Razer Blade 16 Early 2023 RTX 4090 Review: Core i9-13950HX beast with world's first dual-mode mini-LED UHD+ display

[NikoB]

Your trouble, like many other sites that test "gaming" laptops (especially), is that in reality among the buyers of these laptops (as real forums show) players are in the minority. Most buy them now as generic laptops. In general, "gaming" laptops are bought many times less than office/business models, but there is a trend in "gaming" laptops - a universal solution for home or office (and they are already bought there in bulk) as a cheap replacement for powerful overpriced workstations, when the goal is not to carry it around often, but need power, lots of ports, a nice big screen, a keyboard and upgrade options.

I have already written that in reality people who buy "gaming" laptops are interested not for games, but for business, work - they are interested in the noise in the "maximum performance" profile when the load is exclusively on the processor, without a video card. And with a video card, only those who work with the appropriate applications. They, these segments of buyers, take such laptops on purpose, counting on a lower noise level in operation than in ordinary business lines, where the cooling system is obviously weaker at a lower speed of processors and video cards in "silent" profiles, as and "gaming" laptop. Those, the calculation is initially based on the fact that since the laptop is larger and heavier, then its cooling system, when working only with the processor, will be as quiet as possible.

But your tests do not show the noise levels under load only on the processor in the maximum performance profile (and preferably with some kind of average undervolting for the Intel platform). And load cpu cores up to 50% with small load for igpu/dgpu.

For example, I'm sorting through old archives on one HDD, I found an old game from 2010 Modern Warfare 2, and so on my already old Dell G5 5587 with GTX1050 in ultra settings for fhd, practically does not start coolers while playing it, and all this is in the profile of the maximum performance from the PSU - i.e. it is generally silent most of the time, as it is 99% of the time when surfing with 2-3 browsers at the same time and dozens of tabs in each - I can sit with it laptop in heavy surfing for several hours and never hear the of coolers. It's just a pleasure to sit in complete silence and at maximum processor burst performance. As well as when playing videos on YouTube 4k@60fps. Emphasizing - all this is in the profile of "maximum performance", i.e. with maximum PL1/PL2 values, but with undervolting. But the load is quite serious, at least in such a game, right? Is it possible to recognize such a model as successful in terms of noise for an office and a below-average load for a processor and a video card? Absolutely. I have never seen a quieter laptop. But it weighs 2.83kg and is the size of a 17.3" laptop, even though it's a 15.6" model. That is why the new Legion 7 2023 models began to increase in weight - the cooling system simply cannot cope with such a monstrous consumption of 200W+. It's abnormal and wild.

It is very symbolic that right now in 2023, last Friday, Gordon Moore, the author of Moore's "law", died at the sunset of silicon technologies, when they have completely reached an impasse (everyone can already see this clearly, even ordinary inhabitants) and at the moment of the most powerful for decades a magnetic storm on the planet.

RIP Gordon and RIP silicon technologies. It's the end of what's possible with the rise in laptop consumption and everything else.

[RobertJasiek]

If I understand you correctly, you want to keep the noise constant, at say 43 dB, and perform the test to see what kind of results come, is that right?

No. I am aware that achieving a constant, say, 43dB on different devices may be impossible. Rather, I suggest, e.g., 43dB as an "up to" limit. Unless a device is generally too loud with tasks involving the dGPU, every device should have some modes / settings with noise rather close to, but at most, 43dB. A tester might set any suitable, reasonable "medium" modes and settings to achieve such. An enduser might spend days on finding and tuning such but a tester has less time and needs to find some reasonably close compromise.

"Some keep on whirring their fans for no reason, while others stay absolutely silent"

When talking about GPU load, we can ignore such Idle problems of bad devices. However, for GPU, a different problem might occur if the manufacturer has not done its job: sudden, unexpected maximum fan noise at times.

"slightest of load above idle. And in these modes, the max fan noise almost always hits 45 dB and above."

A review should point out such bad driver behaviour.
 
"The FurMark setting that you suggest might work for one particular notebook, but it may not offer the same fan noise in another."

I do not suggest the same fan noise but I suggest approximately the same so that an upper limit, such as 43dB is not exceeded.

"So there wouldn't be any way to standardize the test."

I know that it is not standardised because slight noise variance is tolerated for the test. The test shall not detect whether different devices can reach exactly the same moderate noise level. Instead, the test shall detect a) whether different devices can have a moderate noise level while still having roughly, say, 2/3 performance. Such a test is not a competition of the maximum medium performance, but is a filter detecting all those devices that can have both sufficiently moderate noise and sufficient clearly above average speed.

Current NBC testing detects almost none. Such a test would detect all such devices, and flag those for which such a balance is impossible or too hard to achieve.


"Even if I were to accomplish that, FurMark is not really representative of any practical scenario, for which you want to take up this whole exercise in the first place."

Furmark has, IMO, similar GPU load and CPU partial usage as Blender/GPU, V-Ray, TimeSpy/GPU/High_resolution, Geekbench/CUDA_or_RTX, Go_playing_DNNs etc. Furmark may not be a practical software but is a benchmark standard approximising such practical softwares. Therefore, some of your collegues think Furmark is good for the purpose. You might also test the practical softwares but then you would have to do several such noise and performance tests. Just one Furmark test is done more easily.

"Also, using other software can cause fragmentation too."

Use Furmark to avoid fragmentation of noise tests.

"You might want to check noise levels while doing intensive Photoshop work while someone else would want to know about AutoCAD."

This is becoming unfair. Image, Audio, Video, Music softwares can behave like Load Average but some such rendering tasks or heavy 3D-CAD can be like a) Furmark or b) Load Maximum. Load Average and Load Maximum noise are already measured. What is missing is, say, Furmark.

"The advantage in using synthetic tests"

Such as Furmark.

"It's just how the fan curves are designed."

In their unfortunate defaults favouring high TDPs and RPMs.

"Now, the same test can also make do with just 39 dB noise in theory. But often, the fan curves are conservative and designed to maximize cooling performance, so they often ramp up even on light workloads."

And this is why it is all the more important to also test outside the defaults. Everybody knows that notebooks can be (way) too loud. The intesting aspect is how well they perform at acceptable noise.

[NikoB]

In fact, when buying a new laptop, the owner expects it to have "maximum performance"(profile) with minimum noise. But games and heavy calculations are different and not so often used on laptops in reality, outside of gaming scenarios.

But if a new laptop is not capable of being quiet at maximum impulse performance on all cores - then what is the point of spending money on it? If the owner is forced to immediately lower its speed relative to the peaks advertised in the reviews in the "maximum performance" profile?

My laptop can do this in silent mode, while the burst performance is maximum - which is extremely beneficial for the responsiveness in surfing. And what's the point in a new one if, in order to achieve the same noiselessness in surfing, you need to lower PL1/PL2 by 2-3 times, when the processor turns into a pumpkin and there is no such significant difference with my 5 year old anymore? Why should new laptop buyers pay so much now in 2023? If they don't get a silent laptop at maximum impulse performance (PL2 is always maximum), for surfing?

I do not understand this. This is money down the drain. Especially with the current bullied 2 times from an adequate level, prices.

That is why today only Apple laptops look in the eyes of buyers, more and more, as the only adequate choice in this regard, despite other obvious shortcomings, while maintaining a good level of performance even on battery power.

[RobertJasiek]

Niko, for your office usage, you have made the right decision to keep using your older, often silent notebook! Our speed needs may differ though.

[RB15 User]

Jarrod'sTech got an almost 21k Time Spy graphics score from his Blade 16 RTX 4090. Did you guys do the Blade a bit dirty during GPU/game benchmarking by scaling back the GPU setting in Synapse to Medium? NBC has done this before with previous Blade 15 testing, putting the Blade in the Balanced profile vs. Gaming (as per the old Synapse) for gaming benchmarks.

It's important to get this right because Jarrod's 21k score means it's faster than the Zephyrus Duo 16, Eluktronics Mech-17 GP2 (a bigger laptop) and less than 200 points (<1%) off the Scar 18 (a much bigger laptop) as per your other test results.

[check your test]

please check your test equipment which test response times - in most of your recent reviews - appears to be too high  - please check your last 6-8 reviews in at least 4 of them, regardless of screen type, the response times are above 50ms

Hi there. The problem with miniLEDs is that the constant PWM flickering is so strong that it often overlaps with the response time curves. This makes it a tad bit cumbersome to properly define the curves in the oscilloscope. Even if we try and narrow down to individual peaks, the acquisition is not sufficient for proper quantification.

Can you let me know which other non-miniLED laptops you've come across on the site with 50 ms+ response times? Thank you.

for non miniled - the 'recently'- "Lenovo ThinkPad T14 G3 review" (intel ver) ,  "A Chromebook for MacBook Pro 14 users: HP Dragonfly Pro Chromebook review"   and  "HP Dragonfly Pro laptop review: AMD Ryzen 7 7736U makes a splash"       

All of them are IPS types

[Vaidyanathan]

Jarrod'sTech got an almost 21k Time Spy graphics score from his Blade 16 RTX 4090. Did you guys do the Blade a bit dirty during GPU/game benchmarking by scaling back the GPU setting in Synapse to Medium? NBC has done this before with previous Blade 15 testing, putting the Blade in the Balanced profile vs. Gaming (as per the old Synapse) for gaming benchmarks.

It's important to get this right because Jarrod's 21k score means it's faster than the Zephyrus Duo 16, Eluktronics Mech-17 GP2 (a bigger laptop) and less than 200 points (<1%) off the Scar 18 (a much bigger laptop) as per your other test results.

Hi. Not sure about Jarrod's testing. But I haven't changed the power profiles anytime during the entire review except for battery run times as described. I re-ran the test again (Custom: CPU-Boost/GPU High) just to be sure and got a similar score of 19,881 for the Time Spy graphics test.

[Nedim]

This review hit so much attention, but pixel response times are a big let down 👎🏻

[S.Yu]

I'm disappointed by this model and I think most people would agree. The Blade 16 is supposed to be beefed up Blade 15 that focuses on extracting sustainable performance, because it's a whole lot thicker yet that thickness isn't really put to use anywhere else, like on a mechanical keyboard or on speakers that actually fit this device class. Too many corners are cut for its thickness and price, also I'm still having trouble comprehending the performance differences in light of the theory that all 4070-4090 mobile cards consume no more than 105W.
--------
As a side note, displaying black is not the way to measure light bleed of local dimming panels, you need 1% white tested at different regions to prevent all the backlight from shutting off which is effectively cheating, like fake contrast numbers from projectors with auto iris, it's not the true global contrast.

[NikoB]

As a side note, displaying black is not the way to measure light bleed of local dimming panels, you need 1% white tested at different regions to prevent all the backlight from shutting off which is effectively cheating, like fake contrast numbers from projectors with auto iris, it's not the true global contrast.

You're just writing about the standard ANSI contrast test, where a checkerboard is displayed - black and white squares.

On IPS without multi-zone backlighting, this is essentially the case (although there will be backlight), but on miniLED, the contrast in such a test will immediately drop by 10 times, as one would expect.

The reviewers here don't do an ANSI test, they do an On/Off test, which is beneficial for multi-zone backlighting and for example in projectors where the ANSI contrast is 10-100 times lower than the On/Off contrast.

It should also be taken into account that the authors write too little about obvious halos around contrasting objects, at the borders of illumination zones.

In any case, miniLED panels are a crutch in front of microLED, a complete analog of AMOLED, but, as everyone hopes, without its key drawbacks, such as low resource, flickering and glare screen. But if there is low-frequency flickering there too (or too little resource and not very accurate color reproduction), we, the users, also don't need them, just like the flickering AMOLED.

As a result, the technology that offers the highest contrast, flicker-free, high working resource, color stability and accuracy will win. At the lowest price. So far, the end of this war is far away. But today regular 4k@120-165Hz IPS is the best choice with a contrast ratio from 1500:1+.

[S.Yu]

As a side note, displaying black is not the way to measure light bleed of local dimming panels, you need 1% white tested at different regions to prevent all the backlight from shutting off which is effectively cheating, like fake contrast numbers from projectors with auto iris, it's not the true global contrast.

You're just writing about the standard ANSI contrast test, where a checkerboard is displayed - black and white squares.

On IPS without multi-zone backlighting, this is essentially the case (although there will be backlight), but on miniLED, the contrast in such a test will immediately drop by 10 times, as one would expect.

The reviewers here don't do an ANSI test, they do an On/Off test, which is beneficial for multi-zone backlighting and for example in projectors where the ANSI contrast is 10-100 times lower than the On/Off contrast.

It should also be taken into account that the authors write too little about obvious halos around contrasting objects, at the borders of illumination zones.

ANSI is 50% APL, but yes, it would work, it's just not as clear as a white spot sweeping across the screen exposing backlight bloom(yes also called halos sometimes) but I don't know what the latter is called. On/off, yes, that's what this here is called and it's ineffective for this type of backlight.

[Bogdan]

Thanks for great review (as always). Can you advise how to squeeze tested 503 mins H.264 / 345.5 mins WiFi v1.3 out of its 85W battery? In your review you got Idle Minimum / 27.5W and Idle Average / 34.4W. So even in idle mode, it is hard to get more than 180 mins. Let me know what I miss.
Thanks in advance, Bogdan.

[A]

Thanks for great review (as always). Can you advise how to squeeze tested 503 mins H.264 / 345.5 mins WiFi v1.3 out of its 85W battery? In your review you got Idle Minimum / 27.5W and Idle Average / 34.4W. So even in idle mode, it is hard to get more than 180 mins. Let me know what I miss.
Thanks in advance, Bogdan.

Power consumption is measured in AC mode on wall socket, battery life is measured on battery.
On battery x86 laptops are a pale shadow of their AC performance/consumption.