Why does the 8250U require such a high voltage?

This difference has been bugging me over some time ever since I got this laptop, but why does the 8250U (and presumably other 8th gen U series chips require such a high voltage to be stable?

I ran a Prime95 Stress Test, set Speedshift EPP to 0, and used TPL-Speedshift Limits in Throttlestop to control the maximum frequency of the CPU, starting to 0.4GHz all the way to the maximum Turbo Frequency of 3.4GHz on all 4 cores. The 8250U turbos to 3.4 regardless whether the load is on 1 core or all 4 cores.

All the tests were done with a -125mV undervolt.

Then I graphed the voltages and package power consumption. The graphs are here, and the excel spreadsheet is attached.


To me, the most efficient operating speeds for my 8250U was in the blue bars, between 1.6GHz and 2.3GHz. Beyond 2.3GHz, the Core Voltage ramps up significantly and power consumption shoots up.

The graph also explains how my 8250U could be significantly less efficient than my friend's 7700HQ. We both ran TS Bench 32M with the speedshift value set to 0 and a -125mV undervolt and no throttling. Guess what?

The 7700HQ completed the 32M Test in 7.3 seconds with an average package power consumption of 18W and average voltage of about 0.97v
The 8250U completed the 32M Test in 8.0 seconds with an average power of 25W and voltage of around 1.11v
Yes, both processors just so happened to have an equally stable undervolt amount and equally similar max quad core turbo clock of 3.4GHz. As well, my 8250U just so happened to have a similar Prime 95 power consumption and Passmark CPUMark performance when the TDP Limit was raised to 44W

I simply couldn't believe this and made my friend send a screenshot. So my 8250U, at times, will use more power yet perform worse under some workloads, probably due to its higher turbo voltage. But sure, in many other day to day workloads, my 8250U is vastly more efficient compared to the 7700HQ, but that all changes when turboboost kicks in. I get it, TurboBoost was never supposed to be continuously utilised, nor was it ever meant to be efficient, but why the big difference between a <strike>high performance<strike> HQ CPU and a "low power" U CPU?

Also to note, the 7700HQ idles at a lower voltage (0.6v @ 800MHz) whilst my 8250U idles higher (0.72v), both at a -125mV undervolt. Interestingly enough, my dual core i7-6500U would idle at 0.55v and sustain 3.0GHz turbo at 0.97v. But that's a dual core CPU and it may not be related.

I read online that CPUs use more power when they are hotter... I've never noticed an unstable undervolt that was stable on 40C but locked up at 90C

If I disable turbo (And thus limiting the CPU to the 1.8GHz TDP Level 2 limit) I can undervolt to -200mV. Doing so results in the performance slightly better than the stock 7500U but with almost half the power consumption

I understand that CPUs are complex and that I'm comparing CPUs of different generation, but isn't 8th gen supposed to be a refresh of Kabylake? So what is causing the difference between the voltages and power consumption of these mobile chips?

 

8th gen is nothing more than 7th gen with moar cores.
An 8700K is a 6 core 7700K. They even reach the same clocks with 8700k slightly better since the process is more mature.

And a HQ is NOT a high performance CPU. Dear God please don't call it that.
You just made @Papusan vomit even more. Oh dear.....



P.S.

HQ CPU=Heavily Qrippled.
HK CPU= Heavily Kancerous.

 

Could you and your friend compare results in same way with Cinebench R15 ? As I understand your BGA manage 3.4GHz who is equal clocks as 7700hq on 4 cores load. Thanks

Same IPC. Skylake-Kaby and Coffee Lake. All the same for single cores load.

 

I will have to do in in the next coming weeks. It's school holidays and lord knows what my friends do (waste their lives playing games 12hrs a day. that's what everyone at my school does)

Well you see, back when I was 10 years old, I received my first ever computing device with a DUAL CORE AMD E2-1800 running at a blistering 1.7GHz which took 3 minutes to boot up to Windows 7 and another minute to open Internet Explorer (don't kill me for using it. I was just a child) Then I saw a laptop with the 4700HQ with literally 12x the performance of my E2. I have never owned any device with anything better than a U CPU.

 

Not the same!!!
+1 bin speed needed for BGA CPU compared to LGA CPU!!!

learn to swim, Poopoosan !

 

Who said I compared LGA vs BGA?

BGA as well have Skylake, Kaby and soon new Coffee. Would you think Coffee BGA have better IPC than the BGA predecessors And have Kaby better IPC than Skylake in the BGA family?

No I think I can swim

 

I suppose its like the old analogy of two cars: a Porsche doing 30mph will use a lot more fuel than a Mini, but a Porsche doing 100mph will use a lot less fuel than a Mini doing the same speed as it doesn't need to work as hard. The U line of CPUs are probably pretty far down the binning process and therefore are poor quality silicon compared to the MID range 7700HQ ( @Papusan ). I'd assume the silicon is pretty bad as generally lower quality chips need a higher voltage to sustain a given clock speed.

 

Yeah, Intel is smart, they pack in their scam into max boost who will be turned down after a few sec. See... single round Cinebench R15 and the power seeping out as the air from a punctured balloon

 

I believe this can be avoided in an unlocked Bios.
I turned my 7820HK into a 7700HQ by changing EC registers (it locked the TDP at 45W sustained and 62W for 28 seconds, like locked chip).
Then if you change IMON SLOPE to 25 or 50 (thank you @Vistar Shook , your CPU will report its using half the TDP), then you can pass 45W unlimited time.

I have not tested this on 7700HQ real chips but I think others can try. But they have to unlock their Bios.

If they are afraid, maybe they can use "RU" and find the register and change it to 25 or 50 in EFI prompt, but that depends if it accepts a DECIMAL or HEXADECIMAL value. And this may not work, usually the EFI prompt and setup strings are changed with "enabled" and "disabled" values of 0x01 and 0x00. Not sure if you need to convert 25 or 50 to hexadecimal or not then use 0x32 (32 hex=50 decimal).

 

Yes it can be done with the HQ chips and an unlocked BIOS, the hard part is getting the chip to use more than 45W anyways.

 

Awww, I feel so sorry for you guys with locked BIOSs (Ugh what is the plural for BIOS? BIOSes?) All I have to do is raise the PL2 Long Duration

Anyway I ran the Cinebench R15 @Papusan and came out with a score of 740, which, according to the 7700HQ's page by Notebook Check my 8250U is on par, or slightly better than the average. The package power consumption was 30W constantly with a core voltage of about .98v and core temp topping out at 65C

 

Yes the 8250U is on par with the HQ (afterall, both 4 cores/8 threads) but if you run CB15 many times in the row you can see it cannot sustain that performance, unless the PL is removed. So the difference is mostly allowed TDP and perhaps binning.

 

No, I already removed the PL. Or at least I changed that to 30W. It could sustain that performance indefinitely as long as the thermal limits allow, which it does. It's how I run cities skylines.

That's the thing. I can change the PL on all my acer laptops as well as my Xiaomi, but it seems other manufacturers lock it down or have terrible VRMs that can't handle the current (Ahem XPS). My cooling is great as well

 

Indeed removing the PL limit it performs like a HQ, but to answer your question, the difference to the HQ, or the HK or the desktop cpus is binning. The U chips are usually lower quality, so require more volts and watts at the higher clocks compared to the HQ. The same goes when comparing the laptop HK cpus to the desktop K cpus. The HK cpus also require more voltage and wattage at higher clocks than the K cpus, which can clock much higher because they are better quality.

 

Or 7820Hk in the Alienwares

Almost on level with unlocked i7-7820Hk in the new Alienwares Keep on pushing


And you beat it in temps LOOOL


@Che0063 set the Cinebench process's priority to realtime if you haven't done it. I want to see Alienware scores and with lower temp, LOOL

 

How did you remove the PL limit for 8250? With XTU? or RWeverything?
Not sure what the 4 core clock speeds are for the i7-8550U, I think 3.7GHz, so if the PL is bypassed, then it should peform better than the i7-7820HK at stock clocks (3.5GHz x 4 cores), although at higher voltage and watts.

 

@Papusan

We did it! Full 3.4GHz Turbo on all cores!

I actually had to run the test twice, once with the CineBench set to Realtime, which got me 753, and another set to high, which got me 752. This is because when I set the Cinebench to realtime, the laptop would lock up because (presumably) Windows was allocating everything to Cinebench and not leaving enough to other processes, but when the benchmark finished windows would resume like nothing happened

@Vistar Shook Yes you can use XTU. Download and install Intel's Extreme Tuning Utility. I call this the lazy method because Intel XTU is slow, constantly uses CPU cycles in the background, and resets the moment I sneeze in the wrong direction. But hey, some people like coloured blue bars and lines. But I go into RW-Everything>Memory>FED15900>Offset 160.


The first underlined is what you are interested in. That sets your Long term power limit. Kudos to @Falkentyne and @unclewebb for finding this method. Not me! It's the 6th post of this page

 

I have to smile. The cooling in the 2 kilos Xiaomi Mi Notebook Pro is more effective than the TRIPOD for unlcked BGA in the Alienware Just look at the temp results in Cinebench R15. This made my day.

Lack only one point

 

Super. Good Job.

 

Good job squeezing that much out of it.

 

great that you compile data into a chart. to answer some question, its just intel's chip. most of their cpu behave similar way and will be most efficient at a certain frequency range, of course thats based off their current manufactured silicon quality. intel marketing can call it w/e they wanted, U, HQ, QM, MQ etc but ultimately it falls into what you have recorded, basically your graph summarizes it, though results may still vary between chips but they all relate to one another.

 

Wow, performance has definitely tracked backwards since BGA and the pursuit of thin and light laptops when you have 4-5 year old laptops outperforming many modern laptops. Those scores are right on par with Apple's new top of the line Macbook Pro 15's too.

 

I seem to remember a former Intel engineer on Quora talking about how some chips on a wafer perform better a lower voltage (better perf/w) and worse at a higher voltage and are therefore binned as mobile CPUs. The opposite is true for desktop chips. It certainly explains why my i7 7700HQ idles at 800 MHz at 580 mV and consumes around 700 mW of power. I highly doubt that you could do this with a 7700K despite it being "binned better". Please correct me if I'm wrong.

Useful link: https://pdfs.semanticscholar.org/06d8/9bb120032e7cb30651820bb4fd15dd51dd49.pdf

 

This might've been true for higher tier -QM/MQ and -XM/MX PGA processors in the past, but these days mobile BGA is a dumping ground for low bin chips that can't make it into desktops, requiring more voltage at the same frequency and not being able to overclock as well. Conversely, you could run desktop CPUs underclocked to BGA levels with better efficiency.

 

Proof? Do you have benchmarks of a 7700K vs a 7500U/7700HQ running 800 MHz with the accompanying voltage and power consumption figures?

 

Read this: https://www.quora.com/How-does-Intel-design-and-produce-so-many-models-of-CPUs

 

I wont talk about 800MHz. I have never run my chips this way. But 2016 or there around I tested BGA clocks in benchmarks with my [email protected] and compared with 6700Hq. My 6700K (not the best binned and not even tweaked for lowest voltage) run around 13% more efficient than these so-called High Performance i7 BGA chips. Same was it vs. 6820Hk. All of the BGA chips used more power. And most of them high undervolted.

 

800MHz?

Many systems don't even idle that low.

No I was talking about 6820HK/7820HK vs. 6700K/7700K at turbo clocks and overclocking. Numerous examples of the HK chips' garbage silicon quality plastered in these forums. Like the 6820HK struggling to get above 4GHz with high volts pumped through it. My last BGA CPU was a 4720HQ that did 3.6 at 1.0444V at its max undervolt, while typical 4770/4770K do 3.7 at around 1V stock.

 

Oh you guys mean this power guzzling cancer chip?
Package power was 115W.

One
Hundred
And
Fifteen
Watts.



Brother @Papusan's chip uses less power to do 5 ghz.

 

LGA@50x.


Or 10 point lower score in Cinebench R15 but with 34% less power.

 

You're all missing the point. The issue under debate is whether a i7 7700K is more efficient than a i7 7700HQ which is running at stock clocks, i.e. 3.4 GHz on all 4 cores. Here's a screenshot of my i7 7700HQ (undervolted by -100 mV) running prime95 (max power consumption preset) at 3.4 GHz on all cores @ 980 mV.



At idle:


Bear in mind that I lost the silicon lottery and I can barely undervolt my chip by -100 mV. The vast majority of i7 7700HQ undervolts that I have seen are stable at -130 mV, i.e. 3.4 GHz @ 950 mV at full load. I once saw a chip go as low as 875 mV @ 3.4 GHz.

Running a i7 7700HQ at 3.4 GHz and 875 mV and 0.950 mV yields power consumption figures of 28.0 W and 33W respectively. I challenge you achieve better results on a i7 7700K.

 

Oh and please post screenshots of your clocks, voltages and power consumption at idle. I want to see if you can get as low as 825 mW!

 

hm lets see, my 7700k stock load voltage under p95 avx is 1.240v on average. i run 4.8 ghz stable with -40mV offset, thats 1.200v load. ive done under/overvolting and stability testing in the range of 4.4 (stock max. turbo boost all cores) up to 5.1 ghz (with 5.2-5.4 ghz only useable for quick benches and validation purposes). so the lowest i can offer is 4.4 ghz with -145mV p95 avx stable offset, so thats 1.095v at load.

id be curious to know how much voltage i could additionally shave off that when lowering my clocks another full ghz to 3.4, hmmm...

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

8550U (2nd highest tier i7 kbl-r) here, can go to 0.4v on throttlestop

undervolted under battery saver setting, with turbo disabled and SST 135

 

Jesus at those voltages it would consume less power than a Snapdragon 835!

It's also worth pointing out that we are only considering performance per watt at full load. The vast majority of everyday workloads like browsing the internet, using Word and watching videos will only require short amounts of CPU power at a specific time (e.g. when rendering a web page) meaning the CPU will be idling for most the time. U and H processors consume very little power at idle (around 1 W) and will therefore give you much better battery life than a power hungry desktop chip.

 

Please do test this. It seems that you won the silicon lottery, so it's fair then to compare your results to 875 mV i7 7700HQ in my previous post.

 

Web browsing + discord messenger average is around 0.9-1.4w

 

I don’t think anyone is missing the point but.....

... this really seems like a race to the bottom.

Get your 7700HQ running under load at >=4.2 on all cores (which is stock for 7700K) and then we’ll compare who sips the least kool aid.

 

Jaybee,
use Throttlestop to underclock the 7700K multipliers and voltage manually (it can go down to 0.6v, and 800 mhz but don't go that low). Set it to the same clocks as gorilla then do the same test and give the watts shown. Then we'll see your CPU will use FAR less power than the BGA turd will.

 

Yep race to the bottom....and closer to the bottom they all start looking the same. After all, the binning process is to select the chips that can clock higher not lower.

Enviado de meu Pixel 2 usando Tapatalk

 

Really convenient for you to completely HIDE the
1) prime95 window while it's running, while NEGLECTING to tell us whether you were using AVX or FMA3 or AVX disabled (Core 2/Type 0 FFT small FFT's. I assume you said max power consumption so you used the middle option? No one uses that setting. We all use a) Small FFT Power virus (the usual one for max cpu cache/core hitting), 2) Blend (balanced/RAM testing, but hits the small FFT's sometimes), c) fixed FFT 1344K custom for vcore stability. ).,

2) the max POWER DRAW button in Throttlestop, 3) the max power draw area in HWinfo64!
What are you trying to prove here by hiding stuff while trying to argue things?

I'm going to teach you how to PROPERLY run these tests. Going to be a few minutes before the screenshots. Laptop's busy doing work.

 

you won't see enthusiast testing that, because it is unrealistic and during usage the power drain is multiple times of that wattage. like taking a vehicle to know how much gas it consumes while its on but not driving it rofl seems kinda pointless. btw, there are so MANY things that can go wrong with measuring so i'll list a few for you.

- hwinfo isnt exactly accurate, i'd take throttlestop over it any day if its related to CPU.
- your screenshot idle wattage is nothing but a peak at minimum, what is the average?
- what window power profile are you using
- what settings within that particular window power profile are you using
- what OEM/ODM bios settings are put in place? ( ie an ultra book design would probably have most on power saving mode, not just the cpu but graphics, PCIE devices, chipset, storage and since all are connected to CPU, thus would affect your average wattage completely even during idle).

 

Ok now this is how you do tests properly without being a Jabroni.

1) 3.4 ghz, 0.990v (matching your undervolt exactly), prime95 SMALL FFT, AVX DISABLED.



2) 3.4 ghz 0.990v, prime95 "Max power consumption", AVX disabled.

 

I used FMA3, 36K Lucas Lehmer iterations.

As for point 2, I used HWInfo because it stores the MINIMUM power value. Also the power values in HWInfo and ThrottleStop are exactly the same.

 

Oh cool NOW you tell me, AFTER i already did AVX disabled and AVX enabled. Now I have to do FMA3.
And tell me how did you bypass the 45W TDP? You should have been power capped at 28 seconds at 45W.

Only two people on this forum know how to bypass that.
Me and my brother @Vistar Shook

Since you neglected to give us proper information here's AVX porn for you with FMA 3 coming after.

 

Blend Preset

 

OK lol I think I've reached the limit of my cooling. Prime 95 on option 1 (I think it was FPU stress) runs the CPU right up to the 44W Turbo Limit I've set, and the CPU can't even turbo to 3.4GHz even under a -125mV undervolt. The temperatures climb up until 90C 40 seconds later, and I stop the test because I don't want to ruin my liquid metal application (or at least keep it for longer).

I can run prime95 under a 33W Turbo Limit and everything will stay nice and cool under 80C. So my heatsink can handle 2x the power from my CPU but not 3x. That's pretty good for me.

@Falkentyne I was jealous of those temperatures, but lets see if your heatsink can handle 90-130W of power

See this is what I mean. Both you and @RampantGorilla sustained 3.4GHz turbos with lower voltages than my 8250U.

 

3400mhz 0.995v, FMA3 "max power consumption"
P.S. @RampantGorilla Your temps are horrible. 6C core temp difference at 34.5W ? Where's brother @Papusan when I need him?

 

My heatsink can handle it. The EC shuts the laptop off however.
Because of:
CPU_CrtT = 1.CPU_ThtlT = 1.CPU_ThtlT = 0.SYS_CrtT = 1.SYS_ThtlT = 1.SYS_ThtlT = 0
If CRT=1 is set to a 0 then I could draw unlimited power or destroy the VRM's. Then buy a Clevo P870 TMG @Papusan
But no one knows how to safely edit the EC file to change this without an instant brick.

BTW I have the world record for 7820HK on cinebench. I knocked @Vistar Shook to 2nd place.
Submitted to HWbot. Power draw was 114W. Can barely compete 2 runs without the EC shutting off the laptop when the temps reach 90C.




Here was a previous run with TS open. Forgot to press the power max draw button but it was 114W.

 

I can't seem to increase the power limits for the small FFT Test, I've set it to 50W in Throttlestop, but it doesn't seem to have an effect. I think it's locked in the BIOS. Here's the screenshot anyway.

The Original max power consumption stress test (large FFTs)