Clevo Overclocker's Lounge

What's the source on this? I'll be pretty irked if the motherboard ends up being a downgrade over the DM.

 

That would be some serious BS being pulled, if its true.
But then again, if it is, then the Razer Blade Thunderbolt 3 port is worse. Since the scores that im seeing on those laptops are even lower/similar.

USB 3.1 bandwidth maxes out at 10Gbps.

https://thunderbolttechnology.net/blog/difference-between-usb-c-and-thunderbolt-3

https://thunderbolttechnology.net/blog/difference-between-usb-c-and-thunderbolt-3

I would kinda understand if that was the case with the DM2/DM3 , but somewhat skeptical about it being the case with DM1. Specially since the former have 2 x USB-C ports.

Even the BIOS reads as Alpine ridge, which is indeed the TB3 controller.

 

Well...Source would be some of the other venders on here. It was like one of the main questions I was asking about. Haven't you notice how hush hush the adds are about the speed of type-c?
And the DM is only 10 as well.


@bloodhawk
So does this one. So does my Asus, but in the end.... It's only 10. (Says it's thunderbolt)
So it looks like software is not reading Alpine...

 

If the controller was not an Intel one, i would be inclined to agree with you. But the controller on the mother board is an Intel Alpine Ridge (DSL6340, 1 port) and on the TB3 HDK that im using is a (DSL6540, 2 port). The drivers and the firmware come directly from intel as well.
Only the USB-C 3.1 controller is limited to 10Gbps.

Also if that were the case , then my system should not have been able to run 2 x 4k Displays @ 60hz simultaneously.

 

I was just sharing what was told to me. And if all you got was 18k, that pretty much says it's not 40 or 30 or 20 and it's under 10, but that is just my opinion on it. I'm rooting for ya!!

 

Im sure there is just a lot of confusion and misinformation going around specially because there is a low power chip as well that operates on the PCIe 3.0 x2 interface. (JHL6240)

Also 10gbps , will tank the score even more.



And if that controller was being used in the likes of the Razer Blade, they wouldn't be able to charge over the TB3 interface.

I definitely know you are on our side good sir. And thank you for bringing this up, i will dig up as much as i can. Will it be possible for you to take a picture of the controller chip on your motherboard?
Personally all this doesn't affect me much, since this GPU and another one coming next month are going into my desktop, but if your information is accurate, this will be some seriously shady **** being pulled.

 

The last GPU-Z looks normal. Thunderbolt 3 eGPU solutions are supposed to run at PCI/e 3.0 x4. The connection you're seeing is correct.

PCI/e 3.0 delivers 985MB/s per lane, or just under 16GB/s in a x16 connection.

Thunderbolt 3 is 40Gb/s, which equates to 5GB/s (divided by 8), or PCI/e 3.0 x5.

Since x5 doesn't really work for video cards, you're limited to the x4 connection type.

In other words, you DO have the full bandwidth of Thunderbolt 3 in that situation, and neither the DM1 nor the DM2/3 are "gimped". If you really were at the 10Gb/s bandwidth limitation, you'd be running your video card at PCI/e 3.0 x1, or PCI/e 1.1 at x4... you'd be losing to my 780Ms at that point.

If you find that eGPU solutions bottleneck the 1080, then you're correctly benching it at the limit of the technology for the most part. You said ASUS managed to cut overhead; that might very well be true. Won't argue that. But either way, you did say there was a 3-5% reduction in speed. Stronger GPUs and next generation cards will be more hurt by this, methinks.

 

Force enabled the controller and it's the 1577, which is supposedly the DSL6540. The DM has the 1575, which is the DSL6340. Both should be the full 40Gbps.

 

Dont know about ASUS, but Dell's AGA definitely did, but completely bypassing the controller and having a proprietary interface.

But i doubt ASUS' implementation will be better, maybe 5% tops. Specially because they still limited by the amount of lanes, Skylake as to offer.

 

Sorry, I had thought you said ASUS' model did it better. Alienware's eGPU solution connects to a primary x4 adapter and requires a reboot, making it the proper solution for eGPUs as it changes the primary graphics adapter and should get rid of overhead. That's correct.

 

Ah didnt know the technical side of how the AGA worked. But that would definitely help.
I want to test if completely disabling the MXM GPU, helps in any ways. Or if ill even be able to boot for that matter

 

I have no clue about this as I only have 1 4k monitor@60hz

Yes, there is allot of that going around. That is for sure. And that's why we are all here, to hash it out and get the real numbers instead of the hearsay numbers.

Darn, I can't do a x4 run to see if you are right on target. That would help us narrow things down or maybe give some insight.

Edit:
Also, my chip is thunderbolt chip is intel.

 

@tgipier did some PCI/e 1.1 x16 runs (~3.0 x4 speeds) on his Titan X Pascal. Said there was barely a difference in performance for the most part. So it probably comes down to the overhead of the eGPU solution, plus the fact that they use the whole hybrid graphics scenario.

 

Im easily about 4k points off target fora proper PCIe 3.0 x4 link. Thats pretty much the overhead i guess. Ill keep on looking into this, to see if there is anyway to close the gap.

http://forum.notebookreview.com/threads/testing-possible-egpu-bottleneck-with-titan-xp.794641/page-3

What you said, and the fact that there is a conversion going on from TB to PCIe. eGPU solutions, arent really meant for GPUs to be used with laptops. Most are just to enable external PCIe card support.

 

Yeah. Except everybody and their mother is reporting about how good they are, and how amazing it is for eGPUs that don't require "an annoying reboot"... they don't know what they're talking about. Even people here are extremely excited with machines like the P7xxZM and P7xxDM. It isn't worth it. And even if it was, I am also certain that PCI/e 3.0 x4 will bottleneck in some cases. I'm sure it does for 980Tis and stuff. I think it's pretty reliant on the scenarios though, like what game, what resolution, what framerate, colour space/bits per colour/etc, among other things.

By the time Volta and Navi are out, eGPUs, even on a x4 connection, should have a pretty heavy bottleneck. I remember people complaining about 5% back in the day with Optimus, so the eGPU love now confuses me. But that's why I always said if someone wanted to get an eGPU solution to buy laptops designed for it, like the GS30 Shadow back in the day, as that was the best solution.

Maybe if the downsides were actually reported on, the situation might be better off.

 

Oh yeah. Even right now. For TB3 based eGPU's ,anything over the 1070 is a waste of money.

How much loss are we looking at compared to the MXM 1080?

 

MXM 1080s get clean over 120fps as far as I remember, but I could be wrong.

If that's your 1080 and not your 980, then that's a HUGE downgrade.

 

Other data has to go on that x4 link too like sound, USB and hard disk transfers so you won't get the full x4 link most of the time.

 

EDIT:

I can't post internal documentation, but all Clevo TB3 offerings are in fact 40Gb/s!

- PCIe mode (gen 3 x 4): 32Gb/s (31.5Gb/s effective)
- USB 3.1 mode (gen 2) > 10GB/s (9.7Gb/s effective) each for up to 2 x simultaneous devices

Make sure to use an active TB3 cable or you won't get proper throughput!!!

Spoiler

I would also consider Meaker's answer and de-activate the lanes of everything in the BIOS that may get in the way of using the full bandwidth.

From what I have seen Notebook systems using actual Mobile CPUs all have shown decreased performance over Notebooks with Deskop CPUs when used in a eGPU setups.

BTW: Which TB3 firmware are you using atm? The Mod from a while back or stock?

 

@johnksss maybe the ppl u talked to were talking about the second type C port on the phoenix 2. afaik, that one really is just usb 3.1 without tb3 functionality. especially since in official clevo specs tb3 is only specifically mentioned on the first type c port, whereas the other is just type c, nothing else... very suspicious, if u ask me

Sent from my Huawei Mate 8 NXT-AL10

 

If you use both ports you will half the available bandwidth to the other.

 

I'll try to find where it was posted here on nbr.

As to the bandwidth... Here is a 1070 run I just did right now.


Edit: Found the error in my ways!
Ok, I found it.

So I need to find a true thunderbolt enclosure to test the max bandwidth.

Thanks for clarifying for me!

 

Will your setup run a nvme drive on a pcie adapter?

 

Oh yeah, for sure. That's why I was saying. The 1080 is a waste of money for eGPUs over TB3.
Compared to a 989, that's about 30-40% better. Not really a downgrade.

The mod from a while back.

 

Yeap.

 

I am getting a razer core for my blade 2016... do you find performance scales better in 4K res? Or is it still viable for 1080p?


Sent from my iPhone using Tapatalk

 

Actually, I cannot find documentation that shows how many chipset lanes HM170 has. I know Z170 has 20 PCI/e 3.0 lanes, separate from the 16 PCI/e 3.0 CPU lanes, and it's a huge step up from Z97 and Z78 only having 8 PCI/e 2.0 lanes. It's why NVME is better on skylake and you can use more of the drives without cutting into the PCI/e lanes of the dGPU, and there was a discussion about how if someone wanted to use multiple NVME drives, it might be better to grab a Z170 over an entry level X99 CPU.

But if you're saying mobile CPUs are losing performance, then the chipset is more than likely bandwidth starved somehow. I know every USB 3 port is a PCI/e lane from the chipset, etc. But then you have some notebooks with few USB ports, few NVME ports, and mobile skylake that don't seem to have the bandwidth, and then you have machines like the GT73VR with 5 USB ports, 1 USB 3 Gen 2 Type C/TB3 port, multiple NVME x4 drives plus other x2 drives and a SATA 3 drive, as well as wifi etc.

I don't get it. Are there different PCH chipsets that fall under the bulk HM170 name? Because I saw Notebookcheck's review of the P651RE6 and their CPU-Z mainboard shot said the southbridge was IDA14E or something, and the MSI models have IDA150. Is that a key?

 

With how things stand right now. Anything over a GTX1070 , wont be worth it. No matter what.

 

Everything that isn't a lane straight off the CPU has to pass through DMI, which caps at [email protected] (not exactly, but it's a rough approximation) for anything from Skylake up. If you're hooking the NVMe drives up to the PCH on X99, it's all capped at [email protected]. The problem is that if you're writing to a drive at 1GB/s, that's 1GB/s less of outbound bandwidth that can feed the TB3 controller.

 

So all the lanes are basically pointless since the chipset has a direct bandwidth limitation?

If that's the case how do they even offer 40Gb/s for TB3? It can't achieve that at PCI/e 3.0 x4 speeds; it can only achieve around 31.9Gb/s. And further to that fact, multiple NVME drives in RAID 0 would be perf capped hard O_O. The minute you cross 2 drives (considering max transfer for them is around 2.2GB/s in sequential) you hit a cap around 4GB/s.

 

That's exactly what happens. There's barely even a point to RAID0 for drives like the SM961 because the single drive sequential performance already almost saturates the link. RAID0 950s top out around 3.3GB/s I believe, due to overhead and whatnot. The protocol and physical layer for TB3 are capable of 40Gb/s, but nothing can feed it at the moment. Presumably they're counting on something in the future being capable of doing so before they finish the TB4 spec. I'm not entirely certain where the controller lives, but if it was hooked up to an x8 link from the CPU, it could handle it just fine.

 

I didn't mean that... I have a buddy that has a Titan X Pascal in his Razer Core and it definitely it fairing well vs a 1070 and 1080. But I was more or less wondering if the performance was scaling properly on 1080p vs 4k res. I forget where I read somewhere you get more performance loss running 1080 vs 4k. Just trying to debunk or confirm.

 

Gotcha. I don't have a 4k monitor as home, but I'll try to run a test in loopback on the internal panel.
Will it be possible to get his Benchmark scores so that we can compare ?

 

Sure im super busy at work... But I'll get the link for you in a few. His firestrike ultra score's were a little better than a overclocked 1080 desktop card... So honestly not bad taking into account the loss all egpu units seem to get.

 

Alrighty, just did this -
Removed all USB / M.2 SSD'S /WLAN cards.
Disabled those ports in the BIOS.
Disabled onboard audio.
Limited the MXM 980 to Gen1 and x1.
Booted off a SATA 3 SSD with a fresh copy of Windows 10.

The Graphics score still doesn't want to cross 183xx.

@Prema San, any ideas ?

Also I had my friend run a few tests on his core and Stealth , using a 1080 Founders edition , and his score isn't crossing 174xx.

FireStrike Ultra :

http://www.3dmark.com/3dm/15191071

 

Thats not bad for a stock 1080...., whats the issue here?

 

We think that TB3 is bottle-necking hard

With that FS Ultra result, im like wooot!?

 

Here you go!

http://www.3dmark.com/fs/10111432

 

There is something funky with TB3 tbh.
The Titan XP should be further ahead by at least another 500 graphics points.
On desktops the XP easily does, 6700+

 

I always knew there was no real point to RAID 0-ing newer SSDs because the 4K random read/write gets eaten by overhead and it'd be best to leave them separate, but that's just stupid what you're describing.

Well, the more you know. But yeah, I'm glad I see people doing actual eGPU testing so I know more than just theory when I tell people it's a stupid idea unless you get a machine directly designed for it. It's already hard enough telling them that their garbage ULV i7s are worse than the most low-end desktop i3 selling today and that it'll bottleneck the GPUs hard.

Kaby lake mobile had better have a minimum clockspeed of 3.6GHz if they're intending to feed 1060s and up these days. 6700HQ is not good enough. I refuse to recommend the GT62VR for that purpose alone, because it doesn't come with 6820HKs.

 

Well here's the thing... It doesn't seem to matter which card is used. There is always some overhead loss. Based on internal testing of how much new GPUs actually use theoretically the bandwidth from tb3 should be enough. People say oh it's capped blah blah... But clearly it's not capped because as long as faster GPUs still score higher there's definitely no cap the bandwidth is hitting.

Now could it be drivers? Firmware? Maybe... But I honestly don't know. The Razer Core is so damn sexy looking that I'll pay the 500 for it. Im a closet Razer freak. I just hope I can get more information on these tb3 external housings and nail down why the loss is like 20 percent.

 

Yeap, that was the main motive, and to be able to eek out a bit of extra performance for a few months till my desktop is assembled.

I have a PE4C v4.1 coming in next week. So keep an eye out for more tests. I have messaged Prema, lets see if knows of any way to push the TB3 interface to the limits.

Yeah if someone really wants an all in one enclosure, the Core is the best option out there.
Personally for me, this is all just an experiment, and being able to get about 30% more performance, is pretty decent. This 1080 is going to be paired in my desktop with another one as it is. In the mean time, lets see if TB3 can be pushed further.

 

The Kabylake Series 200 motherboard chipset is supposed to solve this DMI PCIE bottleneck and add more lanes as well.

The Kabylake CPU will work with the current 100 series motherboard chipset but those bandwidth fixes are only available when paired with the Kaby Lake Series 200 motherboard chipset.

On Haswell/Broadwell we experienced a similar bandwidth limitation with M.2 SATA RAID0.

My 3x M.2 RAID0 throughput was 1.5GB/sec, adding a 4th M.2 brought the throughput up to 1.65GB/sec on average instead of the theoretical 1.8GB/sec DMI limit due to the bandwidth consumed by additional overhead.

 

I'm right with you I love this external GPU stuff. I actually just ordered the new Razer 14 with the FHD matte screen and 1060 card. Paring that with the Core will be perfect for at home gaming (with my new Asus ROG 4k monitor) and having the 1060 on the go.

I'm currently fed up with all the clevo crap and not being able to upgrade my 980m sli the way I wanted... So that goes on the back burner.

 

Well that's great and all, but we still don't even have Skylake-E yet, and this means all current notebooks are probably going to be obsolete in this regard very soon.

 

Yup, that was my point, we need to get another motherboard chipset to improve bandwidth before considering investing in a 1 time non-upgradeable / upgradeable laptop.

It's always the gotcha details that getcha with these things, even knowing 90% of the details through research doesn't get you that hands on info as to what the functional range of the features really represent.

 

Yeah. Early adopting tech is never a good idea usually, though if you're the right person you can help the community a lot =D.

 

Oh yeah. Specially when it comes to the amount of PCIe lanes external interfaces are able to utilize.

btw. Just noticed, since i dont have an iGPU, i cant seem to get the internal display to use the eGPU. The display works, but the drivers dont. So games/benchmarks are not able to use the GPU.

 

Is that pure stock titan XP? I can run a comparison if needed..... I need to know the settings used.

 

Seems to be, but where is that 2012 Mhz coming from lol.