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

[Vaidyanathan]

Is the attached ICC profile Standard or Advanced Color? Also, what if you enable HDR? Is the profile still active? Thank you

Hi mario64. The attached profile is the result of our calibration routine. It is not possible to select color profiles in HDR mode.

[Vaidyanathan]

If the panel meets the bandwidth requirements for 120 Hz 3840x2400, shouldn't the 1920x1200 rate be 480 Hz rather than 240 Hz? it should even be able to exceed 240 Hz when set to 2560x1600, and 200 Hz on 2880x1800.

Theoretically yes. But this is something only Razer can answer. I'm planning to ask them in the next few days about what actually goes behind the scenes.

[Vaidyanathan]

The author himself saw that despite the declaration of hardware integer switching in fhd, in reality bad bilinear anti-aliasing switching is apparently used. But how is this possible if NVidia at the driver level gives an integer switch from 2019? I would like to receive detailed explanations from the author.

Drivers perform native scaling and, in fact, fill the empty spaces with black if the native and logical resolution are not exactly divisible. What I meant by that sentence was that resizing from 2400p to 1200p is a direct scale down to 25% resolution-wise. Most likely this happens at the panel level. You make a good point as to then how the image is not as sharp, but this is something that only Razer can best answer.

In the test with CBR15, you should always show graphs for all possible factory profiles, so that a potential buyer can see how much the speed really drops in each profile, coupled with the noise level.

I show how various CPU parameters vary in each performance mode in the HWinfo graph just after the CPU benchmark data. That should give you a fair idea of how well the CPU can sustain (or not sustain) its clocks.

[NikoB]

Drivers perform native scaling and, in fact, fill the empty spaces with black if the native and logical resolution are not exactly divisible

Just a world problem that at the level of the electronics of the panels, a normal integer switching to fhd for 4k panels has never been done. And this has been going on for 10 years. That is why NVidia eventually, after a lot of negatives from users, was forced to independently make an integer resize and, as far as I know, it is done exactly 4 points 4k to one point fhd. But since the driver does it, because of this it loses slightly in performance. This should always be done at the panel electronics level, and not in the OS, then there will be no loss in update speed and there will be no dependence on the driver under any OS and even without it. Here is the problem.

If the picture does not look clear in fhd on a 4k panel, then there is something wrong with an integer exact decrease in resolution by 4 times. Although at the code level, even from the point of view of a schoolboy, the problem is not worth a damn.

I show how various CPU parameters vary in each performance mode in the HWinfo graph just after the CPU benchmark data. That should give you a fair idea of how well the CPU can sustain (or not sustain) its clocks.

It is necessary to clearly show on the CBR15 graph how the processor performance drops relative to the fastest (and noisiest) profile. This is exactly what is not on the NB charts. Moreover, it is often seen that the performance in each profile gradually begins to decline with long calculations, and does not stabilize, but to please the laptop manufacturer, the graph (when such a trend is visible) does not continue further, at least for 30-40 minutes to dot over i.

And it is desirable to immediately indicate the noise next to each option on the graph in brackets. Because the noise measured below takes into account the load on cpu+igpu/dgpu, and more often people are interested in noise when only cpu is working (and cpu+igpu when playing 4k @ 60fps video content for a long time on YouTube and other streaming services). And also locally through, for example, MPC-BE with MADVR (which automatically converts Rec.2020 to Rec.709) for HDR video content at 4k@24-60fps.

That's what potential buyers need - to evaluate such scenarios.

And the key scenario is also interesting - maximum performance profile, but the load on the cpu cores is not more than 35% (averaged, taking into account the impulse). This is the most typical real use case. But there is no noise in this mode in the reviews either...

[Vaidyanathan]

Moreover, it is often seen that the performance in each profile gradually begins to decline with long calculations, and does not stabilize, but to please the laptop manufacturer, the graph (when such a trend is visible) does not continue further, at least for 30-40 minutes to dot over i.

This is not true. I understand you'd want to see as much data as possible, and it's perfectly fine, but please do not make such false assumptions that things are done to please OEMs.

The loop test runs for 25 cycles as standard, which should be generally sufficient to see signs of throttling. There are limits to how much time a reviewer can spend on a particular test and the resources that can be allocated.

The aim of a laptop review is to evaluate the laptop as a whole. If a lot of energy and time are spent in understanding 10 different performance modes alone, then it goes beyond the scope of the review.

Regarding CPU-only load, we do mention Prime95 performance in the Stress test section. CPU+GPU loads while playback of 4K video are shown in the DPC latency screenshots. MPC-HC+madVR is not something that can be replicated well. The function that you mentioned (Rec.2020 to Rec.709) does not work well on certain GPUs and drivers IIRC.

The idea is to use tests that are replicable as much as possible while thoroughly evaluating whether the particular device works well for the advertised demographic.

Hope this helps. Having said that, we definitely value your insights and feedback, so keep them coming :)

[RobertJasiek]

thoroughly evaluating whether the particular device works well for the advertised demographic.

This is bad. You should not repeat the manufacturer's PR. Instead, it should be evaluated what the device can do.

The aim of a laptop review is to evaluate the laptop as a whole.

Now, this sounds better. Therefore, to evaluate it as a whole,...

Regarding CPU-only load [...] CPU+GPU loads

... there should also be evaluation of heavy GPU load while CPU usage is light.

10 different performance modes alone, then it goes beyond the scope of the review.

Idle
CPU load
CPU + GPU load
GPU load during light CPU usage

are 4 (not 10) essential performance modes.

It is not the number of performance modes alone that creates complexity but it is the combination of the three aforementioned load scenarios with different
- softwares
- fan modes
- power targets / TDPs

Recall your / NBC's proclaimed aim to evaluate the laptop as a whole. You do not do so by not answering whether a dGPU notebook has at least one fan mode and at least one power target / TDP setting so that in particular GPU load during light CPU usage is possible at moderate noise in the range 37 ~ 43dB and still good relative speed of roughly 2/3. Although each notebook has different settings, such can - and should - be evaluated for each notebook!

For the ca. 10 RTX 4000 notebook tests of NBC thus far, none has evaluated this. Therefore, NBC's own aim to evaluate the laptop as a whole is unfulfilled. You can claim NBC aims all day long but until your tests fulfil your aims, they are not any better than manufacturers' PR: empty bubbles.

[Vaidyanathan]

This is bad. You should not repeat the manufacturer's PR. Instead, it should be evaluated what the device can do.

Beg to differ. It's not in any way trumpeting the PR. If a gaming notebook is being marketed as such, then it should be evaluated accordingly is what I meant.

Like I said, the tests that we do as a whole should give a more than fair idea of a laptop's performance while also setting a common denominator for comparisons with the existing database. 

While we always try to do as extensive evaluation as possible given the time and other constraints we have to work with (which is actually quite difficult, especially for trending products) — and why not since our readers are very enthusiastic, understand the tech behind stuff and value such info — it's not always possible to accommodate every scenario out there. I believe that the combination of tests we run covers most aspects of CPU and GPU performance and reasonably possible combinations of the two.

And it's not just the processors alone. When I say holistic, it means including other aspects such as display, noise, networking, and other stuff too.

It's one thing to recommend suggestions, and you know that we're pretty receptive to those and value them. Both of us have interacted well many times on several topics.

But misconstruing the whole process as some imaginary PR exercise belittles the amount of hard work that goes into publishing such a piece.

there should also be evaluation of heavy GPU load while CPU usage is light.

Feel free to suggest any test that you think would represent this scenario.

[RobertJasiek]

"I believe that the combination of tests we run covers most aspects of CPU and GPU performance and reasonably possible combinations of the two."

Performance analysis of the CPU alone has been much more detailed than I need. Performance analysis of the GPU varies but has become better and now is often (not always) at least good enough with respect to (rather) maximal performance modes. Therefore, I hardly criticise such any more.

What I criticise is the very insufficient testing of noise and the relation between medium noise and lower than maximal performance. For dGPU notebooks, NBC does not cover most aspects / combinations of CPU and GPU performance and noise. NBC tests little about noise and often even does not clearly specify the test conditions / modes.

Usually, NBC tests about noise in dB for
- Idle (no information on the GPU)
- Load Average (no or too little information on the GPU)
- Maximum Load (CPU and GPU load with maximum noise in typical high mode usually without overclocking etc.)

Sometimes, NBC also tests
- Witcher 3,
which represents GPU load with moderate CPU usage for 3D gaming.

Usually, NBC does not test GPU load with moderate CPU usage for software that is not 3D gaming.

For NBC's noise tests, NBC only occasionally states achieved  (absolute or relative) speeds and the used settings for them. Always stating dB values, speed values and the used settings would be more meaningful.

As someone not using 3D games but using other softwares with GPU load with moderate CPU usage, NBC's noise tests tell me the following for such other softwares:
- Idle: nothing.
- Average Load: almost nothing. It is only likely a lower bound for the noise of "other softwares with GPU load with moderate CPU usage".
- Maximum Load: almost nothing. It is only an upper bound for the noise of "other softwares with GPU load with moderate CPU usage".
- Witcher 3 if done in a loud mode: Either "other softwares with GPU load with moderate CPU usage" is also too loud or "other softwares with GPU load with moderate CPU usage" could have much lower noise in a better chosen / configured medium fan / power target mode.
- Witcher 3 if done in a medium noise mode: More likely than not the measured noise value +-5 dB is the noise of "other softwares with GPU load with moderate CPU usage". In practice, this often means that I only know that the noise is either low enough or too high so I either can or cannot buy the tested notebook WRT its noise.
- Witcher 3 if done in an unspecified noise mode: I can infer almost nothing.

"suggest any test that you think would represent this scenario."

Furmark speed benchmark in some fan mode / power target setting etc. so that the noise approaches at most 43dB, which is the maximum noise I consider acceptable for GPU load with moderate CPU usage in a notebook. (If the test necessarily exceeds 43dB, state this or "the test fails due to too high noise".)

If you prefer, say, 40dB as such a test limit - also fine.

[Vaidyanathan]

NBC tests little about noise and often even does not clearly specify the test conditions / modes.

In the Fan noise section, I mention different performance profiles and the noise levels under each stress condition in each of those profiles as á table. I also show the fan noise graphs separately for each performance mode.

Witcher 3 fan noise is included in every review to show CPU+GPU in gaming. In the very rare case it is not, the reasons would have been mentioned.

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?

If so, it's not practically feasible to limit a device at a particular sound pressure level. Each notebook has its own fan curve. Some keep on whirring their fans for no reason, while others stay absolutely silent. In this case, say if I select the Custom profile with CPU boost and GPU high, the fans keep coming on and off whenever there's a slightest of load above idle. And in these modes, the max fan noise almost always hits 45 dB and above.

The FurMark setting that you suggest might work for one particular notebook, but it may not offer the same fan noise in another. So there wouldn't be any way to standardize the test. 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.

Also, using other software can cause fragmentation too. You might want to check noise levels while doing intensive Photoshop work while someone else would want to know about AutoCAD. The advantage in using synthetic tests is that you will know what the minimum and maximum fan noise levels for a chosen power profile are. By that you can get a fairly decent idea of how it might work for your use case, depending on which component you stress the most.

The artificial load max stress of Prime95+FurMark is only to push the hardware to the hilt. Load average represents very light to medium load on the GPU. Even with this, gaming laptops often hit 43 dB+. It's just how the fan curves are designed. 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.

Witcher 3 represents a more real-world test. It may not matter to non-gamers. The goal here is not to look at any game performance per se but to see how a CPU+GPU real-world stress can influence thermals, noise etc.

In short, I don't think it's practical to set a certain fan noise and work around that. There are simply too many variations in how OEMs design these things. I hope I understood what you meant.

[RobertJasiek]

Your noise test in this particular review is somewhat more detailed than in the other RTX 4000 notebook reviews. So somewhat more helpful.

In this particular review, there are 13 noise values and some graphs. Among the noise values, these have some meaning for me:

Balanced Witcher 40,69dB
Custom Load Maximum 46,72dB
Custom Witcher 46,65dB

The following is essentially useless for me:
- Silent (From other sources, I have heard that Silent fan mode is too slow for GPU load.)
- Idle (Essentially no GPU use.)
- Load Average (Little GPU use does not represent my expected GPU load.)
- Balanced Load Maximum 38,2dB (Illogical value, smaller than Balanced Witcher 40,69dB, therefore the pretended maximum is not a maximum, so I cannot trust this value.)
- Off / Environment (Not GPU load.)

Now, what do the three partially useful values tell me?
- Custom Witcher is almost the same as Custom Load Maximum so does not add new information.
- Custom Load Maximum 46,72dB tells me two things: a) It is higher than my tolerated 43dB so I do not have a simple upper bound for my usage. b) For a "gaming" notebook, it is relatively rather low nevertheless so chances are good that GPU load of my usage might be sufficiently less noisy than the measured CPU+GPU load. However, I cannot be sure that GPU load of my usage is at most 43dB.
- Balanced Witcher 40,69dB and the not so small Balanced PL1 54W tell me: a) Balanced fan mode does reduce noise and for Witcher significantly. b) Probably, GPU load of a different software is 40,69dB +-5dB so it is likely but not guaranteed that it is at most 43dB in Balanced fan mode. c) The relative speed of Balanced Witcher compared to Custom Witcher is unknown (not stated in the text or diagrams). Therefore, I do not know if Balanced fan mode provides sufficient software speeds (at least ca. 2/3) or insufficient speeds (ca. 3/5 or less). So although (a) and (b) look promising, they are actually meaningless without any speeds given in (c). Hence, the value Balanced Witcher 40,69dB is meaningless. (I might study other, external reviews for statements on speeds in Balanced fan modes so as to maybe retrieve meaning. However, only within your review, there is no meaning of this value yet.)

Hence, the only meaningful value for me is Custom Load Maximum 46,72dB and this value alone, although somewhat promising, gives me this conclusion after reading your review despite it being somewhat more detailed for noise tests than most other NBC reviews: I do not know whether I can or cannot buy the notebook as to noise because it remains unclear whether non-3D-game GPU load can run at at most 43dB and at least ca. 2/3 speed.

(To your general discussion, I will answer later.)

[RobertJasiek]

For now, also a short note on the noise graphs:

The blue bars frequency / noise graph tells me nothing because I only hear one loudness and my ears cannot interpret freqency-dependent dB values.

The three coloured frequency / noise graphs tell me relative difference between ambient and different fan modes but I cannot use this information at all. The dB values in these kinds of graphs top out much lower than the measured dB values in review texts. Therefore, I cannot infer any absolute noise value from any curve for any particular fan mode. Hence, these graphs are essentially useless.

In some other reviews, I can recognise that one particular fan mode is louder than another of a curve is completely above another curve, but such would be obvious anyway and not contain any new information.

Instead of useless graphs, there should be more meaningful noise values in the review texts.

[Vaidyanathan]

Balanced Load Maximum 38,2dB (Illogical value, smaller than Balanced Witcher 40,69dB, therefore the pretended maximum is not a maximum, so I cannot trust this value.)

Though expected, it is not always necessary that load max values should be greater than Witcher 3. Sometimes, the vice versa is also possible.  All depends on how the device handles a particular stress condition.

the only meaningful value for me is Custom Load Maximum 46,72dB and this value alone

Of course, since that's the chosen mode for testing. You should base your conclusions about the device based primarily on the Custom mode since that's the one chosen.

The measurements are based on the testing conditions specified in the beginning of the Performance section. The final rating is also awarded based on the initially specified condition and not otherwise.

In this case, Custom was the chosen mode and hence the idea is that whatever measurements have been done should correspond to this. For eg, I cannot set the fan curve to be Silent for taking measurements and say it's great while the rest of the testing has been done in Custom.

However, just so that readers also get a sense of what would the noise levels in other modes be like, I also look at these as well. But do note that this is not always possible. The time we get to do the entire gamut of benchmarking and measurements before the device has to be returned back is already very very limited.

I do not know whether I can or cannot buy the notebook as to noise because it remains unclear whether non-3D-game GPU load can run at at most 43dB and at least ca. 2/3 speed.

I understand your point about perf comparisons vis-a-vis noise, but remember these are stress tests and not benchmarks as such. So, there's nothing to compare here. At best, I can probably indicate the fps in Witcher 3. But again, this cannot be compared with the game's fps data in the Gaming benchmark section as the stress test only needs the character to be stationary whereas the gaming benchmark has a proper sequence.

What we are trying to see is how the device revs up its fans when a known stressor is given. It's not the goal to draw a correlation between various modes, benchmarks, and resulting fan noise. Choosing a particular benchmark here to show trends would be difficult. You like GPU load with minimal CPU load, someone else might prefer Minesweeper :D.

The three coloured frequency / noise graphs tell me relative difference between ambient and different fan modes

I show those graphs exactly for the same reason. The absolute loudness values are indicated in the table above instead.

Instead of useless graphs,...

Wouldn't call them useless :)

[RobertJasiek]

If Load Maximum does not measure maximum noise as I expected, what does it measure?

"You should base your conclusions about the device based primarily on the Custom mode since that's the one chosen." The Custom mode is specified with "Custom mode with CPU Boost and GPU High options", which opposes my software usage because I prefer a low powered CPU without boost and GPU without high options and possibly a tamed GPU. Custom CPU at at PL1 110W and PL2 130W is very far from what I would use. I guess I would, if possible, set PL1 = PL2 = 45W or lower if still with similar GPU load software speed while CPU would be used around 25%.

"these are stress tests and not benchmarks as such. So, there's nothing to compare here. At best, I can probably indicate the fps in Witcher 3. But again, this cannot be compared with the game's fps data in the Gaming benchmark section as the stress test only needs the character to be stationary whereas the gaming benchmark has a proper sequence."

See how much of a problem it is to use 3D game benchmarks instead of compute / work benchmarks to characterise the latter! One needs the latter to characterise it! Then there would also be more meaningful comparison between speed / performance at maximum noise versus at medium noise for medium mode(s).

"It's not the goal to draw a correlation between various modes, benchmarks, and resulting fan noise. Choosing a particular benchmark here to show trends would be difficult. You like GPU load with minimal CPU load, someone else might prefer Minesweeper"

Sigh. When will you / NBC understand that, for modern dGPU devices, correlation between various modes, benchmarks, and resulting fan noise is the most important test subject?! Modern dGPU devices often allow a great flexibility of configuration so that every user can, hopefully, choose settings meeting his preferred noise at still reasonable speed / performance. Therefore, it must be the primary task of reviews / tests to characterise this correlation!

Since different users have different noise tolerances and speed aims, the best would be a series of correlation tests. E.g., some Youtubers show speed at a GPU's 10% increments of power target. They can do so because then they test little else. I understand that NBC also tests every less important aspect of a device so lacks time for a detailed series of the most important noise / speed correleration. Some test compromise can be needed. However, testing only 1 intermediate GPU load setting, such as Balanced Witcher, is not enough by far! It becomes even much worse for devices with wide ranges between average load noise and maximum load noise.

[Vaidyanathan]

I understand that NBC also tests every less important aspect of a device so lacks time for a detailed series of the most important noise / speed correleration

That's not true. What you think might be unimportant may be of importance to someone else. Our aim is to give as complete of an overview as possible of how the device looks and performs.

Your points are noted, though. Thanks for taking the time out to detail them.

But I think I've explained the current rationale quite well, so I'll leave it here for now. Of course, nothing is static and test methodologies keep changing. We also have to strike a good balance between potential newer test methods and data historicity.

[RobertJasiek]

What you think might be unimportant may be of importance to someone else.

Sure. - When I have qualified relative importance, it is my view for my purchase decision of a mobile dGPU device. (Others with similar interests of moderate noise rather than maximum performance have expressed similar concerns. If I want an iGPU device, my preference is rather that it does not even have any dGPU so that noise is lower and battery life longer.)