave that specific bit.
I don't know if it's drivers. It could very well be thermally throttling or maybe the chips really are shitty. We just don't know yet, but you still left stuff that was already said on reddit.
Bro he didn't even used any laptop of these laptop to do benchmarks
I used to do competitive analysis & strategy at Qualcomm and I owned all the benchmarks. They are generally legit test scores. From what you showed, the base frequency seems to be the actual issue. Either it's by binning or that device is doing something else. It's hard to determine from consumer as they don't have ideal test condition. If board and chips are the same, it's likely the user is doing something like installing stuff or updates or whatever that caused the thermal limitation. Just lots of variables here. But just wanted to point out that they generally don't have the practice to pay benchmark vendors to mislead users (unlike some other suppliers).
also I left long ago so things might changed...
Appreciate your insight on this. Thank you!
Most consumer don't test for any scores. I suspect the low frequency ones are low power version. It's like AMD's 5700U.
Windows updates in the background? ... This is a issue that most people do not realize, you need to wait on fresh systems a while after booting (sometimes ages!) for all the updates to be installed. And that easily explains the score issues. But yea, clickbait titles from so many outlets.
Can't be just the frequencies. The frequency drops by 20% and the single core score drops by more than 40% ? That doesn't add up. It should be more like 10-15% performance difference.
Damn, if the base frequency is the issue and you need "ideal conditions" to not get a massive spike at ~1800 1T GB6 points (notice how it is bimodal), then there's something funky here.
Over 100+ scores like that in GB6, out of some ~150 results. It doesn't mean QC _paid_ GB6 (lol), but it can just mean the chip is not reliably fast in a typical environment.
Let's wait for reviews to be sure, but I'd be much more likely to guess a big ole bug on a launch device than binning. :(
The fix is already out. The reddit got it as well. The single core score is 2700-2800 and the multi core score is approx. 14000.
Recall got recalled, shipping machines with faults requiring fixes, if Apple had a rollout this buggy the internet would be having a field day
Plot twist: underperforming chips are actually Exynos 🤯
plausible
That will be crazy if Exynos first laptop chip entry is already on that scale of production
Lol
Samsung: just a prank broski
They do the same with their smartphones so it won't be surprising 😂😂
Mate, the industry needs more objective people like you. Seriously, you aren't an Apple fan boy, etc. You are saying and compiling the facts as it is, without agenda without bias. Thanks for that.
Except that he only answered the question in the title in the last 35 seconds, wasting the first five minutes giving use examples of inconsistencies and stringing us along. He's biased toward getting views for mostly insubstantial content on TH-cam.
If anything he's being too generous, these companies KNOW the first thing users are going to do with a brand new platform is run a bunch of bench tests
The reviewers who run these tests on a macbook in its default mode simply hit run and then report the first value. It's usually fairly consistent
There's no excuse for releasing underbaked products that run inconsistently on bench tests
Nice digging! Also, wasn't these chips capable of running in different TDP configurations, wonder if that is tied to that base frequency difference you saw. Comparisons are gonna be tricky.They are not making it easy for themselves 😄
That also caught my attention. Some do not even indicate what type of processor they are selling, whether 80 or 84, and now we also have to look at the frequency?
Also remember that in addition to the different Snapdragon X Elite Chips with different frequencies, these also suck down more power for the higher end Chips. I'm not sure that it's adequate any longer to only report GeekBench Scores without also reporting Power Draw ...
That is a good point.. but in that case everybody will be shocked if realize that even Nvidia use 600W+ in benchmark to show you the performance of the average 450W card...
For some reason people think it's ok to put the device into battery saver mode to compare to a macbook default mode for battery life, then put it into performance mode to compare to a macbook's default mode for performance tests
If the product was so good they wouldn't have to do this, and they wouldn't be making every comparison to a base model M3 chip in a passively cooled Air chassis just so they can punch down to make it look good
TH-cam is littered with sponsored reviews and not all reviewers are forthcoming with this information
Happy to see, you did some investigation on your own.
This difference likely stems from the fact that each Snapdragon X core cluster has two operating modes: energy-efficient and high-performance. It's unclear how switching between these modes functions for each cluster. There might have been a bug with the switching process. Therefore, I suspect benchmark scores around 1750 are achieved when the core is in energy-efficient mode. This similar how Apple Silicon work in power-saving mode.
Really good and competent analysis, congratulations.
You've gotten yourself a new subscriber. Keep up the great work 👍
Another great video, Alex 🎉
Thanks again!
Thank you for this great video so that we can look closer to what we should look for when purchase a Snapdragon X-Elite chip PC.
Yoooo..
That information is GOLD, dude! Thanks!
I got hyped to buy one, but it looks like I'll wait on to see what Samsung will do in Brazil.
I wonder if some of these scores are coming from test units or engineering samples. That would explain the different mobo and the different frequency. The other thing that might explain this is some combination of thermal throttling (if the unit has been in heavy use before running the benchmark) or if it’s under battery power in a power saving mode.
Bottom line is that we can’t trust any of these benchmark results until they’re released to the public and media outlets can run their own independent tests along with publishing their methodology. But it looks like a best case scenario is that the chips are capable performers and might even live up to the hype.
I suspect a bios bug/error causing it to run in a lower powerstate, or simply some kind of user error. Good video, hopefully the cause can be found so people are not missing out on performance they should have.
Thank you Alex. Very accurate.
Is the score discrepancy for the X1E80 anything to do with battery saving/performance slider? Most typically they're set to balanced by default. Setting them to Max Performance could explain the difference you're seeing in these results from the same chip.
Love your content, but please, fix that TH-cam plate in the background 😂😂😂 I can't pay attention to anything else 😂😂
one day:) thanks for watching, even though it’s so hard for you;)
OCD hits hards sometimes 😂
If all what you see is that TH-cam plate behind him, then twist or align your head to the appropriate position of agreement to the TH-cam Plate 🤣🤣
🤪
@@AZisk Hi @AZisk I am about to start a software development career and I have been thinking the best laptop to buy until I came across your TH-cam videos, right now I am thinking of buying MacBook Pro M3 and M3 Max. I have the budget for the laptop but I am still confused and I don’t know the one to get Or should I go for windows? Please I need your help🙏 🙏 I confused
Cool video. This new X Elite is exciting and scores looking promising!
Good unbiased and informative coverage
4:45 I believe they have overclocked or depending on the brands the chips are undervolted for better cooling or something else
There are 4 X Elite SKUs, listed is on Qualcomm website too. First 3 are pretty similar but the last one X1E-78-100 has NO BOOST CLOCKS and is pretty much X Plus but with 2 extra cores.For X1E-80-10 0 vs X1E-84-100 the 84 has 400 MHz more base clock and 200 MHz more 1-2-core boost clock :) And allegedly the SoC can use a very wide range of power so Windows power settings and each model TDP range can cause the results to vary even more.
Because to prepare for the launch of the chip they will need to add some benchmark score to geekbench with the device in the best condition to have high score. But because geekbench is also open to the test of other users so we see mix results here.
This makes no sense. The CPU was cap at 2.52 GHz, so it was not working at its full potential. The article clearly said that this is due to driver issues. These products are not even released yet.
I don't understand why this is not mentioned.
becauuse fud sells newspapers and being honest rarely does, unless your honest is more sensation than the fud, of course. Just make sure it's a truth WANT are willing to hear.
Doesn't the SoC deals with its own clock speed, why does the driver have to do anything with it? The GPU shouldn't matter in these benchmarks so drivers shouldn't be an issue.
I believe you mean the OS scheduler by "driver" or another system component ment to deal with tasks in the user space. So unless those machine use diffrent version of windows 11 preview, which those versions can have diffrent schedular optimizations, it shouldn't matter. I Don't think you can install a CPU driver.
maybe it has a native overclocking software that overrides the cores speeds but turns out it was underclocking it from the start
Well, you forgot that the Windows is a multitasking device, if 8 task using 8 cores, than the system will automatically limit the frequency less like in a separated environment where you are only benchmarking.. will you close all your other apps just to get more performance in the current app, if that is the reason why the performance is limited..?
Same
Great video
I would have thought that it is battery (saver) vs AC power but the frequencies on the order page seem to match the lower GeekBench scores. So maybe the higher scores are from overclocking? We'll know soon enough.
probably the are making the test on a economy mode or so and the other one in performance mode. IDK, but it is very difficult they would have 2 different motherboards. It would make maintenance of those notebooks a nightmare. A experienced company like Samsung would not make a silly mistake like that... or would they?
Excellent investigation thank you. This is why I am holding judgment until I see reviews for all Snapdragon laptops from independent reviewers.
There may very well be firmware or Hardware Hardware differences between laptop manufacturers that could affect the base speed
Tom's Hardware has been in a downfall for 10 years.
I remember at one point there were admin banning people in comments even when saying small corrections to articles and a lot of articles beeing just ads.
They will never learn... They have been reliable in the past... But not anymore.
A big part of the problem is that they were bought out by Future Plc, just like many other sites. Old staff probably left, leaving freelancers to come and fill the vacancies.
@@Slavolko I didn't know... A great shame...
Maybe plugged vs unplugged?
good theory. although expectations are that it shouldn’t matter.
Apple proved you can have full performance unplugged or plugged in. Why would snapdragon be different?
@@FenrirKi couldn’t have said it better
Part of those benchmarks were plugged in vs unplugged, basically no difference. I'm curious if in windows they set it to "performance" mode
I’m more concerned by that single core score. Multi-core can be explained by some battery saving feature. Losing 30-40% single core is completely unacceptable. Even Intel laptops don’t typically lose single core performance while their multi core tanks.
There are plenty of high and low NT960XMA (16") and NP940XMA (14") scores on Geekbench, and the reported base frequency doesn't seem to be the culprit. For instance, one of the most recent scores is from a X1E80100 16" at 4.01 GHz with a score of 1726/10330.
It's odd that Samsung shipped like this, but I doubt this is hardware-based and expect it'll get resolved quickly with a firmware update.
Suggestions for the next YT video- Running LLMs on Windows ARM and comparison with M3.
Please also consider showing token per sec/NPU usage and the whole array of metrics with this new PC💻...
Dear Alex, I would be interested what you think about the AI features of Apple Intelligence vs Copilot+PCs. Especially for productivity (coding, editing and writing documents, etc.). Thanks in advance
Maybe some fw updates can fix things up
Good point to notice
I really want to know what the power draw of these chips are.
I have a feeling it's just background tasks or some of these devices being on an older firmware version
I have nothing against product segmentation, but that kind of branding is only going to cause confusion in the long run
I waited for this review. Please VMs, Docker, games, video edit. Thank you! 🇨🇦
It has to be on-battery power saving mode. Maybe that hasn't been changed from x64 Windows when ported to ARM, and those who are getting good scores are either plugging their laptop or manually overriding power plan.
Your videos are so good man. Definitely going to click Join to your channel.
How will Qualcomm's product managers spin this one?
Could the difference in performance be different laptop sizes? It seems the X1E-80-100 is available in both 14 and 16 inch and perhaps the 14 inch is thermal throttling or something.
Not even the iPhone throttling while the benchmark measure it's single core performance, so a much bigger sized body with more efficient passive cooling can't perform worst like a tiny phone (just because of the size, the surface where the colder air can reach the device is bigger, so cooling it more efficiently, and in Apple case the iPhone A chip and the Macbook M chip have the same cores, just in different amount, even if Qualcomm not use the same cores like they have in the phone, there won't be that big difference between 2 ARM design, can handle the sigle score test without throttling)
Is it just battery vs plugged-in to power?
Maybe board firmware is conservative in this first version to keep chip safe . But, we will need definitely your review to find the good one 😅
THANK you so much for doing you r own research and not following a single tweet. I talked to the guy who posted the original tweet, even telling him to WAIT becausr the laptops JUST came out, and there are MULTIPLE variants.,
Interesting! What kind of motherboard is built on Microsoft Surface laptop with x Elite 12 core? Isn't further specified this detail on the site.
I am curious how much power will that NPU draw from the battery? If I don't do ML will it be turned off or will continue to run in the background. If it think of it now, this could put a lot of pressuer on the battery while not plugged. And overall will shorten the battery life.
Turn it off? 😂 what are you saying my friend always on always listening.
@@albertsitoe7340:))))))
if the npu is anything like apple’s, it will sip power and only when used. but idk how this one will work for sure
@AZisk Time will tell. Need to dig deeper in how these "ai" laptops ... phones work. If the npu is constantly running LLM optimizations and fine tuning on your device, this can accelerate the ageing of your device 😅. Be sure when you decide to review these machines, I will be more than interested to hear your opinion 😁
Mh, I ordered the Asus S15 with a 78100, seeing the result of the 80100 so low, I'm close to cancel....
I didn't think the result would be so different...
The problem is that here, apart from the Surface (way too expensive) and the Yoga (same chip), none of them have 32GB of ram... and 16 is really too small.
Good in depth research Alex! Very strange that the Samsung is selling the same laptop with different moderboards when there is no option to choose from on their own online store… 🤔 So basically their customers are paying for a lucky bag 🤣
4:40 The power management plan is also different which can greatly control the boosting of the chip speed during benchmark.
Its just balanced in Spanish i think
I have the feeling ARM PCs will be a short lived low volume experience, 8-10years. There are new techniques in computing as we reach physical limits meaning a change will come but that change will also need to emulate x86 and probably won't bother with ARM.
Maybe he was using powersaving mode, based on the benchmarks we can assume that
The really interesting comparisons are between the snapdragon’s and the new M4 chip from Apple which is just on the new iPad Pro 13.
Maybe some of battery mode? doing benchmarks while having apps in the background.
How about power saving settings?
Are there other laptops to benchmark? Samsung laptops run hot and do throttle a lot.
Sorry slightly of topic yet relevant to the video, DE Havilland Hornet or Westland Whirlwind?
Thats insane, if some chips come at a higher frequency than others even on the same architecture, whats the actual score and why are some at an advantage and others at a disadvantage
The biggest problem is the X1E-78. It does not have an dual coore boost which means that it will probably underperform in single core. In the geekbench browser for the X1E-78 The best you find is 2450. Meanwhile the X1E-84 and X1E-80 you can find examples of it performing like qualcomm initially announced with the X1E-84 aeounf 2950 and the X1E-80 at around 2750. The Problem heare is that the X1E-78 is in most laptops that are available.
Agreed! Something is up with the X1E-78. There's a Geekbench result for the base model X1P-64. Why does a 10-core chip have the same multi-core performance as a 12-core?
My take is unless you really need OLED or more ram, just buy the Snapdragon X Plus base model.
I'm just glad these devices motivated Google to release Chrome for Windows 11 ARM. I'm revisiting my 8CX gen2 HP now that I don't have to take the emulation hit to run Chrome. I'm not sure if I will purchase a Snapdragon Elite PC yet. I'll probably wait for the reviews on the Asus PZ13 before I finally decide.
I can guarante that these scores are odd ones out using an extreme power saving mode.
It's the Samsung Android issue again, but now with windows laptops (same model, different chip/board, different performance), I wonder if it is based on the country (exynos vs snapdragon), or if it's just different suppliers not delivering consistently and really making it random
I believe the np940 are sell on Brazil because of the "equilibrado" power plan, that is "balanced" and maybe samsung has make this to deliver a cheap product here in Brazil
The tilted play button is tripping my OCD lol
Thanks for sharing my experience with this Laptop. Device it's great so far but the scores are not what we hope and paid for.
It's also on the news that the recall won't be rolled out due to security reasons. At least for now. It's a bad idea to implement it full stop in my view. Let's say I opt out of it, I can't control if a business is using it. So I can't trust my data with any outside source. Companies, doctors, dentists, council, government. Any are vulnerable points.
is there a binning problem?
1:11, this is just a first version of snapdragon nuvia windos chip, cant wait for second gen.
Thy are trying to put the windows PC copilo Plus down. I saw some of the benchmark with people that actually do these things and they're pretty good and they're faster than the M3. Not in the single cord but on the multi-core for sure. But very close very fast
Maybe they are planning to offer them in different performance versions, it would be wise. Also, they can be unethical and compare prices using the low-cost machines and performance using the expensive ones. Also could be immature software.
Pretty sure the Twitter guy also Said it was because of them lowering the frequency of the chip to still have good battery life.
At the end of the day, in my opinion, it makes no sense for ARM to make 2 versions of the higher end X ELITE processer. Should have just made the higher wattage one to prevent all of this confusion.
it's driver issue, wait after 18 June after more update
ASUS Vivobook S 15 seems to have scores all over. 16 GB / 32 GB RAM seems not to explain. Is it related to SSD (512 GB vs 1 TB), just like Apple case earlier (256 GB vs 512 GB)?
I'm still waiting for Qualcomm to provide instructions on how to order the Dev Kit.
These manufacturers need to make their products less confusing. It's too easy to think you've got one thing when you have something else.
tbh i dont understand the fuzz about those chips
if you consider power draw they are not nearly as efficient as newer m series
W to this tech investigator! do you think intel core ultra 7 155H is good enough for gaming laptop? i found msi laptop with rtx 4050 with 144hz refresh rate and with intel core ultra 7 155H which i wanted balance of performance and decent battery life in window laptops. Or should i go with arm processors cuz 155H is still x86
If you want a gaming laptop you should definitely avoid arm processors for now. I think the ultra 7 155h will be fine paired with the rtx 4050.
Geekbench is an awful benchmark in my opinion.
good job sir, that's why I don't trust random reviewers, but I trust this channel that speaks facts!
The scores lower than 2000 were due to being run on low-power mode, which may be the default when received. Other scores suggest 2400-2700 single core. Which is solid.
But that shouldn't matter for single core. And it also should be reflected in the frequency, which is only 3.4 vs 4.0, clearly not 40% difference as the scores suggests.
Maybe just depends what power was used, if it is a battery saver it could influence the clock speed, similar to Apple Low power mode
It shouldn't matter for single core
The clock speed limit more likely because it's a real world test, and not an isolated benchmark.. i mean if you have 8 opened app, than the system will use 8 cores to run them, and 8 cores in use have different max frequency like just 1 core in use...
@winnetou17 it should make a difference for everything
maybe it's a driver version that lead to different results for same SoC
The performance variance may be down to form factor. The Snapdragon X Elite 84 sku is only available in the Samsung Galaxy Book4 Edge 16-inch model, whereas the 80 sku is available in both 14-inch and 16-inch model. The other thing that would put me off from purchasing this co-pilot+ laptop is the 512GB and 1TB storage is eUFS. I would want Fast NVMe in a laptop of this price.
anyone knows if the silm 7x is any good? Close to the m2 pro ?
There's something seriously wrong. Looking at only benchmarks for the Vivobook S 15 with the S5507QAD motherboard reveals that units with 2350 to 2450 single-core scores have multi-core scores in the 7600 to 9800 range, while units with 1400 to 1800 single-core scores typically have 11,000+ multi-core scores.
Based on what I'm reading, the performance issues appear to be related to thermal throttling of the CPU. It makes me wonder what the ambient room temperatures are for the laptops being tested.
Given the architecture of these things you'd hope they wouldn't be getting all that hot in the first place unless under sustained heavy loads. Most benchmarks are pretty quick. If nothing else it shows how good or bad the cooling solutions are for the machines they're in.
Damn so what are we going to do when buying x elite laptop?
Check the geekbench score for that particular laptop model instead of processor model?
I guess this is going to cause significant headache for consumers since they will expect same performance from all laptops with exactly same processor model !! This was not the case with apple, they have single processor model with single laptop model. Hence performance was kinda deterministic.
Also, what exactly is going on here? What exactly is determining different frequencies of the same processor model across different laptops? How exactly geekbench measures cpu frequency? By some rumtime heuristics or by some OS API call? Btw I guess this must also be the case with old intel ones, right?
One more thing, i guess the ones with higher scores will have poorer battery life?
Even though Samsung might use different motherboards I highly doubt that they are using different cooling solutions so all chips should run at a similar frequency unless there is some throttling going on due to too much stress during benchmarking, gaming whatever.
It's also hard for me to believe that the Samsung quality team never encountered this enormous difference in performance between the two motherboard models.
Probably best to wait for the 2nd version of these CPU's.
I believe results are plugged (fast) vs unplugged (not so fast).
I'm interested to see how they perform with Virtual Box and Docker
am only waiting to see how these snapdragon chips handle gamers, i'll judge from there
at leasts on battery and plugged in are close
The Samsung Book 4 has power draw issue that can be fixed in the BIOS with a firmware update.
This is very odd. I don't think is due to clock differences because when Qualcomm presented X Elite last year it was said the (until then 4.3GHz, now 4.2GHz) boost was only regarding 2 cores. An increase of 800MHz of 2 cores running already at 3.4GHz base wouldn't help an ~1800 points SC Geekbench 6 score to achieve 2900 points much less the MC score do so in a much higher level. The power consumption also doesn't seem to be the culprit because during the same October 2023 presentation, Qualcomm showed two different X Elite models, an A and a B ones, the first with a 23W TDP and the second with a 80W, but the MC score increased just around 10%. I remember that I thought (and saw many people wondering about that) why they made a model with an almost 4x higher TDP just to get a lame better performance. My opinion, here: Maybe they did so thinking that the performance would scale well with the TDP getting higher, but in the end it was awful? Anyway, this discrepancy is a conundrum now and if it is indeed the Mobo the culprit we'll only find out after June 18.
Battery option on save mode bug