Dell Precision 7670 & Dell Precision 7770 owner's thread

[jorgeregula]

2 hours ago, win32asmguy said:

By chance does anyone happen to know the sustained TDP for the 7670 A4500 GPU? From Notebookcheck's review I can see the 7670 A5500 is about 102W but wondering if the A4500 is the same.

 

Same for the A4500, 102W.

 

Anyone knows if there is anything to be done with Afterburner or Nvidia Inspector to improve performance within that power limit? The voltage curve in afterburner only starts at 1230 Mhz / 700 mV so there's not much room for undervolting I guess? Is there any way to reduce the idle power draw for the GPU?

 

On my 7670 (12850HX, a4500) Timespy result is 10059.

[Ionising_Radiation]

1 hour ago, jorgeregula said:

so there's not much room for undervolting I guess?

Essentially, yes. Desktop GPUs run at almost 3× the frequency, and the voltage is frequently much higher than necessary, allowing for plenty of room to undervolt, whereas notebooks already run under power and voltage constraints. 

[1610ftw]

17 hours ago, MyPC8MyBrain said:

my best 7770 3DMARK Time Spy score so far 12145, allot of tinkering to find the right undervolt balance,
what's interesting is with the right undervolt settings i am finally able to maintain 4,423 MHz average clock frequency throughout the test (with all my previous attempts i got good scores but cpu always ramped down to 2000Mhz), still getting slight thermal throttling but not enough to ramp the whole system down,
25.5k scores in CB23 are a regular occurrences now, trying to hit that illusive 26k CB23 score,
(id share my tests graphs and TS settings but i am not allowed to upload anymore images )

https://www.3dmark.com/3dm/85032368

(with LM TIM application on both cpu and dgpu)
 

TS > FIVR:
CPU Core           -146.5 (-150 effective) with  IccMax 185.00
CPU P Cache     -73.2 (-75 effective)
Cache Ratio       > Min/Max value 8/36
Turbo Groups     > all performance cores set to 47
(everything else left to its default settings)
 

MSI Afterburner Settings
Core Clock (MHz) +225
Memory Clock(MHz) +1400

 

Way to go, looks like with liquid metal you can get some added performance out of it!

The CPX17 can do almost 14K with a bit of Afterburner Tweaking so it is a shame Dell is leaving so much power on the table by setting the power limit so low.

 

Still an excellent improvement over stock and it would be cool to know how much your 7770 does pull during TimeSpy with the slim and the fat power supply and how much of a difference it makes in scores.

 

Have you checked how much power you can sustain with your CPU now that you have applied liquid metal?

I am pretty sure that it will be at least 120W as otherwise I doubt that you would make it to a 25.5K CR23 score.

 

 

 

 

[MyPC8MyBrain]

1 hour ago, 1610ftw said:

it would be cool to know how much your 7770 does pull during TimeSpy with the slim and the fat power supply and how much of a difference it makes in scores

the system pulls noticeably allot more power from the wall with the 330 brick,
280w is the average sustained pull form the wall during Time Spy bench,
with the slim brick it pulling a max of 235w but averaged much lower 180w sustained,

 

1 hour ago, 1610ftw said:

how much of a difference it makes in scores

difference in scores is 0

 

1 hour ago, 1610ftw said:

Have you checked how much power you can sustain with your CPU now that you have applied liquid metal?

during CB23 bench with 25.5k scores if i recall correctly it wont even reach PL1 limit, it utilized 81w max for the whole run, it wont thermal throttle either unless i hammer it with 4-5 consecutive runs without cooling for a minute, even fans don't ramp up through the run,
 

i haven't tested my current config with my special sauce yet,
i believe adding that will get me over the hump to the 26k score,
 
 

[win32asmguy]

9 hours ago, jorgeregula said:

 

Same for the A4500, 102W.

 

Anyone knows if there is anything to be done with Afterburner or Nvidia Inspector to improve performance within that power limit? The voltage curve in afterburner only starts at 1230 Mhz / 700 mV so there's not much room for undervolting I guess? Is there any way to reduce the idle power draw for the GPU?

 

On my 7670 (12850HX, a4500) Timespy result is 10059.

 

Thank you for the information. It is interesting that the A4500, A5500 and 3080Ti are all around the same power limit and performance levels.

 

Does afterburner allow manipulation of the voltage curve, overclocking, or power limit with any drivers? In that past I thought it was completed locked.

 

If the setup works for you, using the Intel GPU to drive all displays definitely can reduce idle power draw. The Nvidia GPU will also be in a different higher power state depending on if you have many displays connected or if the displays are set to a higher refresh rate. Obviously be sure to have IA AC LoadLine set correctly if it is not otherwise CPU voltage will be significantly overcompensated.

[Aaron44126]

17 minutes ago, win32asmguy said:

Thank you for the information. It is interesting that the A4500, A5500 and 3080Ti are all around the same power limit and performance levels.

 

Does afterburner allow manipulation of the voltage curve, overclocking, or power limit with any drivers? In that past I thought it was completed locked.

 

I feel like 4000/5000 cards have had the same power limit (& roughly the same performance) going all of the way back to the Pascal cards in the Precision 7X30 line (2018), which I think was the first time that the 4000-level card hit 100W.  Since then, there's been no reason to get the 5000 card, really, unless you need other capabilities that it provides.  (Until this generation, they had different vRAM amounts at least...)  Benchmarks have shown the performance difference otherwise to be pretty consistently less than 5% off.

 

This is reflective of the "low" power limits that NVIDIA is putting on these power-hungry desktop GPU dies that they are shoving into laptops.  Because they have the same power limit, they have the same performance, with the higher-end GPUs getting a very slight nudge up because they are able to run more cores at a lower clock speed (a little bit more efficient).

 

Actually, I sort of suspect that the A3000 GPU would be competitive against the top-end cards as well, at least if you are not pushing vRAM above 12GB.  NVIDIA says it supports up to 130W TGP so it should have the same power limit as the A4500 and A5500 in these systems (or at least in the Precision 7670).  I'd love to see some benchmarks on this.

 

Regarding undervolting, previously the pro/Quadro/RTX line has been locked down, but starting last year with the pro Ampere line, NVIDIA seems to be allowing undervolting in the pro GPUs using tools like MSI Afterburner.  (I have not tried it yet, but other users did report being able to undervolt the GPUs in the Precision 7X60.)

[jorgeregula]

15 hours ago, MyPC8MyBrain said:

the system pulls noticeably allot more power from the wall with the 330 brick,
280w is the average sustained pull form the wall during Time Spy bench,
with the slim brick it pulling a max of 235w but averaged much lower 180w sustained,

 

difference in scores is 0

 

I would also be interested in seeing if there is any difference in scores when connected to the WD19DCS dock. The device clearly shows in the BIOS it is then connected to a 210 W power supply, instead of a 240 W.

 

Perhaps the battery takes up the slack and provides some extra juice?

[Aaron44126]

1 hour ago, jorgeregula said:

I would also be interested in seeing if there is any difference in scores when connected to the WD19DCS dock. The device clearly shows in the BIOS it is then connected to a 210 W power supply, instead of a 240 W.

 

Perhaps the battery takes up the slack and provides some extra juice?

On past systems, performance was slightly reduced when using dock power only. You can avoid it by attaching a 240W power adapter directly to the system in addition to the dock cable.

[MyPC8MyBrain]

noticed some weird hardware error issues recently,
not sure if they were there the whole time or started recently,
i tried to eliminate variables to try and narrow this down but i have no lead atm,

it is not caused by the under or overvolting, its not a driver compatibility issues (went back to original OEM image),
seems to be related to the dgpu but no additional info can be found even system logs are very vague,

it wont happen during CB23 run nor when running Valley bench,

but if i open Afterburner or 3DMARK just to the main interface not even running tests yet it will trigger these hardware errors instantly,

can someone please help validate if this is just on my system?
 

 

 

 

 

 

 

[Aaron44126]

BIOS update 1.8.0.

https://www.dell.com/support/home/en-us/drivers/driversdetails?driverid=JG0V9

- Firmware updates to address security vulnerabilities including (Common Vulnerabilities and Exposures - CVE) such as CVE-2022-3365 and CVE-2022-30339.

 

—————

 

45 minutes ago, MyPC8MyBrain said:

can someone please help validate if this is just on my system?

 

I am also seeing WHEA events.  They seem to come in "bursts".  The most recent I have is a batch of 15 which occurred within 20 seconds of each other, about two days ago.  They don't occur often.  I see a few bursts on December 13 and 14, but then I have a break all of the way back to December 2.  (My system is powered on 24/7.)

 

I haven't noticed any actual problems timed with these events, so I'm just going to ignore it.

[MyPC8MyBrain]

maybe its just placebo...
after BIOS 1.8.0 update my system seems to idle 3-4 degrees cooler (package is usually 41c at idle),

[MyPC8MyBrain]

here are the graphs for my best score mentioned earlier, note cpu is holding 4.7Mhz throughout the run, 
this is very different from previous runs with early cpu undervolt and gpu overclock setting, with all my previous settings that purple line would dip to around 1.7MHz almost instantly (yes this is replicable),
with 30w withheld from 150w = 120w which what we all able to get in effect,
25% performance missing from an already capped 175w dGpu to 150w we only get 120w from,
if Dell allowed us to regain the full limited 150w i bet we could close the gap we currently see,

 

 

  

https://www.3dmark.com/3dm/85032368

[win32asmguy]

4 hours ago, MyPC8MyBrain said:

here are the graphs for my best score mentioned earlier, note cpu is holding 4.7Mhz throughout the run, 
this is very different from previous runs with early cpu undervolt and gpu overclock setting, with all my previous settings that purple line would dip to around 1.7MHz almost instantly (yes this is replicable),
with 30w withheld from 150w = 120w which what we all able to get in effect,
25% performance missing from an already capped 175w dGpu to 150w we only get 120w from,
if Dell allowed us to regain the full limited 150w i bet we could close the gap we currently see,

 

Have you tried flashing Aaron's 3080Ti vbios to your machine? IIRC his was delivered pretty early on so it may have different base TDP or DB2.0 configuration settings. 130W pull should at least boost your score up a bit even if it cannot reach 150W.

[Aaron44126]

Huh, if there is a different vBIOS version floating around, @MyPC8MyBrain can you report what vBIOS version number you are seeing?

[MyPC8MyBrain]

i should still have a copy of Aaron's vbios, it wouldn't flash the old 7670,
my current 7770 vbios is 94.03.19.00.EE

 

[Aaron44126]

11 minutes ago, MyPC8MyBrain said:

my current 7770 vbios is 94.03.19.00.EE

That’s the same as mine. I’ll have to double check what power limit I am observing. Pretty sure I was seeing 130W when this system was new.

[MyPC8MyBrain]

i made bit of a mess with the info i submitted earlier,
here is live capture with info present live so there's no confusion,
not trying to break records just to capture power usage during runs

 

with 240w brick

CB23 Run

 

3DMARK Time Spy Run

 

 

330w brick

 

CB23 330w Run

 

3DMARK Time Spy Run 330w

 

AIDA64 Run 330w (290w+)

 

 

[DarginMahkum]

Could drawing considerably over the specified ratings damage the hardware? I remember someone tweaking an Alienware 17 R2 said that the power rails of his laptop were damaged.

[PHVM_BR]

On 12/16/2022 at 11:54 AM, MyPC8MyBrain said:

here are the graphs for my best score mentioned earlier, note cpu is holding 4.7Mhz throughout the run, 
this is very different from previous runs with early cpu undervolt and gpu overclock setting, with all my previous settings that purple line would dip to around 1.7MHz almost instantly (yes this is replicable),
with 30w withheld from 150w = 120w which what we all able to get in effect,
25% performance missing from an already capped 175w dGpu to 150w we only get 120w from,
if Dell allowed us to regain the full limited 150w i bet we could close the gap we currently see,

 

 

The processor clock is fixed at 4.7GHz during graphics tests 1 and 2 where CPU load is low. In the processor test there is a slight drop in the clock rate.

GPU clock is dropping considerably in graphics test 2.

My hypothesis:

You've set the PL1 to 85W and checked the MMIO box blocking dynamic power cap adjustments and this makes the processor power work at a very high cap during TimeSpy. This means that Dynamic Boost has no "room" to kick in and the GPU is limited to base TGP (maybe 125W???).

If you limit the CPU power during the benchmark to somewhere between 45 and 55W, you will likely see an increase in the 3080Ti's power consumption, increasing its clock rate. With this your global score will increase, as timespy gives much more weight to the graphic score.

[MyPC8MyBrain]

2 hours ago, PHVM_BR said:

My hypothesis:

the cpu is at 4.7MHz because that's the highest its allowed to ramp up (all cores locked via TS at 4.7), 
power plan aka Ultimate Performance places cpu to its highest state until PL2 rules are broken and then cpu is placed within PL1 power limits followed by reducing MHz to reduce thermals, the graph is showing that the cpu was able to run through Time Spy bench without breaking PL2 rules which would have placed it in PL1 limits,
the unit is essentially maintaining manageable PL2 terms during the bench with undervolt,

[PHVM_BR]

44 minutes ago, MyPC8MyBrain said:

the cpu is at 4.7MHz because that's the highest its allowed to ramp up (all cores locked via TS at 4.7), 
power plan aka Ultimate Performance places cpu to its highest state until PL2 rules are broken and then cpu is placed within PL1 power limits followed by reducing MHz to reduce thermals, the graph is showing that the cpu was able to run through Time Spy bench without breaking PL2 rules which would have placed it in PL1 limits,
the unit is essentially maintaining manageable PL2 terms during the bench with undervolt,

Exactly, it's at 4.7GHz due to the high-performance power plan, but it's only holding that plateau because the load during the graphics tests is low.

 

In the CPU test the load on the processor is much higher.

 

Check, run Time Spy on one screen and leave HWiNFO open on another.

 

Or do you believe that this Precision (even with liquid metal) can simultaneously maintain 157W from the CPU plus 120W from the GPU?

[Aaron44126]

4 minutes ago, PHVM_BR said:

Or do you believe that this Precision (even with liquid metal) can simultaneously maintain 157W from the CPU plus 120W from the GPU?

 

It doesn't necessarily run the CPU at maximum power just because it's running at the maximum clock speed.  (...Though, it would use less power if it were allowed to clock down.)

[PHVM_BR]

22 minutes ago, Aaron44126 said:

 

It doesn't necessarily run the CPU at maximum power just because it's running at the maximum clock speed.  (...Though, it would use less power if it were allowed to clock down.)

But that's exactly what I'm saying.

 

Clock is fixed at the maximum during graphics tests 1 and 2 because the load on the cpu is low.

[MyPC8MyBrain]

locking PL1/2 to 50w will not set the cpu to lower MHz, few cores can still go as high as they can within that 50w limit, the only way is locking cores manually, doing so will defeat the premise of your theory limiting the systems full potential, if thermals will allow power plan rules will place cpu cores at highest regardless,
 

59 minutes ago, PHVM_BR said:

can simultaneously maintain 157W from the CPU plus 120W from the GPU?

it doesn't need too, 157w happens when all cores utilized to the max, they usually alternate with only few holding high MHz until thermals saturate the system and it starts ramping down,

 

during Time Spy runs i have seen both cpu reaching its 157w PL2 limit and dgpu using max 120w (with HWiNfo), you can observe wattage draw during different segments of Time Spay right off the wall in the videos i posted few posts back, i also seen the system push almost 300w in AIDA64 bench (with 330w brick),

[PHVM_BR]

53 minutes ago, MyPC8MyBrain said:

locking PL1/2 to 50w will not set the cpu to lower MHz, few cores can still go as high as they can within that 50w limit

What I said about blocking the PL1 at 45/55W has nothing to do with lowering MHz.

 

This is to leave room for Dynamic Boost to increase GPU power, increasing the graphics score.

 

You probably need to also limit the PL2 as your liquid metal system must try to maintain 157W during the entire PL2 actuation period and this will not allow the dynamic Boost to kick in

 

Obviously you will have a loss in cpu score but a gain in graphics score.

 

 

 

53 minutes ago, MyPC8MyBrain said:

during Time Spy runs i have seen both cpu reaching its 157w PL2 limit and dgpu using max 120w (with HWiNfo), you can observe wattage draw during different segments of Time Spay right off the wall in the videos i posted few posts back,

Not simultaneously!

 

Only in the cpu test the 157W must be reached because in this test the load on the cpu is high and on the GPU it is lower.

 

According to your video, during cpu test the consumption of the whole system is ~225W.

In the graphics tests this consumption is around 190W.

 

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.