Dell XPS 15 9570 benchmarks + temps

Where exactly are those settings for EPP? I’ve been doing it through throttlestop.

Also, I never considered whether power manager affects clock speed as well. I thought it was merely a fan management tool, which indirectly affects clock speed. But, I would think the optimized (or optimal?) setting would essentially be smart enough to turn into the ultra mode during an intensive task.

Has anyone tested this in a benchmark?

 

Since you are running Windows 1709

Windows Control Panel>Power Options>change plan settings (for your selected power plan) >change advanced power settings

Look in "processor power management" > "Processor energy performance preference policy" (dropdowns will be either 0-100% or 0-255)

If they don't show up, you can reference this post to enable the dropdowns, which I clipped from the speedshift thread referenced above

[EDITING]
1. Make sure you are ADMIN in Windows PowerShell (or cmd)

2. Rather than use the GUID Alias (SUB_PROCESSOR ), you might need to use the Subgroup GUID (54533251-82be-4824-96c1-47b60b740d00). For example, to just show "Processor energy performance preference policy", use the first line rather than the second line below:

powercfg –attributes SUB_PROCESSOR 36687f9e-e3a5-4dbf-b1dc-15eb381c6863 -ATTRIB_HIDE

powercfg –attributes 54533251-82be-4824-96c1-47b60b740d00 36687f9e-e3a5-4dbf-b1dc-15eb381c6863 -ATTRIB_HIDE

You can print out your power configuration GUIDs to confirm if you like but I haven't seen that these have changed recently.

3. Sometimes copying code from a browser adds some strange characters or spaces. Sometimes pasting the code into a notepad then into PowerShell (minding the spaces) can resolve that. [END EDITING]

If you open the Dell Command I Power Manager, the four settings state that they will adjust fans and performance. I can't read what registers Dell is changing; I can hear changes in fan behaviour and Quiet is indeed Quieter.

No idea.​

 

Ah, got it. I am running the new windows update, so that is why I don't see it (I think). Throttlestop seems to be doing a good job with it.

 

Well if location or procedure has changed with new windows update, please post that on the SpeedShift thread.

 

I'm very impressed with the CPU actually. CB loop score is nearly double that of the i7-7700HQ 9560 (1128 vs 680). CPU is pulling 60W consistently throughout the test.

 

entering those commands gives me an invalid parameter error.

 

1) Finally Eason's post was the hint. Switching in Bios to S3 Sleep brings back the old energy options.

2) Temps: I repasted both the cpu and gpu with Grizzly liquid metal. I only tested with prime; all [email protected] MHZ, now 70-72°C, before repasting 90 -97°C.

 

How long did you test prime for? I found that my initial temps were much lower but as the heatsink became saturated with heat the average was ~6C lower. Also mine stock wouldn't average near 90C due to power throttling.

 

Not long, about 20 min.
Next days I'm gonna post some pics. Dell did a really poor pastejob (as Apple too).

 

Did you notice an increase in clock speed or did the test ran at 2.700 MHZ before and after the liquid metal repaste?

 

And those are your calculated max and averages? I'm having trouble believing that:
1. you got the system to run near the 90C mark as an average temp
2. you had a 20C drop after LM and a 20 min prime95 run.

Would love to see some temp logs.

 

So thanks to @abujafar for making me go back and look at the realbench logs.

Throttling definitely happens when the DIMM temps reach 63C, just like @lugiank noticed with his 9560.

I put thermal pad over the RAM (cheap 2mm stuff that came with my NVMe heatsink for my desktop) and ran Realbench again (results in the first post). This time in a 15min run the DIMM temps got up to 62C but didn't get to 63C like they did before. The results are pretty impressive. Temps are a little higher from the higher clocks but not unbearable, GPU speed did drop a little from the higher temps but could likely be resolved with an undervolt on the 1050ti...

Sadly running a 30min realbench allowed the DIMM temps to creep back up to 63C and caused immediate throttling. The 8GB sticks in my laptop are dual sided and that may be attributing to the heat buildup. The good thing is the throttling didn't happen for the rest of the test like before and it was able to recover after a minute or so (unlike before when the system just kept bouncing off of that 63C mark).

Up next is iunlock's mod, here's to hoping.

 

Anyone got an i9-8950HK model here that has before and after numbers (idle, cpu load, cpu+gpu load)? I am wondering if I should bother repasting when mine arrives tomorrow.

Also, and not directly related to this thread sorry, but any other issues I should be aware of to check for?

 

Sadly no one I know has the i9. My guess is your results would be similar to mine including clock speeds. The repaste for me was worth it even for just a few degrees cooler temps and much more stable max temps. I would say in terms of issues, just use your laptop and see if anything sticks out for you. Plug in headphones, type on the keyboard, check backlight, check the usb ports, etc..

 

Interesting find that the RAM temperature is causing throttling, my 16GB turned up (ordered specifically for the 9570) and they have metal heat sinks either side, looking like finned aluminium strips.

The same as these

https://www.amazon.co.uk/Ballistix-...1&keywords=Ballistix+Sport+LT+BLS2C8G4S26BFSD

Hopefully this may help with the temperatures. If not how about padding the dimm sensor up to these heat sinks somehow, maybe padding right across the 2 modules in one go?

 

You'll have to run realbench and see what your dimm temps are getting up to. I'm not a big fan of sinking the sensor, we will see what happens with iunlock's mod.

 

Will do, just waiting for the darn thing to be delivered

Looking forward to seeing how you get on with the mod and hoping it works well.

 

@custom90gt your findings are very interesting! If Indeed the DIMM sensor is the only problem for the 9570, is iunlock mod even necessary?
We should just focus on cooling ram RAM (or fooling the ram sensor)!

If you have HWInfo64 logs, please share them.

 

Thanks Eason.

Three things that might help:

1. Make sure you are ADMIN in Windows PowerShell (or cmd)

2. Rather than use the GUID Alias (SUB_PROCESSOR ), you might need to use the Subgroup GUID (54533251-82be-4824-96c1-47b60b740d00). For example, to just show "Processor energy performance preference policy", use the first line rather than the second line below:

powercfg –attributes SUB_PROCESSOR 36687f9e-e3a5-4dbf-b1dc-15eb381c6863 -ATTRIB_HIDE

powercfg –attributes 54533251-82be-4824-96c1-47b60b740d00 36687f9e-e3a5-4dbf-b1dc-15eb381c6863 -ATTRIB_HIDE

You can print out your power configuration GUIDs to confirm if you like but I haven't seen that these have changed recently.

3. Sometimes copying code from a browser adds some strange characters or spaces. Sometimes pasting the code into a notepad then into PowerShell (minding the spaces) can resolve that.

 

So iunlocks mod may take some of the heat away from the motherboard itself and help with the temps, we shall see. What are you looking for in the logs? I can send you my last run for 30 minutes and the ram padded if you'd like? I literally have 61 logs of temps with hwinfo...

 

True, it can help to avoid the build-up of heat.

Just wanted to see what was the frequency response of the system. I also wanted to see if there are other sources of throttling. If what you are saying is correct, it means that there's NO VRM throttling, just a throttling due to the DIMM sensor.

 

Mind it is in fact good that the laptop throttles if the heat builds up. Some folks are reporting freezing and restarts under load, as DPTF doesn't seem to be working. Better throttle than just freeze/reset and possibly even fry.

But if you want better, you'll obviously have to take some VRM heat away, it is plain to see from the thermal images. At least pad the VRMs, or do the iunlock mod, or find something that works reasonably well a bit less fussy.

 

I sent you the log. Let us know if you see anything I don't.

 

Well I've wasted the whole day playing with the laptop and doing different versions of iunlock's mod. Sadly it looks like even though most of the average temps are lower than just padding the ram, the system throttled more... I'm actually pretty confused, but I've been running 30 minutes of realbench stress at a time and that seems to be what the numbers point to. I will run the stress tests like I said I would, but I will likely remove the mod. I even went so far as to build a "copper bridge" from the backside of the ram to the back panel, lol..

 

I think keeping a 62C sensor cool is pretty tough. It is close to room temps so there is little margin for air cooling.

Also you are fighting hot stagnant ambient air in the case and the hot motherboard which heatsinks everything.

 

Yeah I get it. Was just hoping it would help a little. Oh well at least XPS users seemingly won't have to go through too much craziness to get their laptops to run decently. I may try some better thermal pad on the ram to see if that helps any. I don't know if I have anything thick enough in fujipoly but I may stack two 1mm pads to see if it works.

I'm also considering going back to liquid metal to see what we can do there.

 

Agreed - liquid metal might help here. It will allow the radiators to push out a lot more heat from the system. So running at the same speeds, everything should be significantly cooler. You might still need to cap CPU speeds or knock out a few cores. But you won't know until you try.

I assume you used a bit of hi-temp sealing tape between the ducts and radiators to ensure max airflow goes out the back of the laptop.

If you can prop the back rubber foot up just one cm, that will allow the fans to suck in a lot more air.

All a delicate balancing act. Thanks for the interesting posts @custom90gt

 

Undervolted for CB but no performance improvement. Original: 1128. UV: 1115.6

 

Did you try padding just the ram and vrm sensors (not ram itself) to the backplate, making sure that the extra VRM heatsinks are isolated from the backplate as much as possible?

 

How about a very small thermal sticky aluminium heat sink placed on top of the DIMM sensor? At least it would wick some of the heat away from the sensor itself. Unsure how much room there is in between the DIMM Modules but if this is what's causing the throttling its worth trying to get it cooler somehow surely?

Something like this if it would fit and be firmly attached

https://www.amazon.com/dp/B00637X42A/ref=psdc_2998409011_t3_B00E36SEB4

Failing that Dell should rewrite the system Bios to allow this sensor to get warmer than 62C as that isn't that hot in the big scheme of things.

 

I haven't and won't pad the sensors themselves. The heatsinks for the VRM touch the bottom of the cut copper plate and are not directly attached to the backplate. I am trying the 4th iteration without the plate right now, but don't know if I would be comfortable having aluminum heatsinks only connected by thermaltape...

Those heatsinks are way way way too big for that, besides we need to cool the ram and not the sensor itself. perhaps it is getting incorrect or inflated readings, but I would be worried to take too much heat away and have something happen...

 

Good point, and I'm sure you will get to the bottom of this issue pretty soon and have a proper fix for it. I've been chasing Dell up for mine as it said ETA yesterday but still no sign of it. Everything ready to go and itching to get it stripped down and rebuilt....

 

Dang I figured you received it yesterday, that's a bummer. I hate when stuff doesn't come when it says it will...

 

Like @pressing said, I'd try thermal pads preferably arctic cooling 6W/mK of .5mm or 1.0mm pads sandwiched on top of each other.
If DIMM means RAM chips then I reduced some 5-10C temps on max load on my PC and the temps can be checked using ePSA test whilst running cpu test in thorough mode.
@judal57 used thermal pads on over the board and that's reduced like 10C under heavy overclock of his CPU/GPU.

 

Posted some info about the iunlock mod. Unsure if I'll keep it or not at this point.

 

Looking at individual sensors, the only ones that had really good improvements were the PCH and the first Ambient sensor.

Using some compressed air it looks like the first ambient sensor is here:



That would make sense that it would be cooler. Trying another run with just sinking the VRM below and to the left/right of it and then using thermal pad between the heatsink and the bottom cover (a must for me to feel secure taking this laptop with me places). I removed the rest of the heatsinks besides the PCH.

 

Its going to be here within the next 72hrs is all they can advise today, darn it! Although its not here yet I now have the Tag Code and from Dell Support I can see all the component parts and it looks like it only gets the HDD bracket - no mention of the rubber mounts or cable to the mobo, although it may be there but I do doubt it. Why just send the bracket without the essential parts, doesn't make sense to me!

Part Number
Quantity
Description
3FDY3 1 BRACKET, HDD, 7MM, 9550/5510
5D91C 1 BATTERY, PRIMARY, 56WHR, 3C, LITHIUM, LOGIC
98MKC 8 SCREW, M2X4, K SCREW HEAD, MICROSOFT, BLACK OXIDE, NYLOK

 

Understood, thanks. I'm hoping my replacement RAM chips won't need any additional cooling as they do have heatsink strips on them already, but yet to try them out in anger. I have a pack of the 6W/mK Artic cooling pads ready to go if the machine ever actually gets here!

 

That would make a lot of sense as the VRM's close to it would surely leak heat to the ambient sensor - making it think it the machine was a lot warmer than it would be otherwise. I'm starting to think Dell test these machines inside a walk in Freezer!

 

Alright the results section now includes Mk5 as does the iunlock graph. I think I'm done messing with this laptop for a while, wasted far too much time on it.

I'll see how it runs under general use with the mods. The bottom cover does get warmer than without them, but it's by no means hot. I'll try to run some benchmarks sometime and see if the mods actually got me anything, lol. If anything this system is so much more stable than the way Dell shipped it out...

Oh and a side note, it looks like bios 1.1.4 is back up on the site... Wonder if it's the same thing or if they changed anything.

 

Ok so according to Dell Support the 56WHR battery model only comes with the metal caddy bracket, Dell P.N. 3FDY3 (1x 3FDY3 - Metal HDD bracket). Some units may include this part even if a 2.5" drive wasn't ordered from Dell originally.

To add a 2.5" SSD or HDD you will need to order either from Dell directly or Amazon dot com (as they also have them available at the time of writing this)

1x XDYGX - Cable (Motherboard to SSD/HDD)
2x 3XYT5 - Rubber side isolator (either side of SSD/HDD to fit into caddy bracket & fit snugly).

Screws, well as seasoned laptop/pc tweakers I'm sure you'll all have a few spares kicking about!

30 minutes to get through, finally to someone who knew what I was talking about. They were able to check these parts are totally compatible with the XPS 15 9570 and cost around 30 UK pounds delivered - for parts that in my humble opinion should be included with the machine in the first place - but at least they are on their way and allow me more storage space.

Now if they machine turns up with these in place I'm going to scream!

Oh and apologies to @custom90gt for placing this in his excellent Benchmarks & Temps thread by mistake!

 

Starting to put in new benchmarks after the mods. Does anyone have a video encoding project you'd like me to run? While I can't do a totally stock run I can do a mods vs mods+undervolt comparison.

 

Realbench score after mods looks pretty darn good if I do say so myself...

 

I was asking about the pads to RAM in response to
: I put thermal pad over the RAM (cheap 2mm stuff that came with my NVMe heatsink for my desktop) and ran Realbench again (results in the first post). This time in a 15min run the DIMM temps got up to 62C but didn't get to 63C like they did before.

I mean, I understand your position regarding faking the measurement. But if we are pretty certain that the ram temperature causes throttling but is in fact okay while the thottling threshold isn't and we can't change it, I guess padding it to the backplane shouldn't be such a sin. Mind this may not even work if you heat it up the backplate by padding the VRMs to it. You can use a cheapo pad that won't change the temperature that much. Just a few degrees lower out of the throttling zone but so that it can throttle again if the temperature rises further.

I guess the copper plate is important, because it is intended to act as additional fin surface for the redirected air that should be removing VRM heat out via the small central grille. Also if it stretches to the fan areas, they are also pretty cool in the thermal images. And I think in the central part over the VRMs it better be a bit isolated from the backplate, otherwise things may go bad in the long run as with quality pads.

 

I've done hours and hours of testing and haven't had any issue with sinking the ram/pch/mosfets from the VRM to the backplate.

 

Hmm, so you've connected the VRM right there as well as the heat spreader over the CPU/GPU to the chassis? and this has improved your performance under sustained load?

 

Those are pretty good results overall and a nice bump in performance @custom90gt .

Real balancing act with this chassis. . .

 

Yeah, I go from the mosfet to the small heatsink to a low thermal conductivity thermal pad to the bottom of the case. If that blathering makes any sense. After 30 minutes of realbench the laptop feels warm on the bottom where the vrm and ram is connected but by no means is it hot. I only have like 100 different logs if anyone wants more info. I'm doing some encoding tests now just for fun.

 

hmmm so are you saying you bridge the mosfet to the heatsink, then heatsink to bottom of the case? got a pic?

 

I put up a pic under the iunlock section earlier. Basically it goes mosfet-heatsink-thermal pad-bottom case.

Doing a 1hr realbench stress test as we speak. Then all I have to do is a Heaven bench (forgot to run it) and then put the results table up. I'm shocked about the encoding difference for longer files...