Precision 7680 / Precision 7780 pre-release discussion (Raptor Lake, 2023)

[win32asmguy]

6 hours ago, Etern4l said:

Single score performance doesn't really scale much anymore, but for multicore Raptor Lake blows Alder Lake out of the water: +33-40% on the desktop side, and seeing similar if not larger improvements with the mobile chips:

 

https://www.notebookcheck.net/Intel-Core-i9-13980HX-Processor-Benchmarks-and-Specs.675757.0.html

https://www.notebookcheck.net/Intel-Core-i9-12950HX-Processor-Benchmarks-and-Specs.618743.0.html

 

CBR23 Multi 20K vs 31K median. Amazing.

 

The single core score is better, just because of higher clock speeds (if you have a paste job good enough to allow over 5ghz on a single core without thermal throttling)

 

With multicore it has the same thermal system and power limits so its essentially robbing performance from the p-cores for additional e-cores. At least it also has better memory options.

 

In the case of the 7780 it also has the unfortunate combined load PL1 throttling down as low as 40W. This is better than the 7670 which goes down as low as 30W in the same scenario. It should really be 55W minimum in the highest performance mode while plugged in.

[Etern4l]

54 minutes ago, win32asmguy said:

 

The single core score is better, just because of higher clock speeds (if you have a paste job good enough to allow over 5ghz on a single core without thermal throttling)

 

Much less impressive an improvement, +13% to be exact in the case. Not clear why a single core, even running at 6GHz would stress even a laptop to the point of throttling.

 

54 minutes ago, win32asmguy said:

With multicore it has the same thermal system and power limits so its essentially robbing performance from the p-cores for additional e-cores. At least it also has better memory options.

 

I don't believe that's what's happening. Optimal utilization of limited thermal solution capacity involves downclocking first and foremost (the marginal cost of additional 1 step in ratios increases sharply at the high end).

Prior to that, the scheduler prioritises putting load on the primary logical CPU of each P-Core, then eCores, then the secondary logical CPU (hyper threading). Makes sense as the performance scaling with HT is only about 50%.

This results in pretty amazing improvements in MT performance with just a single gen, and the same dated fab process. 

[Aaron44126]

11 hours ago, Ionising_Radiation said:

Now that @win32asmguy has a XX80 device, maybe it's time to start an Owners' Thread proper? @Aaron44126, thoughts?

 

Yeah, seems like it; I usually leave it to someone who actually has a system in hand to kick off that thread, but I'll just go ahead and create one if it looks like no one else is going to do it...

[win32asmguy]

2 hours ago, Aaron44126 said:

 

Yeah, seems like it; I usually leave it to someone who actually has a system in hand to kick off that thread, but I'll just go ahead and create one if it looks like no one else is going to do it...

 

I went ahead and created one, will try to post a review soon but I have yet to create a rough draft.

 

6 hours ago, Etern4l said:

Much less impressive an improvement, +13% to be exact in the case. Not clear why a single core, even running at 6GHz would stress even a laptop to the point of throttling.

 

In single core I have seen the 13950hx draw up to 1.502v stock at 5.5ghz. Not sure where that falls compared to the 13900k, but it can draw a decent amount of power at that speed.

[IamTechknow]

@win32asmguy Thanks for sharing the benchmark results! Was undervolting available on the system or is it locked?

[Etern4l]

21 hours ago, win32asmguy said:

 

I went ahead and created one, will try to post a review soon but I have yet to create a rough draft.

 

 

In single core I have seen the 13950hx draw up to 1.502v stock at 5.5ghz. Not sure where that falls compared to the 13900k, but it can draw a decent amount of power at that speed.

 

That voltage looks high, but that's OK. CPUs can draw very little power at high voltage. Power draw is a function of both voltage and current, the latter always being regulated based on load. 

For example, I run my 13900K at fixed voltage (VCORE I believe) and power draw varies from 15 to 260W, clearly depending in the load. 

 

Puy another way, nax voltage determines max power draw at full CPU load. Consequently, I wouldn't expect any significant power draw, therefore temps, with just a single core load on a 24 core CPU. 

[1610ftw]

21 hours ago, win32asmguy said:

In the case of the 7780 it also has the unfortunate combined load PL1 throttling down as low as 40W. This is better than the 7670 which goes down as low as 30W in the same scenario. It should really be 55W minimum in the highest performance mode while plugged in.

 

This handling of situations where sustained power is required can only be called pathetic. While it is not only Dell one would think that with all the feedback they are getting and with their Alienware laptop branch they would be willing to at least allow sustained loads of 80W provided that certain temperature thresholds aren't exceeded.

 

And if the heat sink isn't beefy enough it would be easy enough to improve it. There are several aftermarket heat sinks out there for Clevo laptops that show us that it does not take huge sales numbers to do a redesign and if Dell is too lazy they may want to ask one of the guys who designed some of these monsters:

 

 

 

 

[win32asmguy]

23 hours ago, 1610ftw said:

This handling of situations where sustained power is required can only be called pathetic. While it is not only Dell one would think that with all the feedback they are getting and with their Alienware laptop branch they would be willing to at least allow sustained loads of 80W provided that certain temperature thresholds aren't exceeded.

 

And if the heat sink isn't beefy enough it would be easy enough to improve it. There are several aftermarket heat sinks out there for Clevo laptops that show us that it does not take huge sales numbers to do a redesign and if Dell is too lazy they may want to ask one of the guys who designed some of these monsters:

 

Yeah after looking into what causes this, it apparently is the Intel "Innovation Platform Framework" which is just a new name for DPTF. Throttlestop can disable its effects or the drivers can be blocked from installation.

 

So it can go higher with that disabled but it still runs into thermal throttling. PTM7950 repaste is the next step.

 

An aftermarket heatsink is an interesting idea, but I am not sure how they could improve on it. Vapor chamber? Additional heatpipes? A thicker bottom panel to allow for higher CFM fans and larger fin stacks?

[1610ftw]

19 minutes ago, win32asmguy said:

 

Yeah after looking into what causes this, it apparently is the Intel "Innovation Platform Framework" which is just a new name for DPTF. Throttlestop can disable its effects or the drivers can be blocked from installation.

 

So it can go higher with that disabled but it still runs into thermal throttling. PTM7950 repaste is the next step.

 

An aftermarket heatsink is an interesting idea, but I am not sure how they could improve on it. Vapor chamber? Additional heatpipes? A thicker bottom panel to allow for higher CFM fans and larger fin stacks?

 

Looking forward to what you can achieve with Throttlestop and PTM7950 - beyond 30K should easily be possible in single runs.

 

I was not so much thinking that an aftermarket heatsink would be nice but I was showing them as an example that they were produced with very low volume for a limited market so surely Dell could have taken the time to improve upon the heatsink of the 7780 with its much higher expected sales volume. It would also help numbers if they just went with a single heatsink for all models with a dGPU that is based on the requirements of the top configuration. I am sure that not too many people would complain about too much cooling.

[win32asmguy]

On 4/25/2023 at 11:33 AM, 1610ftw said:

Looking forward to what you can achieve with Throttlestop and PTM7950 - beyond 30K should easily be possible in single runs.

 

I was not so much thinking that an aftermarket heatsink would be nice but I was showing them as an example that they were produced with very low volume for a limited market so surely Dell could have taken the time to improve upon the heatsink of the 7780 with its much higher expected sales volume. It would also help numbers if they just went with a single heatsink for all models with a dGPU that is based on the requirements of the top configuration. I am sure that not too many people would complain about too much cooling.

 

It has been a few days and so far it seems like CBR23 Multi has gone up slightly (best seen was 29,900) and CBR23 Single can now sustain 5.5ghz, scoring 2150. Also Timespy Graphics scores 16,600 which is up a bit as well.

 

I also got the bios dumped, and tried to do some undervolting which was not successful. The "undervolt protection" remains enabled regardless of the value set to the hidden UEFI variable similar to the experience reported on the XPS 17. Setting voltage offsets for P-cores also did not have any effect. I also tried modifying AC/DC LL to both 1.35 which did not boost scores any.

[1610ftw]

4 hours ago, win32asmguy said:

 

It has been a few days and so far it seems like CBR23 Multi has gone up slightly (best seen was 29,900) and CBR23 Single can now sustain 5.5ghz, scoring 2150. Also Timespy Graphics scores 16,600 which is up a bit as well.

 

I also got the bios dumped, and tried to do some undervolting which was not successful. The "undervolt protection" remains enabled regardless of the value set to the hidden UEFI variable similar to the experience reported on the XPS 17. Setting voltage offsets for P-cores also did not have any effect. I also tried modifying AC/DC LL to both 1.35 which did not boost scores any.

 

Preventing any undervolting is a big no-no in my book.

 

Dell really has dropped the ball this generation with their liquid metal disaster with the AW18 and now they prevent undervolting in the 7780. Do they have an official stance on the matter?

 

In any case I bet you are happy with the single core performance!

[vooze]

On 5/2/2023 at 12:44 AM, 1610ftw said:

Preventing any undervolting is a big no-no in my book.

 

Dell really has dropped the ball this generation with their liquid metal disaster with the AW18 and now they prevent undervolting in the 7780. Do they have an official stance on the matter?

 

In any case I bet you are happy with the single core performance!

 

Undervolting is not really priority in the enterprise where they sell 99% of their precisions.

[yslalan]

It surprised me that the Precision 5680 is designed to have a removable GPU daughter board.

 

Dell released a video 2 weeks ago:

 

 

[1610ftw]

On 5/3/2023 at 12:33 PM, vooze said:

 

Undervolting is not really priority in the enterprise where they sell 99% of their precisions.

 

With the performance that they achieve it should be.

[Aaron44126]

Dell Precision 5480 and 5680 support pages are up, so those systems are presumably launching soon.

 

https://www.dell.com/support/home/en-us/product-support/product/precision-14-5480-laptop

https://www.dell.com/support/home/en-us/product-support/product/precision-16-5680-laptop

[AL123]

5 hours ago, Aaron44126 said:

Dell Precision 5480 and 5680 support pages are up, so those systems are presumably launching soon.

 

https://www.dell.com/support/home/en-us/product-support/product/precision-14-5480-laptop

https://www.dell.com/support/home/en-us/product-support/product/precision-16-5680-laptop

Yeh should be tomorrow 🙂 

looking forward to testing out a 5680 atleast although it is soldered ram which puts some off but we will just max it out from factory to avoid regrets In future, easier to justify that for a company vs an individual.
 

 Think it will be a good option for our marketing team who want a mix of power for adobe suite and portability, will be interesting to compare to the 7680, it’s always hard until you physically hold a decide to really appreciate the size/ weight difference. 

[Aaron44126]

15 hours ago, AL123 said:

Yeh should be tomorrow 🙂

 

5480 and 5680 are now available for order in the US.

https://www.dell.com/en-us/shop/workstations-isv-certified/precision-5480-workstation/spd/precision-14-5480-laptop

https://www.dell.com/en-us/shop/workstations-isv-certified/precision-5680-workstation/spd/precision-16-5680-laptop

[PHVM_BR]

45 minutes ago, Aaron44126 said:

 

5480 and 5680 are now available for order in the US.

https://www.dell.com/en-us/shop/workstations-isv-certified/precision-5480-workstation/spd/precision-14-5480-laptop

https://www.dell.com/en-us/shop/workstations-isv-certified/precision-5680-workstation/spd/precision-16-5680-laptop

5680 has vapor chamber option (owner's manual) but no indication as to which configuration it is available from.

Maybe from RTX 3500 Ada?

[the_professor]

Looking at the Precision 5680 owner's manual the height dimensions are:

• Front height 26.17 mm (1.03 in.)
• Rear height 26.17 mm (1.03 in.)

While the spec sheet mentions:

• Front: 20.05mm, (.80in)
• Rear: 22.17mm, (0.87in)

Quite the difference!

 

[win32asmguy]

2 hours ago, PHVM_BR said:

5680 has vapor chamber option (owner's manual) but no indication as to which configuration it is available from.

Maybe from RTX 3500 Ada?

 

It make sense if it were from the 3500 Ada and up, as the same heatsink should be compatible 4000 Ada, 5000 Ada and even the RTX 4090 mobile option.

[AL123]

3 hours ago, the_professor said:

Looking at the Precision 5680 owner's manual the height dimensions are:

• Front height 26.17 mm (1.03 in.)
• Rear height 26.17 mm (1.03 in.)

While the spec sheet mentions:

• Front: 20.05mm, (.80in)
• Rear: 22.17mm, (0.87in)

Quite the difference!

 

Odd indeed I think when I’ve queried in the past there are some differences between how measurements are reported in different territories but not noticed a difference in owners manual vs spec sheet before.

 

hoping to have my hands on a unit in the next couple of weeks and compare to 7680

[the_professor]

I'm more inclined to believe the spec sheet dimensions looking at this comparison shot from this video:

 

https://www.youtube.com/watch?v=Kho5KL0z0Ss

[AL123]

10 hours ago, the_professor said:

I'm more inclined to believe the spec sheet dimensions looking at this comparison shot from this video:

 

https://www.youtube.com/watch?v=Kho5KL0z0Ss

Good find! if we went with the manual sizes it would be pretty much the same as the 7680 sim model albeit a bit lighter. I also note they don’t quote the weight with the 99.x whr battery only the smaller one which isn’t available at launch. I suspect the 5680  will be more like 2kg with the larger battery still an appreciable amount lighter than the 7680 which would be a bigger thing for us, the 7680 is already pretty compact but quite hefty. 

[yslalan]

Comparing my Precision 7670 (Perf. chassis) and XPS15

Comparing my XPS15, MBP16, and 5770

[vooze]

Just got the first Precision 5680 delivered. It's actually pretty nice.

Performance is very good for i9 (13900H) 45W CPU (it can boost quite high) and it's usually quiet and cool, compared to 7680.

- Screen (FHD) is the same as 7680's FHD screen.

- Keyboard is more like Precision 5570 than 7680, which is okay for me.
- Touchpad is haptic like a MacBook. Will take some getting used to, but so far it's good.
 

Still need to test speakers etc.

It's a good in between device for me (sysadmin) where 7680 was a big heavy and 3580 was not powerful enough.