Dell Precision 7670 & Dell Precision 7770 owner's thread

[Ionising_Radiation]

3 hours ago, Aaron44126 said:

it's a ≈33% performance improvement over stock

Holy crap, that's a lot of performance left on the table...

[operator]

On 9/7/2022 at 12:00 AM, win32asmguy said:

I replaced the thermal pads today on the CPU side with some mid grade 12.8W/mK and now performance has jumped over 15%

 

Would it also make sense to do this on the GPU side also?

[Aaron44126]

55 minutes ago, operator said:

Would it also make sense to do this on the GPU side also?

 

The GPU is not close to thermal throttling before hitting the power limit, so there's not a pressing need to make it cool better.  Undervolting it could eek out some more performance.

[operator]

46 minutes ago, Aaron44126 said:

 

The GPU is not close to thermal throttling before hitting the power limit, so there's not a pressing need to make it cool better.  Undervolting it could eek out some more performance.

 

Understood. Just thought when you do it for CPU side, why not also doing it for GPU side the same time.

[-batab-]

On 9/8/2022 at 9:53 PM, TwistedAndy said:

 

Those results were received during the first run with Turbo Boost enabled. 150W is a huge number. I think, the noise level will be above 50 dBA and incompatible with productive work. Mobile Alder Lake 6+8 and 8+8 CPUs are most effective on the 45-75W range or nearly 14500 - 18000 CB R23 MT with slight undervolting. In terms of noise it will be something between 38 and 45 dBA accordingly.

 

Yes, you can go beyond 80W and get slightly more performance, but the fan noise...

 

If that's the typical use case why not simply opt for a 6900hx which is way more efficient than Alder lake when below 30W and has close performance in the 40-70W range. 

 

If someone plans to not ever exceed the 70-80W for cpu only loads you basically get better thermals, noise and battery life at a lower price if you just pick up an amd laptop. 

 

Or am I missing something? 

[win32asmguy]

1 hour ago, operator said:

Understood. Just thought when you do it for CPU side, why not also doing it for GPU side the same time.

I mostly only did the CPU side because of these reasons:

• Ran out of time and wanted to just get it reassembled
• Lack of proper tools (really need plastic tweezers and scissors / cutting knife and mat to properly create the custom pads)
• No good resource of guides (tried looking back into NBR archives for that, or youtube)
• Issues with eyesight and lighting

I think going back and doing the GPU side is a good idea as Nvidia has a sliding clockspeed scale depending on temperatures, so a core at 60C will benchmark faster than a core at 75C even if its not technically thermally throttling.

 

If I remember correctly HID Evolution used to offer CTO Precisions through their repaste / repad services. It could be quite valuable if a turn key "excellent" job is required.

[Aaron44126]

2 hours ago, operator said:

Understood. Just thought when you do it for CPU side, why not also doing it for GPU side the same time.

 

8 minutes ago, win32asmguy said:

I think going back and doing the GPU side is a good idea as Nvidia has a sliding clockspeed scale depending on temperatures, so a core at 60C will benchmark faster than a core at 75C even if its not technically thermally throttling.

 

I did repaste the GPU core when I was doing the CPU (...sort of required any time you take the unified heatsink off).  I actually never checked 3DMark to see if performance actually improved; I'll do that soon.  When I get around to repadding, I'll look at swapping out the GPU pads as well, but that would be more for the vRAM and VRMs and not for the core.  Maybe it will make a difference...

 

54 minutes ago, -batab- said:

Or am I missing something? 

 

...Just a decent selection of AMD-based workstation-class laptops...

[1610ftw]

9 hours ago, Aaron44126 said:

 

@win32asmguy has managed 20K in one loop / 18.5K sustained but it took some work, both physical (repasting & new thermal pads for VRMs) and software/firmware (IA AC/DC loadline, power limits, voltage levels).

https://notebooktalk.net/topic/632-dell-precision-7670-dell-precision-7770-owners-thread/?do=findComment&comment=13192

 

You can see progress by looking at his posts on page 5 of this thread.

 

Sort of a shame as he mentioned that the score was achieved without undervolting, so basically it's a ≈33% performance improvement over stock (for extended multi-threaded load) and it's all things that Dell could implement for all systems coming out of the factory (using higher quality thermal materials between heatsink & CPU/mobo components, and setting some firmware variables)...

 

I have not had time to dig in any deeper myself.  Given the "issues" with these systems, I'm thinking about re-framing the "review" that I plan to produce as more of a "how to get better performance" guide but finding time to work on it is another story...

 

Thanks, I guess I wasn't sufficiently up to date!

That is the performance i would have expected from the start and it represents a ca. 50% multicore increase over the 7760 which is very impressive!

[operator]

4 hours ago, win32asmguy said:

to properly create the custom pads

 

Maybe a dumb question.... as this material does not conduct electricity, why couldn't you just cut one long stripe that covers all "chips" in a row? Why do they have to be properly made with the correct dimensions of each single "chip"?

[Aaron44126]

26 minutes ago, operator said:

Maybe a dumb question.... as this material does not conduct electricity, why couldn't you just cut one long stripe that covers all "chips" in a row? Why do they have to be properly made with the correct dimensions of each single "chip"?

 

Even combining some, there are a fair amount of pads to cut out.  Check out the bottom part of the heatsink:

Spoiler

 

[SilverAzide]

Just curious if the experts on this forum have any observations on this particular 7770?  It's stock and untweaked in any way.  Results seem pretty similar to the OPs machine?  It's got the RTX 4500, if that matters.

 

The core temps are kind of concerning, especially since they hit the max really easily and quickly, but I don't know much about how recent CPUs are supposed to behave.  My previous machines were a 6600, 6700, and my current daily driver is a 6800.  They almost never hit the max CPU temps, even with all cores pegged at 100% for hours at a time.

 

Also, HWiNFO doesn't show any fans.  This seems typical for Dell, as Martin (the HWiNFO dev) hasn't found the secret to get this info from recent machines.  Just checking if others see the same thing?

 

 

[-batab-]

9 hours ago, Aaron44126 said:

 

 

I did repaste the GPU core when I was doing the CPU (...sort of required any time you take the unified heatsink off).  I actually never checked 3DMark to see if performance actually improved; I'll do that soon.  When I get around to repadding, I'll look at swapping out the GPU pads as well, but that would be more for the vRAM and VRMs and not for the core.  Maybe it will make a difference...

 

 

...Just a decent selection of AMD-based workstation-class laptops...

 

Yeah, fair enough. I was stupidly thinking of any high level laptop including gaming version and not workstation-class only. 

 

So let's make it a different question:

 

If a 6900hx workstation-class was actually available, would it make more sense over the 12900hx options? 

 

I'm always attracted by raw power no matter what and lean towards Intel setups but I wonder if >80W load cases arejust so unbearable (fan noise and whatever) that AMD is simply superior. 

 

Feels like many people, me included, just lean towards Intel setups even when amd is available but I have no clue if I have a real reason for my preference or not. Current laptop TDPs are just getting out of hand and maybe it's just something I shouldn't consider. 

 

Actually, does anyone know if Dell has any plans of introducing amd-based versions? 

[Aaron44126]

7 hours ago, SilverAzide said:

Also, HWiNFO doesn't show any fans.  This seems typical for Dell, as Martin (the HWiNFO dev) hasn't found the secret to get this info from recent machines.  Just checking if others see the same thing?

I posted the “secret” and they don’t seem to have made use of it…

AFAIK, Dell Fan Management is the only tool that can show real-time fan measurements for this system.

 

https://www.hwinfo.com/forum/threads/new-dell-systems-ec-sensor-values-not-available-fix.7482/

 

Anyway, unlike older systems which could run at max turbo indefinitely…. for a few years now it has been normal for these CPUs to hit 100C and thermal throttle at max load.  The turbo boost speed is high enough that you can expect thermal throttling.  The question is rather what wattage/clock speed can you achieve at that temperature?  Improving the thermals should push both of those higher.

 

To keep the fans under control, I keep turbo boost off unless I need it.  I have an article about this linked in my signature.

[1610ftw]

On 9/6/2022 at 2:03 PM, Easa said:

Back to relevant topic, I personally think (and also heard insider rumours) that the 15.6" and 17.3" form factors will eventually cease to exist, with 16" replacing them both, and we might just see a new 18-19" 16:10 form factor emerging which, due to the thin screen borders and aspect ratio, will be similar in outline dimensions to older 17.3". 

 

18 or even 19" 16:10 would be great for workstations. it would also help with necessary real estate for cooling, memory and storage. Did the insider mention a bigger screen size or only that Dell would phase out the 17.3 and 15.6 form factors?

 

In any case in a machine with a bigger screen there would also be enough space for a 2.5" form factor data drive(s) and indeed it would be nice if it could have a multi purpose slot that alternatively could take a second battery but if the battery would be meant to have any capacity it would probably leave enough space for two 2.5" SSDs or one of those NVME monsters with a 2.5" form factor - I think they go up to 30TB now...

 

It would also be great if via that slot or by some other means there would be some kind of connection that allows to use an external GPU without bandwidth restrictions and without having to use some kind of home-made solutions like the ones where the GPU is connected via an NVME SSD slot.

[Aaron44126]

"Speculation time"

 

7 hours ago, -batab- said:

If a 6900hx workstation-class was actually available, would it make more sense over the 12900hx options? 

 

I'm always attracted by raw power no matter what and lean towards Intel setups but I wonder if >80W load cases arejust so unbearable (fan noise and whatever) that AMD is simply superior.

 

6900HX doesn't seem to be able to top 12900HX in terms of top performance.  If the high TDP of the Alder Lake HX CPUs is a concern, there are options to just limit the top power use which will bring fan speeds and system temperature down.  This is why I normally run with turbo boost off, unless I need extreme CPU power.  You can also just use Dell's "Quiet" thermal profile which will cut power limits on the CPU and GPU.

 

7 hours ago, -batab- said:

Actually, does anyone know if Dell has any plans of introducing amd-based versions?

 

I've asked about this before.  Dell generally won't discuss future product plans but...  I rather doubt it.  I think there would be concern that they'd put in all of the engineering effort to offer an AMD option, and in the end, >90% of the sales would still be for the Intel version, so there's no point really.  That is, unless AMD somehow manages to get ahead of Intel in terms of top performance, then maybe people would be willing to give it a closer look.

 

I've heard some users mention that Intel is just better for some types of things, like heavy use of VMs.  AMD "works" but support is just not up to the same level.

 

...There's definitely benefit to having AMD around providing stiff competition, even if they aren't an option for the workstation laptops right now.  For instance, if Ryzen hadn't have shown up as a serious contender, we may not have seen Intel bump up the core count in 8th/9th-gen Coffee Lake CPUs.

 

One other thing; there is some real benefit to having separate performance and efficiency cores.  Although AMD is poo-pooing Intel's offering right now, I fully expect them to have a similar setup in their CPUs in the next 2-3 years.

 

2 hours ago, 1610ftw said:

Did the insider mention a bigger screen size or only that Dell would phase out the 17.3 and 15.6 form factors?

 

At a recent "deep dive" presentation on the Precision mobile workstation lineup, Dell reps committed to keeping the 17-inch form factor around for the foreseeable future.  (They would already be in the design phase of at least the next two generations of mobile Precisions, so 17-inch "7780" and "7790" seem likely to me.)  Based on the industry trend of shrinking workstations generation-over-generation, I rather doubt that we'll see a 18"-19" workstation, or 2.5" drives in a mobile workstation (at least from Dell).

 

Now, if something like that did show up, I'd definitely be interested...  So I'd love to be wrong here.  However, I think that there wouldn't be that much of a market for such a system.

[1610ftw]

1 hour ago, Aaron44126 said:

At a recent "deep dive" presentation on the Precision mobile workstation lineup, Dell reps committed to keeping the 17-inch form factor around for the foreseeable future.  (They would already be in the design phase of at least the next two generations of mobile Precisions, so 17-inch "7780" and "7790" seem likely to me.)  Based on the industry trend of shrinking workstations generation-over-generation, I rather doubt that we'll see a 18"-19" workstation, or 2.5" drives in a mobile workstation (at least from Dell).

 

Now, if something like that did show up, I'd definitely be interested...  So I'd love to be wrong here.  However, I think that there wouldn't be that much of a market for such a system.

 

I checked my 7760 and it would hold an 18" 16:10 screen with slighty slimmer bezels which should be entirely doable so they would not even have to increase the current form factor for that. I can see 18" 16:10 as some kind of common denominator as it would mean that the currently accepted biggest (least tiny) form factor wouldn't have to grow even bigger.

 

It also allows for nice clean steps: 14, 16, 18" screens and maybe 20" at the top? OK, fat chance of 20" happening but 18" would be a good start as with a 16:10 screen it would mean almost 15% more screen real estate in the same form factor.

As 16" 16:10 is just about the same height as a 17.3" 16:9 so it would also make more sense from that vantage point. Having a screen that is wider and taller is certainly more attractive than having a screen that is only wider, especially for a workstation.

 

 

[Aaron44126]

I can see them moving the 17.3" system to 16:10 by keeping the same screen width and increasing the height, basically...  They could even remove the "Dell" logo beneath the screen and have some extra space to work with at the bottom (like they have done with the XPS / Precision 5000 systems).  And maybe you could call that 18".

 

...Though, if the size that they need is not one that is becoming standard in the industry, I sort of doubt that Dell would go through the effort and expense of contracting some screen manufacturer to make custom-sized displays just for this system.

[1610ftw]

7 hours ago, -batab- said:

 

Yeah, fair enough. I was stupidly thinking of any high level laptop including gaming version and not workstation-class only. 

 

So let's make it a different question:

 

If a 6900hx workstation-class was actually available, would it make more sense over the 12900hx options? 

 

I'm always attracted by raw power no matter what and lean towards Intel setups but I wonder if >80W load cases arejust so unbearable (fan noise and whatever) that AMD is simply superior. 

 

Feels like many people, me included, just lean towards Intel setups even when amd is available but I have no clue if I have a real reason for my preference or not. Current laptop TDPs are just getting out of hand and maybe it's just something I shouldn't consider. 

 

Actually, does anyone know if Dell has any plans of introducing amd-based versions? 

 

The solution is to undervolt and limit multipliers. Then you have 14 or 16 cores with Intel instead of just 8 with AMD and that will make a difference for multitasking. I also want support for virtual machines, Thunderbolt and the best compatibility in other areas so unless AMD tempts me with a 16 core unlocked beast that at least offers Thunderbolt 4 I am unlikely to switch over and even then it will probably get released at a time when Intel releases their next gen HX processors with 8 + 16 cores and those will be hard to beat.

[-batab-]

2 hours ago, 1610ftw said:

 

The solution is to undervolt and limit multipliers. Then you have 14 or 16 cores with Intel instead of just 8 with AMD and that will make a difference for multitasking. I also want support for virtual machines, Thunderbolt and the best compatibility in other areas so unless AMD tempts me with a 16 core unlocked beast that at least offers Thunderbolt 4 I am unlikely to switch over and even then it will probably get released at a time when Intel releases their next gen HX processors with 8 + 16 cores and those will be hard to beat.

 

I completely agree as I lean towards Intel. However, no matter the undervolt/optimization: it seems like under 30-40W AMD just wins in terms of efficiency.

 

This is also completely reasonable if we think that Intel scales nicely from 40 to 120W where AMD is super bad. It is quite legit that AMD shows more scaling margin in the idle-browsing TDP levels (5-30W). And this is also the reason why there is no competition in battery life usually. Curiously, Intel also looks more efficient in single core turbo boosted tasks as AMD probably pushed a bit over the limit to compensate for the low single core performance.

 

There are a few reviews by jarrod and some others that actually show nice scaling charts (e.g. image below)

[Aaron44126]

59 minutes ago, -batab- said:

I completely agree as I lean towards Intel. However, no matter the undervolt/optimization: it seems like under 30-40W AMD just wins in terms of efficiency.

 

I wonder if that holds true for the E cores?  The Alder Lake E cores are more efficient at low power levels and the P cores are more efficient at high power levels.  I haven't seen much specifically looking at E cores performance & efficiency because most people doing benchmarks are just interested in the biggest number you can get.  (Getting the OS to properly distribute processes when you want to save power could be a challenge as well.)

[1610ftw]

6 minutes ago, Aaron44126 said:

I can see them moving the 17.3" system to 16:10 by keeping the same screen width and increasing the height, basically...  They could even remove the "Dell" logo beneath the screen and have some extra space to work with at the bottom (like they have done with the XPS / Precision 5000 systems).  And maybe you could call that 18".

 

...Though, if the size that they need is not one that is becoming standard in the industry, I sort of doubt that Dell would go through the effort and expense of contracting some screen manufacturer to make custom-sized displays just for this system.

 

As I said 18" would fit within the dimensions of the bezel - the Dell logo would have to go though but I guess that is OK as they already did that in the XPS/Precision 5000 line. The bezels naturally would have to get slimmer but Dell already does that in the XPS / Precision 5000 systems and Apple has done that, too.

 

When that works out HP, Lenovo, Asus, MSI can get the screen, too and in any case I doubt that it would be much more of an effort than the new memory they implemented - they only have to convince one manufacturer to build a screen for them and Asus has shown that it can be done even for the GX800 that probably sold less than 500 units worldwide and the big Dell workstations certainly will sell in higher numbers than that.

[win32asmguy]

13 minutes ago, Aaron44126 said:

I wonder if that holds true for the E cores?  The Alder Lake E cores are more efficient at low power levels and the P cores are more efficient at high power levels.  I haven't seen much specifically looking at E cores performance & efficiency because most people doing benchmarks are just interested in the biggest number you can get.  (Getting the OS to properly distribute processes when you want to save power could be a challenge as well.)

 

Maybe this is helpful?

 

https://www.techpowerup.com/review/intel-core-i9-12900k-e-cores-only-performance/7.html

 

Granted, the e-cores are subject to the same silicon lottery as P-cores so an excellent or poor sample could inaccurately represent what to expect. They are probably not as efficient at 3.9ghz in the 12900k as they are at 3.4ghz as well, so maybe more testing would be good using a system that can freely adjust the multiplier range.

 

That being said I think the battery life is pretty good on the Precision so long as you take steps to ensure the Nvidia GPU is not activated unless it is needed. You can create a power mode for battery with P-core turbo limited to 4.0-4.2ghz which is definitely in the sweet spot of their efficiency.

[1610ftw]

1 hour ago, -batab- said:

 

I completely agree as I lean towards Intel. However, no matter the undervolt/optimization: it seems like under 30-40W AMD just wins in terms of efficiency.

 

This is also completely reasonable if we think that Intel scales nicely from 40 to 120W where AMD is super bad. It is quite legit that AMD shows more scaling margin in the idle-browsing TDP levels (5-30W). And this is also the reason why there is no competition in battery life usually. Curiously, Intel also looks more efficient in single core turbo boosted tasks as AMD probably pushed a bit over the limit to compensate for the low single core performance.

 

There are a few reviews by jarrod and some others that actually show nice scaling charts (e.g. image below)

 

Thanks for digging that up as I had seen that one, too but was too lazy to look for it 😄

 

On a normal day my 7760 with I would guess a cooling system that is about average has a CPU power consumption between 10 and 80 Watt with most of the consumption being more between 15 and 60. It is sitting on a slow spinning fan and I barely hear it all most of the day, certainly not when it stays below 60 Watt. So I would always look for the processor that is more efficient starting at ca. 50 to at least 80 Watt as I can easily go that high intermittently without being annoyed by fan noise.

Even a single run of CR23 Multi Core will not create much noise as with short power peaks the fans do only get louder for the second half of it and they go down quite fast after that, all in optimized mode.

[1610ftw]

6 minutes ago, win32asmguy said:

That being said I think the battery life is pretty good on the Precision so long as you take steps to ensure the Nvidia GPU is not activated unless it is needed. You can create a power mode for battery with P-core turbo limited to 4.0-4.2ghz which is definitely in the sweet spot of their efficiency.

 

That is an understatement - 8 hours for a mobile workstation with this kind of capability is very impressive and when properly done it seems that these P and E-Core HX processors can achieve good battery life as can also be seen here:

https://www.notebookcheck.net/MSI-Titan-GT77-12UHS-4K-review-Elite-amongst-UHD-DTR-laptops.644290.0.html#toc-4

Even top of the line gaming laptops like the MSI GT77 get up to 7 hours of battery life on wifi or playing videos while the Razer Blade 17 at least almost makes it to 4 and 5 hours respectively. Compare that to the Clevo X170KM-G (Schenker XMG Ultra 17) that basically comes with its own USV but not much more:

 

[-batab-]

1 hour ago, 1610ftw said:

 

That is an understatement - 8 hours for a mobile workstation with this kind of capability is very impressive and when properly done it seems that these P and E-Core HX processors can achieve good battery life as can also be seen here:

https://www.notebookcheck.net/MSI-Titan-GT77-12UHS-4K-review-Elite-amongst-UHD-DTR-laptops.644290.0.html#toc-4

Even top of the line gaming laptops like the MSI GT77 get up to 7 hours of battery life on wifi or playing videos while the Razer Blade 17 at least almost makes it to 4 and 5 hours respectively. Compare that to the Clevo X170KM-G (Schenker XMG Ultra 17) that basically comes with its own USV but not much more:

 

 

8 hours is definitely good and certainly not an issue for such a laptop. This is also the reason why I'm prefering Intel to AMD for this generation (and when talking of 15-17" powerhouses). However, almost every in depth review I found showed a 30 to 100% improvement in battery life when opting for the AMD counterpart. Many reviews even showed that AMD gets better performance when operated on battery due to Intel being TDP throttled (this is my assumption as I don't see any other reason). 

 

This is also fairly reasonable to assume since going for the 80-100W spot would drain a 99Wh battery in way less than a single hour.

 

However I also think that AMD will switch to the big little architecture in a few years. It just seems better even with this first iteration. I think that after a couple of iterations there should also be huge gains in scalability within the low TDP ranges.