Dell Precision 7670 & Dell Precision 7770 owner's thread

[MyPC8MyBrain]

43 minutes ago, PHVM_BR said:

In the graphics tests this consumption is around 190W

therefore my logical conclusion would be... if the test afterwards was able to draw 225w, at the least this suggests dgpu had 35w buffer it choose not to/or was unable to utilize in the dgpu focused phase before!

the premise of lack of AC power was proved wrong with the 330w brick tests i run,
to me it is obvious that something is actively preventing dgpu from even reaching its current 150w declared TGP even with plenty of power available and thermal buffer (small as it may be),

[PHVM_BR]

59 minutes ago, MyPC8MyBrain said:

therefore my logical conclusion would be... if the test afterwards was able to draw 225w, at the least this suggests dgpu had 35w buffer it choose not to/or was unable to utilize in the dgpu focused phase before!

the premise of lack of AC power was proved wrong with the 330w brick tests i run,
to me it is obvious that something is actively preventing dgpu from even reaching its current 150w declared TGP even with plenty of power available and thermal buffer (small as it may be),

 

I believe it's not a thermal cap or power cap issue, it's an imposition of CPU + GPU summed TDP at high GPU loads. Games or TimeSpy for example.

 

My theory is that in the TS cpu test this limit doesn't come into play because the load on the GPU is low.

 

So your CPU reaches 157W + X Watts from the GPU + X Watts from the rest of the system resulting in ~225/230W.

 

Already in the graphics tests, the GPU must be consuming ~120W and the processor limited to ~50W, resulting in ~170W and making it impossible for Dynamic Boost to come into action.

[MyPC8MyBrain]

10 minutes ago, PHVM_BR said:

My theory

speaking with hands on experience your theory didn't hold,

 

12 minutes ago, PHVM_BR said:

it's an imposition of CPU + GPU summed TDP

cpu stress test runs after gpu test, during that time gpu had at least 35w overhead to consume more wattage,
it simply didn't, the test afterwards had no issues demanding and consuming more with thermal headroom,

 

[PHVM_BR]

27 minutes ago, MyPC8MyBrain said:

speaking with hands on experience your theory didn't hold,

 

cpu stress test runs after gpu test, during that time gpu had at least 35w overhead to consume more wattage,
it simply didn't, the test afterwards had no issues demanding and consuming more with thermal headroom,

 

You didn't understand what i wrote.

 

Probably due to my limited (terrible) English...

 

In TS cpu test GPU load (%) is low.

 

The imposition of the CPU + GPU power limit (~170W) only comes into play at high GPU loads (like in the TS graphics tests).

 

All this according to my theory.

 

If you want to confirm whether it's right or not, just follow the real-time CPU and GPU power consumption data during TS.

 

There is no point in checking at the end that the maximum consumption of the CPU was 157W and the GPU was 120W as this is not achieved simultaneously.

[MyPC8MyBrain]

i believe i understand the point you're trying to make, it just didn't work the way you envision it,
locking PL1 and PL2 to 55-65 did nothing but slow the test down to a crawl with a score of 9k, 
dGpu usage wattage/TGP wise remained the same regardless of manual cpu cap i imposed,

sound like you believe the cpu "reserves" its entire TDP range which is simply not the case,

 

[PHVM_BR]

On 12/18/2022 at 4:25 AM, MyPC8MyBrain said:

i believe i understand the point you're trying to make, it just didn't work the way you envision it,
locking PL1 and PL2 to 55-65 did nothing but slow the test down to a crawl with a score of 9k, 
dGpu usage wattage/TGP wise remained the same regardless of manual cpu cap i imposed,

sound like you believe the cpu "reserves" its entire TDP range which is simply not the case,

 

On 12/18/2022 at 1:57 AM, PHVM_BR said:

Edited: checking the total system power consumption on your video during graphics tests it is likely that there is some limitation imposed on CPU + GPU power limit at high simultaneous loads.

 

This limit is probably ~170W.

 

If this is correct then for Dynamic Boost to kick in the CPU power cap should be ~30W.

If the total power limit (CPU+GPU) is 170W you cannot lock the PL1/PL2 to 55 or 65W because that will make the TGP stay at the base value without the extra power of the Dynamic Boost.

 

For Dynamic Boost to work the processor power must be ~30W to leave ~140W for the GPU.

 

Without blocking power caps with ThrottleStop on other systems (depends on Dynamic Boost implementation) at high simultaneous CPU/GPU loads GPU power should run at values between its base TGP and maximum TGP (base TGP + Dynamic Boost) and the power of the processor will vary completing the 170W (in this case) according to the demand.

 

Enough theory for today.

 

Good luck with your adjustments!

 

@MyPC8MyBrain

Take a look:

 

 

https://notebooktalk.net/topic/23-precision-7670-7770-pre-release-discussion-alder-lake/?do=findComment&comment=8307&_rid=574

 

[MyPC8MyBrain]

53 minutes ago, PHVM_BR said:

Take a look:

at what? nothing stated there contradicts or conflicts with my result or suggests a different configuration, 

i have 7770 in hand and tested your theory even with PL1/2 both locked at 30 the dGpu still only draw max 120w out of allotted 150w no matter what steps i take, and I've even went the distance shutting every peripheral off in bios to make sure there's no current waste, i even went further distance and order a 330w Dell power brick and still no go, no matter what i did dGpu will not draw beyond 120w (you can watch the bench i run with 330w power brick few posts back, that's with bios recognizing 330w brick with matching PL4 limit, that test alone rules out any TDP+TGP power limit theories out there).

 

ill disclose few random observation there was one time never to happen again where i saw 130w for a split second, with that i also saw 750w max dGpu power limit, i believe its some math fluke by HWiNfo as it never accrued again, on the other hand 115w-120w max dGpu usage is consistent throughout.

 

i am very happy with CPU performance, dGpu is a BIG disappointment from getting a capped sku to not having it even perform to its cap is another BIG disappointment, 

[win32asmguy]

8 hours ago, MyPC8MyBrain said:

i am very happy with CPU performance, dGpu is a BIG disappointment from getting a capped sku to not having it even perform to its cap is another BIG disappointment

 

I think the best way to be happy with the 3080ti option as a consumer purchase is to only consider it in a refurb model. If you are paying around 2000 then it's perfectly fine for it to perform in the lower bracket. Even with the lower performance it can run games well enough while being a business chassis.

 

[PHVM_BR]

9 hours ago, MyPC8MyBrain said:

at what?

 

I understand that this answer from Dell-Mano_G means that there is a fixed, locked, limit on CPU+GPU power of 162W.

From your videos I believe that the limit is activated with high loads on the GPU.

 

9 hours ago, MyPC8MyBrain said:

i have 7770 in hand and tested your theory even with PL1/2 both locked at 30 the dGpu still only draw max 120w out of allotted 150w

 

Have you tested with Intel Power Max Turbo Boost 3.0 disabled?

 

https://notebooktalk.net/topic/632-dell-precision-7670-dell-precision-7770-owners-thread/?do=findComment&comment=17016&_rid=574

 

perhaps this limitation on joint CPU and GPU loads can be removed with a BIOS mod...

 

It really is very disappointing.

 

I expected on the 7770 at least 180W on the CPU + GPU simultaneously, with values varying between 125+55W and 150+30W.

And without a deliberate blocking of that value...

[MyPC8MyBrain]

On 12/19/2022 at 3:55 AM, PHVM_BR said:

Have you tested with Intel Power Max Turbo Boost 3.0 disabled?

we are all here pretty much disappointed as well (accept maybe one owner),
i have not tested with turbo boost 3 off, and i have not noticed that it disables dgpu boost,

 

On 12/19/2022 at 3:36 AM, win32asmguy said:

I think the best way to be happy with the 3080ti option as a consumer purchase is to only consider it in a refurb model.

agreed 100%, new ordered from Dell the 3080Ti option alone is over $2200, paying that price for sku that's not performing as expected is the same as flushing hard earned money down the toilet, i paid almost $4500 for my current 7770 configuration out of that $2200+ is just for the 3080Ti, for $2000 yes id absolutely keep it and be happy with with its very low (relative) performance, for $4500 I'm afraid it would be a no go for me,
 

 

my dGpu bus is capable and should be running PCIE x16 4.0 lane speeds,
currently it is only running at PCIE x8 4.0/1.1 speeds at best,

can someone please confirm if this only my system please,
(i tested a single nvme in every slot to rule out PCIE lane sharing)

[developer]

3 hours ago, MyPC8MyBrain said:

my dGpu bus is capable and should be running PCIE x16 4.0 lane speeds,
currently it is only running at PCIE x8 4.0/1.1 speeds at best,

can someone please confirm if this only my system please,
(i tested a single nvme in every slot to rule out PCIE lane sharing)

 

Hi,

 

7670 with a3000, showing PCIe x16 4.0 @ x8 2.0

 

I think this is worst than yours.

[Aaron44126]

9 hours ago, MyPC8MyBrain said:

my dGpu bus is capable and should be running PCIE x16 4.0 lane speeds,
currently it is only running at PCIE x8 4.0/1.1 speeds at best,

can someone please confirm if this only my system please,

 

This is normal.  The DGFF connection between the GPU and the motherboard is just eight lanes.  There are not enough pins for 16 lanes.

 

This was also the case for prior-gen Precision 7X60.  Some older systems did run at 16 lanes with an extra DGFF data connector.  PCIe4×8 has the same bandwidth as PCIe3×16, so I think that Dell took the opportunity to free up some board space rather than increase bandwidth when the PCIe4 upgrade came along.  Ampere doesn't really need 16 lanes of PCIe4 to reach full performance (especially when you consider how power-starved the GPU is in this case).

 

6 hours ago, developer said:

7670 with a3000, showing PCIe x16 4.0 @ x8 2.0

 

This is also normal.  Under standard configuration, the PCIe link speed will change dynamically, you won't normally see it running at PCIe4 speed unless the GPU is doing work.  (You can override this behavior with Windows advanced power settings... but there's no real reason to.)

[MyPC8MyBrain]

the replacement for my original replacement should be arriving tomorrow
(to Dell's credit they sent two replacements, the first replacement had some dGpu issue),

i decided to run another test on the original unit, i removed dGpu completely out of the chassis,
the goal was to see how the chassis will perform with the dGpu plate dedicated for cooling without anything below it,

results were interesting AIDA64 was unable to get it to thermal throttle, temps stay at max 78c or below 80,
that's with me changing and locking a new TDP (going off Intel XTU own auto tuning numbers)
i changed and locked PL1 at 110 and PL2 at 165, CB23 single run score is 25700 atm, 

during CB23 run system consumed around 210w sustained (dGpu out of chassis),
cool down is almost instantly back to temps below 40c, and idles at 31c,

 

[Aaron44126]

Interesting that the system boots with nothing in the DGFF area. Precision 7X30-7X60 would fail to boot without a dGPU, or the iGPU pass through card. (In these systems, the mDP and HDMI ports were physically on the dGPU card, so the iGPU pass through card was needed to pass those ports to the integrated GPU.)

 

I wonder what the iGPU spacer card actually does in this system…?

[Aaron44126]

NVIDIA graphics driver 517.66.

https://www.dell.com/support/home/en-us/drivers/driversdetails?driverid=C33HW

[win32asmguy]

Just a heads up, bios 1.8.0 fixes IA AC loadline values to the correct values. Tested this on a fresh machine running 1.6.3 from the factory.

[Aaron44126]

10 minutes ago, win32asmguy said:

Just a heads up, bios 1.8.0 fixes IA AC loadline values to the correct values. Tested this on a fresh machine running 1.6.3 from the factory.

 

Thanks for the heads up.  I updated the OP to reflect this.

 

On 12/14/2022 at 4:50 PM, MyPC8MyBrain said:

interesting, there's another SunOn fan model,

you seem to have a model with dense fin formation like the Delta in one of the systems i received,
the Sunon in my current system has a less dense formation then yours (i noticed it immediately when i saw it)

 

Coming back to this one...

My fans are Delta, not Sunon.  (I have not yet had the assembly replaced.  Not sure if I will, since I have things working good for now.)

[MyPC8MyBrain]

5 hours ago, win32asmguy said:

Just a heads up, bios 1.8.0 fixes IA AC loadline values to the correct values.

better late than never, well done Dell!
i wonder if they implemented other settings as well,
i did notice and reported some perceived "relaxation" with 1.8.0 bios few days back (though i thought it might be placebo or wishful thinking),

[win32asmguy]

1 hour ago, MyPC8MyBrain said:

better late than never, well done Dell!
i wonder if they implemented other settings as well,
i did notice and reported some perceived "relaxation" with 1.8.0 bios few days back (though i thought it might be placebo or wishful thinking),

 

Indeed, glad to see this issue fixed even if there is no mention in the release notes. They technically also fixed the GT DC Loadline setting to be the proper 4.0 ohm setting, but that is more of an issue where load voltage for the iGPU is reported to be higher than the actual value ( ideally you want AC == DC )

 

I also tested Dynamic Boost 2.0 but did not see GPU TDP above 102W, or near 136W combined CPU+GPU under load. So that is still not working with current bios + ec + vbios + 517.66 dell driver.

[MyPC8MyBrain]

35 minutes ago, win32asmguy said:

tested Dynamic Boost 2.0 but did not see GPU TDP above 102W

since @PHVM_BR brought the subject to our attention I've been researching Dynamic Boost 2.0 bit deeper,
from what i learned and possible saw in person is a 5w Dynamic Boost 2.0 for our specific sku,

Dynamic Boost 2.0 doesn't mean it will boost to max TGP rather attempt a prefixed predefined boost value, that boost may be as low as 5w in our sku case, i haven't found a source to verify what exactly Dynamic Boost 2.0 is for our specific sku,

with heat dissipation being a big factor we all verified we can reach 120w under some circumstances, before i test locked my cpu to 30w i only saw short spikes from 115w to 120w during benching, when it was locked i could see 120w sustained (rails were listing 134.5w)

 

[MyPC8MyBrain]

i receive hopefully my final third replacement 7770 yesterday afternoon,

to my surprise it wasn't spitting lava out of the box as experienced before, even on ultimate performance in bios and windows it was happy idling at mid 60c,

decided to tune Dell's original image instead of wiping out and using my enterprise edition, got it to where im almost happy with it enough to get a baseline where performance are with this unit, getting 22500 in CB23 and a 11800 in 3DMARK Time Spy bench, this is with original Dell paste,
 

i have not yet cracked this unit open, with max 25700 for CB23 and 12177 for 3DMARK with LM TIM, i can surmise that there's overall 10% increase potential with LM TIM for the 7770 unit,

most LM TIM gains are from cpu side not much benefits for dgpu from what i can tell, it is prob better to keep dgpu with regular paste because that layer serves as heat shield protection when cpu is revving pumping temps up fast, i think it would be most efficient especially for the way both cpu and dgpu are setup inline on the heat pipe,

 

i am starting to think that the stiffer thermal application Dell uses serves another propose intentionally the compound get stiffer after application to help with bonding due to mobility and lack of full physical support,

[MyPC8MyBrain]

maybe some of you noticed in my videos a small batch file run before bench starts, 
it is a small script i wrote to help switching bios power plan and windows power plan,

with one click it will change from bios "Cool" and windows "Power Saver"

to bios "Ultra Performance" and in windows "Ultimate Performance",

and vice versa when clicked again, 

to have this working you will need to first disable "Modern Standby" 

[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power]
"PlatformAoAcOverride"=dword:00000000
"CsEnabled"=dword:00000000
[-HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\ModernSleep]
[HKEY_LOCAL_MACHINE\SYSTEM\CurrentControlSet\Control\Power\ModernSleep]

restore and activate the classic power plan schema,

install "Dell Command PowerShell Provider" 

Quote

Dell Command | PowerShell Provider is a module that makes BIOS configuration easily manageable through the Windows PowerShell interface.

 

finally paste the following script in *.bat and run 

mode con: cols=40 lines=4
@echo off
powercfg.exe -getactivescheme | findstr /i "saver" >nul && (powercfg -s 4591f705-2228-47e6-acef-44df2d359f8b & powershell -Command "& {Import-Module DellBIOSProvider ; cd DellSmbios:\PreEnabled ; si .\ThermalManagement 'UltraPerformance'}" & echo -= Ultimate Active =-)
if %errorlevel% equ 1 (powercfg -s 04afbba1-f17c-42c8-b76b-7dd8fd264253 & powershell -Command "& {Import-Module DellBIOSProvider ; cd DellSmbios:\PreEnabled ; si .\ThermalManagement 'Cool'}" & echo ** Power saver Active **) 
TIMEOUT 3 > nul

(if you get an error confirm power plan guid in script match your windows version (powercfg /L))
 

want to toggle "Hybrid Graphics" ON/OFF here's a script for that, 
paste the following script in *.bat and run to toggle Hybrid Graphics,

mode con: cols=40 lines=4
@echo off
powershell -Command "& {Import-Module DellBIOSProvider ; cd DellSmbios:\PreEnabled ; gi .\HybridGraphics|Format-Wide CurrentValue}" | findstr /i "Disabled" >nul && (powershell -Command "& {Import-Module DellBIOSProvider ; cd DellSmbios:\PreEnabled ; si .\HybridGraphics 'Enabled'}" & echo -= HybridGraphics Enabled =-)
if %errorlevel% equ 1 (powershell -Command "& {Import-Module DellBIOSProvider ; cd DellSmbios:\PreEnabled ; si .\HybridGraphics 'Disabled'}" & echo ** HybridGraphics Disabled **) 
TIMEOUT 3 > nul

 

using PowerShell we can now change most of bios settings from windows,
craft the right string and send it with PowerShell for instant mode changes,
below are the variable (as of bios rev 1.7.1) that can be accessed with their command and path options,

 

 

 

 

 

[Aaron44126]

3 hours ago, win32asmguy said:

I also tested Dynamic Boost 2.0

 

Is your system showing that Dynamic Boost 2.0 is available?  Mine is not.  (It should be an entirely separate entry in the NVIDIA "System Information" panel...  I've seen it in my Precision 7560.)

[win32asmguy]

33 minutes ago, Aaron44126 said:

 

Is your system showing that Dynamic Boost 2.0 is available?  Mine is not.  (It should be an entirely separate entry in the NVIDIA "System Information" panel...  I've seen it in my Precision 7560.)

 

Yes. Dynamic Boost is available and enabled. I also tested it with Nvidia PCF disabled in the device manager, which then limited the GPU to 90W. So it does appear to get 12W through dynamic boost.

[Aaron44126]

3 minutes ago, win32asmguy said:

Yes. Dynamic Boost is available and enabled. I also tested it with Nvidia PCF disabled in the device manager, which then limited the GPU to 90W. So it does appear to get 12W through dynamic boost.

 

Yeah, so I see "Dynamic Boost" showing enabled, but there should be a separate entry for "Dynamic Boost 2.0", which I am not seeing.

Spoiler