Feedback to Clevo R&D Team: What's missing or broken in your Skylake laptops?

"What's your WPM on the keyboard?"

"It depends how good you are at Irish dancing"

 

Can't you just see Robert Loggia getting down on that thing?

 

Welcome our new typing team to the company (2:20):

 

this is directly from clevo tw. whats broken in the laptop would be to add an additional 4th fan to the laptop, basically 2nd fan for cpu heatsink.

 

Finally, a notebook that can fit enough cooling hardware and battery to satisfy the needs of power users.

 

plot twist, turns out that display isnt 4k

 

Looks like a projector in the booth, likely 720p given the age of it

 

You're right, its 64K

 

Mine has to hinge on one primary complaint: BIOS updates.

I'll provide an example: I own a variant based on the Clevo P751DM2-G (Branded as the origin Eon 15-X), and almost one year after launch, no new BIOS update is available.
Yet, we have other vendors with similar offerings who have been gracious enough to offer BIOS updates for their SKUs, such as Obsidian PC.

Clevo should enforce a Quality Control scheme where all their vendor partners are compelled to avail BIOS updates to their users , including the EC and TPM firmware.

Three pending issues I have open at the moment, all related to the stock BIOS that came with this machine are as follows:

1. Intel's Speed Shift (SST) for both SKL and KBL is not yet available with the stock BIOS,meaning potential performance losses.

2. The Skylake and Kabylake hyperthreading bug remains unpatched.

3. The TPM by Infineon Technologies still runs an older and vulnerable firmware version, which means that the vendor (aka Origin PC) is compelled to release a firmware update to address this, as directed by the security advisory here.

And to top that up, the issue with the VT-d IOMMU grouping issue (that nukes the internal SATA controller when Thunderbolt docks are plugged in unless overriden by the Linux ACS patch, as described here) still persists.

As an Origin PC client, this may be the first and last purchase I make from these guys. We shouldn't have to beg for baseline BIOS updates that actually fix security and usability issues.
For instance, due to the unpatched Infineon TPM firmware issue (which you can confirm by running tpm.msc on Windows 10 Pro), I miss out on both:

1. Full drive encryption with Bitlocker.

2. And what sucks the most, TCG Opal 2.0 Self Encrypting Drive (SED) support. My workflow relies heavily on sedutil and a PBA to unlock my self encrypted drives prior to launching the bootloader, and with an untrusted TPM, I cannot use this feature.

3. Sensible power adapter configuration: Why Origin PC does not include the 330W power adapter as a configuration option for the P751DM2-G variant (Origin Eon 15-X) is baffling. This is a unit, that on stock, will gladly draw approximately 270W from the wall socket under full simultaneous GPU and CPU load. I'm glad to see other vendors (Eurocom, Sager and OBSIDIAN PC) include this sensible option, where others (including System76) fail to do so. Why the 330W power adapter is not the default power supply with an unlocked Kabylake processor and a desktop-grade GPU is an oversight that has serious performance implications for the user down the road, which calls on stronger quality control checks for all suppliers.

Here's where I will issue this public warning (for future reference), on a matter related to a measurement commonly referred to as the TDP. The TDP value you see plastered on Intel's ARK for specific processors is NOT a measure of power draw on the processor under any circumstances, but rather, a hint for the system manufacturer that directs the amount of cooling needed to dissipate heat at a rate that will keep the specific processor within operating temperatures under the worst case scenario. What you should be more concerned about is the actual power draw, measurable through a wall socket meter, and not on utilities running on the host operating system (which is why Intel's RAPL and GPU power draw estimates from NVAPI and AMD's ADL is not considered reliable, but as rough estimates, at best) and this will often dictate actual performance under most workloads.

Remember that under full load, a processor such as the unlocked 7700k and similar mobile processors with unlocked multipliers will draw as much power as is needed to sustain their clock speeds, as as a rule, will often exceed the rated TDP especially when overclocked (also configurable via the power limit setting in the BIOS). So if you have one of these so-called "extreme" gaming laptops with a hint of an unlocked processor underneath, WATCH your power supply's limit closely. It's most likely to be your limiting factor. Quoting some worth while examples:

(a). The same chassis exhibits premature throttling (amperage-based) on the stock 230W adapter, as observed on the Schenker XMG (2017, 15" SKU) an issue that some reviewers have documented on, in notes or in passing: https://www.notebookcheck.net/Schenker-XMG-U507-Clevo-P751DM2-G-Laptop-Review.209180.0.html

(b). The infamous MSI GT80 Titan that CANNOT make use of the dual PSU or the 6920HQ chips well with SLI 980M and SLI 120W 980 solutions except for the purpose of benchmarks. The machine is designed to use a 330W power brick and draw any extra power from the battery, and will run down the battery to 0% after a couple hours of gaming (happens at stock clocks with 120W 980 SLI) at which point the machine will throttle components to make the 330W brick suffice. While attaching a second 330W brick is physically possible, the machine's EC is set to ignore the extra power and use from the battery, and an EC hack is required to make it work properly. This is not acceptable.

I believe, firmly so, that someone who considers a Clevo based laptop that will sink in thousands of dollars does so as an investment, and as such, expects uncompromising feature support as on the date of purchase towards the longevity of the notebook. We applaud these vendors (such as Sager, Eurocom, OBSIDIAN PC, etc) that take this seriously and are prompt with firmware and software updates, but this kind of lapse with other vendors should be monitored, closely.

And another issue, which affects Linux users: Hardware enablement patches are welcome, particulary around issues such as Audio device setup (hands up if your headset jack doesn't work on Linux!) , and to that front, we applaud the likes of System76 and Tuxedo PC. Keep it up!

Edit: After I received my 330W power adapter (Thanks Eurocom), I did re-run some benchmarks and testing confirms that this is the appropriate PSU.

See this:

1. Initial result: http://www.userbenchmark.com/UserRun/5587729

2. Follow up: http://www.userbenchmark.com/UserRun/5952654

The follow up is with the new 330W PSU.

And yes, 3DMark is seeing some substantial gains.

Sky Diver went up by 2000 points
With the 230W brick https://www.3dmark.com/sd/4861157
And with the 330W brick: https://www.3dmark.com/3dm/23514783

Further followup: With @Prema 's help, I was able to get his awesome BIOS flashed on the P751DM2-G and all throttling issues are gone. The AVX offset feature is very useful for Kabylake. And of great satisfaction is the fact that all the issues I had with the stock BIOS, even on Linux, are gone. No more ACPI errors on Linux, etc.

What baffles me is that with such awesome people like Prema around, his BIOS isn't a default on all Clevo systems. Seriously, hire this guy and partner with him as many have done. This is the type of contributor that Clevo's R&D team should back up and work with, and maybe even give him a full time position without the corporate nonsense tied around gag orders and NDAs.

 

Have you emailed Origin your concerns?

 

I have, over several threads. They seem to be taking them seriously, but as for the BIOS updates, not much on that front.

 

Interesting, good luck with them and I find it surprising too.

We do offer files to customers on request and while we do verify them internally before we update them we do try and keep them up to date with the latest from Clevo.

 

I've experienced that before with Sager, having owned a few masterpieces before.

 

Well, Origin has just posted an update on the TPM issue (and also mis-identified it as an MEI security issue) here.

 

You have to be logged in to see that page by the way.

 

CLEVO if you're reading this, PLS FIX THESE:

- laptop chassis redesign, to cool 8 core overclock 9800k will need CPU heatsink with TWO fans.
- must be able to switch between dGPU and iGPU
- more 2.5" sata storage bay, currently 870TM has 2 is fine but more is welcomed

 

maybe add support for 15mm 2.5" drive also.

 

Why not get a desktop?

 

Honeslty, I have a laptop connected to a large display and two large speakers.

The only reason I don't get a desktop is because I need to move between two locations sometimes and I need my processing power on-the-go.

Having a laptop means more power than most compact desktops like this, Plus I don't need to have two full setups in each location.... Although I do have two full setups, one in each location.

In guess it is less expensive right now.

Most people hjave a situation similar to mine...

 

Yup. Originally I had two work places, was traveling more, I am incredibly impatient when it comes to spinning hourglasses. Using a laptop means I could take to pub next door, sit in beer garden instead of corporate office. Now, I can work in my home office, or backyard or downstairs or rear studio. I'm still incredibly impatient when waiting for computers to do things since my mind is usually a few steps ahead of what I can input, and wanting a shiny 17" panel to look at and enough power to power it smoothly when I'm sick of working. No other option.

The huge DTR niche exists because enough people want the product. Simple capitalism! Tired of unimaginitive people not understanding this.

 

- Option has been left open in the P870TM models
- Not a must have feature for most people
- No demand for more from most people (an effective 5 bays with M.2 covers nearly everyone for potential storage)

 

It is better to have only the dGPU, it lowers input lag and display lag by A LOT. I had a few laptops with iGPU, and although the dGPU in there can be very strong, it usually has a lag to itself, sometimes as high as 10 ms. Add that with the lag of your monitor, and you're on to skipping some frames like there's no tomorrow.

 

what georgel said, also no battery for power failure etc etc

- 870TM has it now? iGPU?
- for iGP and dGP? thats cause they dont understand the good thing about it.
- well we asking for em right now, more always good

woah didnt know that, i dont notice much using my old alienware with iGP to game though, dont feel a difference.

 

they are trying to kick us out saying no one needs it, yet they already have pretty much entire market to themselves with all the bga trash and thin hardware junk. there are thousands of options out there for those who want bga soldered filth, they can go for it, but dont come around telling us DTR thick notebook replacement is dumb, because they are dumb and only see their own pov while we can see both sides.

 

It is a real thing, my friend.

You might not fully notice it because they use some predictive tricks to mask it, but the total input lag can go above 30 ms with the iGPU in the way sometimes, especially for large games, leading to a lot of dropped frames on AAA titles. IT can make a huge difference running something on Ulktra high settings that barely reaches 60 fps

Example

Dishonored 2 will barely reach 60 fps with everything at max, on GTX 1080, on a QHD screen. This is with a laptop that has only DGPU or a desktop.

If you place an iGPU in the middle, some frames need to be passed before they are ready, so they are either delayed or dropped, it looks awful...

I do a lot of video work and such, especially with madVR, which uses videos that are 24 frames per second, and adding an iGPU in the way added a ~10 to 20 ms of display lag, sometimes dropping the frame entirely, leading to a very juddery motion. This is not related to anything else, and it is just the extra road the signal has to travel to reach the display because the iGPU exists.

It works kinda like: the dGPU is not connected to the display or anything else, so the signal is streamed through the same interface as the iGPU, taking a much longer road than with dGPU only.

This is kinda why I don't want a dGPU + iGPU laptop....

 

so the mobo design adding igp path will increase latency, thats pretty crappy, but im sure desktop side of things will always allow iGP and dGP to work and capable of switching between them and if its desktop, a lot of them dont have this problem otherwise it'd pointed out by reviewers by now.

im thinking just laptop OEM not doing quality work when making the laptop mobo, im still thinking that even if they were to allow iGP to work and do it right, it should work right.

 

With a hardware switch there is no latency penalty when optimus is disabled. It's the added cost per unit vs how many people would actually use the feature on the larger machines.

 

Hey about the mechanical keyboard thing, I think that is becoming much easier now. I mean, even consumer grade desktop keyboards seem low profile enough to put in laptops. That might be an interesting mod to do

 

Nope, it is a motherboard issue only on laptops...

Basically, they don't have a switch. The data stream doesn't pass to anything, it passes from the CPU and iGPU to the dGPU, then back to the CPU and the iGPU, then to the display or the display port. On dGPU only lappys, the data goes one way to the dGPU, then straight to the monitor or the ports, this is why they don't work at all without the dGPU.

In their advantage, this works fairly well for low power video cards and BGA stuffs, but really this is a very failed concept for high end, where we talk about GTX 1060, 1070 and 1080.

This gets worse with the increase in performance, meaning that the more power is needed to process a frame, the worse this lag gets and the more you're prone to noticing it. High end games at high end settings are basically unplayable, and I saw that on my own skin as I owned laptops that did this. Never again, my friends

And @Meaker@Sager knows about it very well, but he can't hurt his company's sales by openly stating that having the typical optimus switch adds a LOT of latency.

I've never had a hardware switch, so I'm not sure how that works, but I know that most Optimus implementations are really bad and they will have a deep impact on performance, to the point where it is absolutely not worth to own and use a lappy that is high end and has optimus. That hardware switch might help, but I won't speak about something I don't know...

 

Wow never knew that! Would it be hurting performance on 970m laptop? It's bga btw yes sad I know

Sent from my Moto G (5) Plus using Tapatalk

 

Optimus doesn't reduce performance, it adds output lag, which is what Georgel was referring to. The added latency is why Optimus is unsuitable for VR applications.

 

Ok. So regardless of GPU/CPU there's about an extra 30ms of input lag?

Sent from my Moto G (5) Plus using Tapatalk

 

Yes, it degrades the top performance of it...

Yes, theoretically it lets the GPU process things, but adds output lag, but if the frame reaches late enough, it is dropped. Think about it like this, what is the difference between cards, if not framerate... Well Optimus makes it drop frames, and sometimes it drops more than half of the frames because they are delivered way too late.

It is not always so bad, sometimes it is 10 ms, but you have 16ms to present a frame, otherwise it is dropped.

If it was 30, then you'd see no more frames, because all frames would reach it way after they were supposed to reach the display.

The major difference between inpud and output lag is that frames that reach the monitor too late from output lag are just dropped, while input lag can be just lag.

 

Nvidia said 0.2 of a frame. That's 3.33ms at 60Hz, or 2.22ms at the 90Hz VR standard. The lower the frame rate, the higher that added output lag is.

 

Well, I experienced around 10 ms of added lag in practice, and measured by the laptop itself. That means that from the 16ms the GPU has to process the frame, it needs to send the data 10ms faster for it to arrive safely, which means that the GPU can only reach roughly 50% of its potential before the output lag kills it.

Like anything, trust actual measurements, not company marketing specs.

 

Hmm....I still don't really understand...So in order to get steady 60fps, then the GPU has to push a new frame every 16.666...ms. Is that where you get the 16ms? What do you mean by the 50% before output lag kills it?

Sent from my Moto G (5) Plus using Tapatalk

 

That may be what NVIDIA says...but that would be under the very best circumstances, and is rarely achieved in real world use.

 

Getting nvidia and OEMs to work out how to completely power down slave GPUs when not used would be good. Pointless to have a bit of hardware sitting there sucking up to 10W permanently doing absolutely nothing.

 

Yes. to output 60 frames, you need to push a new frame ever 16.6 ms. If out of those you have 10 ms of delay, you either push the frame at 6.6 ms, or it is dropped.

The output lag doesn't do delayed frames, it drops them entirely...

Now imagine adding the input lag of an external monitor to the output lag of the Optimus setup.

Optimus is the optimal lag solution!

 

I honestly recommend people to work with HIDevolution a lot because they know their stuff and are forward about it.

I want my seller to tell me what I'll experience, if my work requires me a dGPU only, I need to know that, this is why they provide a pretty excellent service.

I need to have all my data to make a very conscious purchase.

In all honesty, if I'm spending 1.5K$, I'd better spend another 0.5k$ than to purchase something that won't fit my needs.

Optimus exists for very very low power solutions, a GTX950 doesn't really see this issue as much as a GTX 1060 or 1070, where this delay is just wasted potential.

 

Ok..I think I almost get it, but how come it HAS to push 60 otherwise it skips frames? Also, is the displays input lag taken into account when deciding whether it should drop or not drop frames? Lastly, if the frames are just dropped, why is there increased input lag?

Sent from my Moto G (5) Plus using Tapatalk

 

This needs a bit of clarification

1. It doesn't skip frames, it tries to do as many as it can, if it fails during the render, it drops the said frame. Most GPUs are always under the maximum possible stress, if it is not frame capped, with something like Vertical Sync or such.

2. Nope, only the moment the signal reaches the monitor. Basically, if the GPU knows that the signal travels another 10ms to the monitor, and that it took it 10 ms to render it already, it decides to drop that frame and leave the one that is already there, then start a new frame. IT s much more complex than this, but this is the basic principle. And it usually is a complex motherboard chip that decides this, not the GPU nor the CPU, but considering that everything we talk about is most probably soldered, it is the same motherboard.

3. Output lag, not input lag. Because the signal will always have to travel a longer path than without that optimus thing. Regardless of whether the frame was dropped or not, the next frame still takes that time to reach the display.

Thius is why I say this caps the power of most GPUs at 50%.

Poor GPUs either couldn't process the data faster anyways, so they aren't as affected, or they sometimes just just are affected, but the user blames it on mobile nature of the GPU.

 

Basically, the GPU cannot send a frame forward, if it is too late.

Because

It took the GPU 10 ms to render it

It makes 10 ms to reach the display

20 ms

The next frame has a duration of rende runknown , but it will surely take another 10 ms to travel, so that next one could also be delayed, and the GPU cannot make a predictory calculus of what will happen with too many frames.

Let's say it sends the first 10 frames late, there is a point, where it needs to recover from this thing because it can't predict the next frame.

At 60 fps you imagine this isn't verty bad, but I experienced optimus typical throttle at 24fps...

 

Dang so your saying if I had the same exact laptop non Optimus, I'd be getting double the performance? What exactly does output lag entail?

Sent from my Moto G (5) Plus using Tapatalk

 

All frames being delayed an extra frame/part of in no way affects the GPU's render workload since once the frame is completed the GPU core is onto the next frame (unless told to wait by vsync/frame cap etc) and doesn't care how long it takes once it 'leaves' the GPU for that completed frame to be displayed.

Shuffling data from one spot to the next always takes time, optimus introduces an additional delay in the pipeline, its bad, just a degree of whether the badness is noticeable and observable to the user, or not. This is kind of getting OT though since noone is asking for a MUX switch to game on anyway?


I highly recommend the series of videos by a guy called "battle-nonsense" on youtube. He has some great explanations of how display buffers work, what tearing is, along with technologies such as gsync, vsync, adaptive vsync, low vs high refresh rate monitors, all kinds of things. Along with lag comparison testing which I have seen literally nowhere else.

 

Blur Busters.

 

panel refresh/transition != "click-to-bang" delay

 

Welp, I did developing tests, which you kinda need to take from a developer - just having optimus reduced the performance and added frame drops because the frame would be way too late. That output delay can cause frame drops, opposed to input delay, which cannot do that.

 

It is a Microsoft problem, not a hardware problem. Look at Apple's implementation of switchable GPUs on their MacBooks—they've had it since 2011 at least. Apple's proprietary switchable solution only works on OS X, and is Intel/AMD/nVidia-agnostic. Install Windows on a MacBook Pro, and see the battery life plummet as only the dGPU can be used, albeit with full control over the display.

Key thing is that Microsoft needs to allow proper plug-and-play in Windows—currently PCIe devices work like PS/2—needs a driver install and reboot cycle to work entirely properly.

It is perfectly possible to connect both the iGPU and the dGPU to a multiplexer, with neither as master nor slave. Windows ought to be smart enough to switch to the dGPU upon detecting draw calls from the application in question. OS X can do it, why not Windows?

 

https://www.blurbusters.com/gsync/gsync101-input-lag-tests-and-settings/

https://www.blurbusters.com/category/inputlag/