Dell Precision 7670 & Dell Precision 7770 owner's thread

[vooze]

33 minutes ago, MyPC8MyBrain said:

for whatever reason dell insists on shipping performance chassis with Intel UHD iGpu instead of the new beefier Intel Iris iGpu they ship with none performance chassis, you are likely running the same iGpu in your M6700 from previous generation, the new Iris should be able to get a more respectable result, i wanted to build my 7770 with Iris iGpu but dell straight up refused to let me have Iris chip instead of the antiquated UHD in the performance chassis,

i also requested to have my system build to have the 330w SFF power supply,
dell refused to sell me the 330w with my system config but informed i could by it on my own separately,
 

 

Is it even possible to get Iris / Intel Xe in Alder lake HX? I can't find anything on Google saying so. Only the 45W H models seem to feature that iGPU.

[MyPC8MyBrain]

9 minutes ago, vooze said:

Is it even possible to get Iris / Intel Xe in Alder lake HX

it appears so, clicky here

the above is just the first one i found, there are more listed in the second link below
there are other configs with the newer Iris installed in the 7670 and 7770 specific click here,
(this is again sales and marketing nonsense from both sides Intel and Dell)

[Aaron44126]

51 minutes ago, MyPC8MyBrain said:

i wanted to build my 7770 with Iris iGpu but dell straight up refused to let me have Iris chip instead of the antiquated UHD in the performance chassis,

 

The iGPU is determined by the CPU.  You'd have to complain to Intel about this.  They use the same iGPU in all Alder Lake HX CPUs.  (Well, aside from 12450HX which gets only half of the EUs compared to the rest.)

 

The iGPU in Alder Lake HX does use the same Xe graphics architecture used in the Alder Lake H CPUs.  It is just 32 EUs, compared to 64 or 96 in the Alder Lake H CPUs that they have labeled as "Iris" graphics.

[MyPC8MyBrain]

any 7770 owner here able to score over 12k on 3DMARK Time Spy Benchmark?

[DarginMahkum]

Maybe an Alienware 18 with Dell premium service is the way to go...

 

 

[SvenC]

On 12/5/2022 at 2:52 PM, Aaron44126 said:

Went and ran 3DMark Time Spy against the integrated Intel GPU.  The result that I got was a brilliant... 855.

 

 

What I was interested in here mostly is how well the iGPU compares to the best dGPU from ten years ago.  My Precision 7770 replaced a 10-year-old Precision M6700, which originally shipped with a Quadro K5000M GPU.  The K5000M (at stock clocks) was able to get 6.37 FPS on graphics test 1 and 5.60 FPS on graphics test 2.

 

So, the Alder Lake HX iGPU is sitting at just about exactly 75% performance compared to the 10-year-old Quadro K5000M.  Presumably, an Alder Lake H iGPU (with 3× the execution units compared to HX) would be able to beat it.

Wow. Isnt the hx line igpu less performant than the h line igpu? If yes, than h line igpu could beat your result? I have no real use for a dgpu, so seeing the igpus get so much better sounds great for my demands.

[Aaron44126]

31 minutes ago, SvenC said:

Wow. Isnt the hx line igpu less performant than the h line igpu? If yes, than h line igpu could beat your result? I have no real use for a dgpu, so seeing the igpus get so much better sounds great for my demands.

 

Yes, Alder Lake H has a more powerful iGPU than HX so it should be able to top my result.

[MyPC8MyBrain]

as some of you aware I have been evaluating the 7770 as a replacement for 7670,

I like the 7770 size, ergonomics, and form factor over the 7670 16/10 ratio and narrow stubby build, I do miss the nicer screen option, moving to actual performance the 7770 run even hotter than the 7670 idling out of the box between90-100c, not a good start, run the usual benchmarks with out of the box default settings, needless to say they were not good,

 

 

went about optimizing the OS environment removing bloatware, stopping/disabling services etc.,

this improved system performance to some extent but not what I was expecting, it was still thermal throttling badly,

 

 

I decided to re-paste both the cpu and gpu, after that I dropped roughly 15-20 degrees idle temp,

but throttling was still an issue no matter what I did including undervolting,

 

 

this wasn't good enough, I was still getting relatively poor result,

I decided to re-paste again with LM this time on both gpu and cpu, and so I did,

after that temps dropped roughly by 30c from initial idle temps,
I was finally idling at somewhat acceptable temps,

 

 

 

further digging around I realized how critical having control on cooling is for these systems, but it is non existing, there's no way to increase fan controls, the existing power plans are designed for older gen cpu's and can’t keep up with burst of power, cool in bios and power saver in windows gives the best result when it comes to idle temps, it’s cool and the cpu is so powerful that even in its lowest state it can handle decent daily loads without even breaking a sweat, the issue begins when I want to use dGpu, all power plan mainly block GPU power, accept for ultra-performance in bios plus ultimate performance in windows will allow to fully unleash dGpu full potential, that is when the system having tough time handling the heat and its almost instantly,

 

 

to conclude these system are actually amazing if you don't include a dGpu in the systems configuration, the cpu on its own can perform amazingly, this was not without painstaking hours of fine tuning and undervolting the systems cpu and the insightful help of Unclewebb (TS Author), despite achieving some respectable results, it was still thermal throttling (though rarely) even with dGpu not in the mix,

 

  

 

I did a small experiment where I disconnected the dGpu,

and sure enough the system run flawlessly and performed incredibly well with almost zero thermal issues, i have one more experiment to conduct but i have to wait for a component to arrive, i am curious to see if changing to 330w power supply will further iron out and boost the system further,

 

 

 

 

[PHVM_BR]

3 hours ago, MyPC8MyBrain said:

como alguns de vocês sabem, tenho avaliado o 7770 como substituto do 7670,

Eu gosto do tamanho, ergonomia e fator de forma do 7770 sobre a proporção 16/10 do 7670 e construção estreita e atarracada. 100c, não é um bom começo, execute os benchmarks usuais com configurações padrão prontas para uso, escusado será dizer que eles não eram bons,

 

 

optou por otimizar o ambiente do sistema operacional removendo bloatware, interrompendo/desativando serviços etc.,

isso melhorou o desempenho do sistema até certo ponto, mas não o que eu esperava, ainda estava com um afogamento térmico ruim,

 

 

Decidi colar novamente a CPU e a GPU, depois disso baixei cerca de 15 a 20 graus de temperatura ociosa,

mas a limitação ainda era um problema, não importa o que eu fizesse, incluindo undervolting,

 

 

isso não era bom o suficiente, eu ainda estava obtendo um resultado relativamente ruim,

Decidi colar novamente com LM desta vez tanto na gpu quanto na cpu, e assim fiz,

depois que as temperaturas caíram aproximadamente 30c em relação às temperaturas ociosas iniciais,
eu finalmente estava em marcha lenta em temperaturas aceitáveis,

 

 

 

pesquisando mais, percebi o quão crítico é ter controle sobre o resfriamento para esses sistemas, mas não existe, não há como aumentar os controles do ventilador, os planos de energia existentes são projetados para CPUs de geração mais antiga e não podem acompanhar a explosão de energia , legal na bios e economia de energia no windows dá o melhor resultado quando se trata de tempo ocioso, é legal e a cpu é tão poderosa que mesmo em seu estado mais baixo pode lidar com cargas diárias decentes sem nem suar, o problema começa quando Eu quero usar o dGpu, todo o plano de energia bloqueia principalmente o poder da GPU, aceita ultra-desempenho no BIOS e o melhor desempenho no Windows permitirá liberar todo o potencial do dGpu, ou seja, quando o sistema está tendo dificuldade em lidar com o calor e é quase instantaneamente ,

 

 

para concluir, esses sistemas são realmente incríveis se você não incluir um dGpu na configuração dos sistemas, a CPU por si só pode ter um desempenho incrível. TS Author), apesar de alcançar alguns resultados respeitáveis, ainda era um afogamento térmico (embora raramente), mesmo com dGpu fora do mix,

 

  

 

Fiz um pequeno experimento onde desconectei o dGpu,

e com certeza o sistema funcionou perfeitamente e teve um desempenho incrivelmente bom com quase zero problemas térmicos, tenho mais um experimento para conduzir, mas tenho que esperar a chegada de um componente, estou curioso para ver se a mudança para a fonte de alimentação de 330 W resolverá ainda mais e impulsionar o sistema ainda mais,

 

 

 

 

Your CPU results are fantastic.

What's the sustained power during the 30 minute test?

I imagine the temperature stays at 100°C in this longer test, right?

 

As for the GPU, how many Watts does the 3080Ti consume, average and peak, during a graphics benchmark?

Anyway, I believe 11500 on Time Spy is a fair score for this GPU with ~130W... If you limit the CPU you can still increase the graphics score and consequently increase the overall score.

[MyPC8MyBrain]

Thank you @PHVM_BR
to be fair i don't really have a base for comparison beside online statistics,

i never owned such a powerful mobile station, it is harder to compare this to previous generations, 

 

43 minutes ago, PHVM_BR said:

What's the sustained power during the 30 minute test?

during the test the cpu healed its 157w for quite some time, it settled somewhere around 70w deep into he run,
 

43 minutes ago, PHVM_BR said:

I imagine the temperature stays at 100°C in this longer test, right?

nope, it didn't reach 100c during the 30min run, one core reached 91c at max peak,

 

43 minutes ago, PHVM_BR said:

how many Watts does the 3080Ti consume, average and peak, during a graphics benchmark?

i was not focused on the 3080 wattage consumption,
during bench cycles I could tell the gpu and cpu were fighting over power availability, 
especially when the cpu was in its turbo cycle it was a fine line balancing the two when undervolting just on the merits of lack of power not even in thermal issues territory yet, the 3080 at any point didn't reach over 86c, mostly 74c,
 

43 minutes ago, PHVM_BR said:

11500 on Time Spy is a fair score for this GPU with ~130W...

that is the most disappointing part, i couldn't reach the missing 30% in performance left on the table,
i am quite happy with the cpu itself the 3080 is a big disappointment (any dGpu will be),
it is capped at 150w for the 7770 (the 7670 is capped at 125w),

from the little experiment i run i can confirm without a doubt the gpu is not happy living with the cpu in such tight space, the system performs folds better even when undervolting, i just disconnected the pci bridge and power to the dGPU, i didn't fully remove the pcb which i suspect if i did the result would be even better,

[PHVM_BR]

19 minutes ago, MyPC8MyBrain said:

during the test the cpu healed its 157w for quite some time, it settled somewhere around 70w deep into he run,
 

51 minutes ago, PHVM_BR said:

I imagine the temperature stays at 100°C in this longer test, right?

nope, it didn't reach 100c during the 30min run, one core reached 91c at max peak,

Can you change the PL1 for the CPU to stay above 70/80W?

If maintaining 157W for some time with the maximum of 91°C it would probably end the 30 minutes sustaining 110W (or more)...

 

20 minutes ago, MyPC8MyBrain said:

i was not focused on the 3080 wattage consumption,
during bench cycles I could tell the gpu and cpu were fighting over power availability, 
especially when the cpu was in its turbo cycle it was a fine line balancing the two when undervolting just on the merits of lack of power not even in thermal issues territory yet, the 3080 at any point didn't reach over 86c, mostly 74c,

At 86°C there is already thermal throttling. The temperature limit for using these GPUs is 87°C and at 86°C the clock drops.

 

27 minutes ago, MyPC8MyBrain said:

that is the most disappointing part of the test, i am quite happy with the cpu itself,
the gpu portion is a big disappointment,  it is capped at 150w for the 7770 (the 7670 was capped at 125w),

I believe this GPU should stay around 130W.

If it kept the 150W the rating would be higher...

[Aaron44126]

1 hour ago, PHVM_BR said:

I believe this GPU should stay around 130W.

If it kept the 150W the rating would be higher...

 

This was my experience.  It shows a 150W power limit in NVIDIA control panel, but practically it seems to be limited at 130W.  (HWiNFO64 shows it sitting right at 130W during a 3DMark run the whole time.)

[MyPC8MyBrain]

the GPU cannot ramp up at the same time as the cpu, if it has a none cpu dependent chores yes,
a task requiring both to run together at their peak is where issues begin,
they can do it each on their own but the juggling act of running together is very tall order for 240w power supply, together they demand more peak power than is available to the system and why i ordered a 330w power supply,

the cpu has a locked Power Limit 4 (PL4) value which tells the cpu to power itself down when overall consumption reaches the magic number of 242 (PL4 value should be 0 for the cpu to never ramp itself down), the cpu in its turbo cycle it will consume every bit of its 157w allotted which leaves 83w for the rest of the system including the 3080 that cannot use its full 150w power potential unless the cpu is slowed down somehow to free up power,
 

[Aaron44126]

4 hours ago, MyPC8MyBrain said:

i ordered a 330w power supply

...

the 3080 that cannot use its full 150w power potential unless the cpu is slowed down somehow to free up power,

 

Interested to see if the 330W power supply helps at all, but I doubt it.  Users have tried this with older 17" Precision systems and they would not utilize the extra power.

 

NVIDIA Dynamic Boost 2.0 is "supposed" to shuffle power allocation between the CPU and GPU when both are under load and strike a balance that gives best overall performance in the given situation.  Not sure if that is working well in these systems or not.  I noticed in the Precision 7X60 line that they enabled NVIDIA Dynamic Boost 2.0 in BIOS 1.8.x, and then removed it in a later BIOS update...

[MyPC8MyBrain]

2 hours ago, Aaron44126 said:

they would not utilize the extra power

interesting,
out of curiosity i plugged in a power meter to the wall jack and the power supply to the power meter,

run a 3DMark Time Spy bench, during the 2 initial tests the power usage was around 175w-180w,
during the 3rd and last phase of the test that suppose to be cpu focused stress test,

power usage spiked up over 235w few times,

 

i also noted that when on cool power plan and power saver in windows the draw from the wall is still between 40w-60w at idle, out of that HWInfo reporting cpu using only 5w-6w, when the unit goes to sleep power draw from the wall jumps between 1.8w-3.6w,

[PHVM_BR]

7 hours ago, MyPC8MyBrain said:

a GPU não pode acelerar ao mesmo tempo que a cpu, se não tiver tarefas dependentes da CPU sim,
uma tarefa que exige que ambas sejam executadas juntas em seu pico é onde os problemas começam,
cada um pode fazer isso por conta própria, mas o ato de malabarismo de funcionar juntos é uma tarefa muito difícil para fonte de alimentação de 240w, juntos eles exigem mais potência de pico do que está disponível para o sistema e por que pedi uma fonte de alimentação de 330w,

a cpu tem um valor bloqueado de limite de energia 4 (PL4) que diz à cpu para desligar quando o consumo geral atingir o número mágico de 242 (o valor PL4 deve ser 0 para a cpu nunca diminuir a velocidade), a cpu em seu turbo ciclo, ele consumirá cada bit de seu 157w alocado, o que deixa 83w para o resto do sistema, incluindo o 3080, que não pode usar todo o seu potencial de energia de 150w, a menos que a CPU seja desacelerada de alguma forma para liberar energia,
 

To verify and understand the behavior of how the system allocates power between CPU and GPU, run AIDA64 CPU stress test and Heaven or even Prime95 and Furmark simultaneously.

 

After a period the system should stabilize maintaining fixed values of power for both, or, as dell likes it, the power can vary, increasing and decreasing in cycles.

[pintie]

the power supply is rated 240W / 19,5V*12,31A if you use a powermeter on the Wall it's up to 300W and more.

i testet an old bigger Dell 240W power supply for a docking station - efficiency factor was about 0,82 at full load.

 

[Redrider]

On 11/4/2022 at 7:08 PM, MyPC8MyBrain said:

this is just painful to see, buy the 7770 from dell online outlet store for $2,409.00,
or buy the exact same config for 220% more (or 55% less) from dell online store for $5,336.04,

I've been following the outlet for a while. The 7770 has been removed from the search engine. Plenty of the 7760. 

[MyPC8MyBrain]

34 minutes ago, Redrider said:

The 7770 has been removed from the search engine.

the 7770 in the link i used must have sold by now, the 7770 in general is still listed click here
they been pushing lots of specials for outlet products with deeper discount then before the past two weeks,

[Redrider]

31 minutes ago, MyPC8MyBrain said:

the 7770 in the link i used must have sold by now, the 7770 in general is still listed click here
they been pushing lots of specials for outlet products with deeper discount then before the past two weeks,

Thanks , the old link I was using had them removed.

[MyPC8MyBrain]

my best result for the 7770 so far  

https://www.3dmark.com/spy/33349326

[PHVM_BR]

7 hours ago, MyPC8MyBrain said:

my best result for the 7770 so far  

https://www.3dmark.com/spy/33349326

Huge increase over the last best score!

What settings changes did you make?

[MyPC8MyBrain]

8 hours ago, PHVM_BR said:

What settings changes did you make?

NVidia cheater driver settings, which at this point I suspect majority of 3Dmark scores are saturated with,
my 7770 true score is 12100 so far https://www.3dmark.com/spy/33265268

[win32asmguy]

On 12/8/2022 at 6:06 AM, Aaron44126 said:

Interested to see if the 330W power supply helps at all, but I doubt it.  Users have tried this with older 17" Precision systems and they would not utilize the extra power.

 

NVIDIA Dynamic Boost 2.0 is "supposed" to shuffle power allocation between the CPU and GPU when both are under load and strike a balance that gives best overall performance in the given situation.  Not sure if that is working well in these systems or not.  I noticed in the Precision 7X60 line that they enabled NVIDIA Dynamic Boost 2.0 in BIOS 1.8.x, and then removed it in a later BIOS update...

 

I have a 330W brick from an Alienware 51M R2. While it does recognize and power the Precision 7770, it did not allow draw over 240W. In fact there were still situations where combined full load use was causing battery drain.

 

I would think that if the Dynamic Boost 2.0 limit is 150W then a synthetic load with CPU idle (<10W) and GPU 100% should be able to reach that value, but the highest I ever saw was 120W regardless of driver, bios and operating system combination.

[MyPC8MyBrain]

34 minutes ago, win32asmguy said:

it did not allow draw over 240W

the cpu still has a 242w power limit that has to be removed first to change its behavior,
see below in preparation i have "Power Limit 4" set to 0 (0 means no power restrictions),
if you haven't changed yours it should read 242, that is a hard cap on overall power,
plugging a 330w brick without changing this shouldn't change native cpu behavior, 
that 242 in PL4 field is the instructions to ramp down before wattage consumption reaches 242w,
you could set it to 330 if you feel safer with a power lock,

my brick is out for delivery and should arrive shortly, 

 

 

  

34 minutes ago, win32asmguy said:

the highest I ever saw was 120W

my cpu reach its 157w limit every chance it gets, if i run it in bench without gpu load it will run with 157w for quite some time before it starts ramping down, i suspect native setting might be limiting your cpu, power plan will be limiting power too, you need to be on ultimate power plan in windows and in bios on Ultra Performance to get the max out of cpu and gpu during bench testing,