Framework Intel Core i7-1165G7 benchmarks

I benchmarked the Intel Core i7-1165G7 in Linux against a whole slew of other CPUs I’ve had access to over many years.

I’m using:

and others where it’s harder to interpret or compare the results.

All tests were done plugged in with 100% battery.

Here’s the 1TB WD_BLACK SN850 in the Framework:

versus the same type of drive in my desktop:

(Sorry no write measurement, it can only be done offline.)

So - close but not quite. I’m thinking this may be due to the SN850 in a lower power state in the Framework? Or thermal throttling?

I won’t be getting into Geekbench, it seems I have something wrong. See the thread below.

Not too many results from 3D graphics benchmark glmark2.

glmark2

But it does slightly exceed the performance of the old yet discrete NVIDIA GeForce GT 640 DDR3. It greatly exceeds the older high end Intel “Crystal Well” iGPU in the Core i7-4870HQ. It cannot match the performance of the low-end GDDR5 GPU, the NVIDIA GeForce GT 1030, or the low-end GDDR6 NVIDIA Quadro T400 though.

Lots of data from HardInfo!

This is a multithreaded benchmark and it’s a bit better than the Core i7-4870HQ but not quite up to the level of the Core i7-4790K. It does exceed quite a few desktop CPUs with much higher TDPs.

Here it completely outclasses all my older laptop CPUs, including the powerful Core i7-4870HQ by a small margin. It cannot quite equal the performance of the desktop Core i7-4790K.

This is a single threaded benchmark. It’s the 2nd fastest CPU I’ve ever tested and is just barely beaten by the monster Ryzen 9 5900X (105W TDP). Not bad for a 28W TDP CPU!

I’m unable to explain the strange, seemingly random results here.

This is only available in more recent versions of HardInfo so fewer CPUs were benchmarked. It beats two lower-powered Intel mobile CPUs but not the high performance Core i7-4870HQ. It also cannot match the relatively low-powered desktop Ryzen 3 3200G despite having HyperThreading, but it is at much lower TDP.

Testing fast Fourier transforms on the floating point unit, it seems that Intel’s recent performance here has fallen behind where they were. Older CPUs will beat it as well as all desktop AMD Ryzens. Performance is still decent but not leading.

The 1165G7’s FPU fares much better on raytracing, where it’s the fastest I’ve ever tested.

NBench is exclusively single-threaded, an area where the 1165G7 shines. Only the Ryzen 9 5900X beats it, and not by a huge margin. The Ryzen 7 3700X just barely beats it in integer calculations. There’s no evidence of any floating point weakness in this benchmark. This is especially impressive for a low TDP mobile CPU that has a lithography disadvantage over the AMD Ryzens (10 nm for the 1165G7 vs. 7 nm for the 3000 and 5000 Ryzens).

Conclusion

The Intel Core i7-1165G7 is a very fast CPU period, let alone the fact that it’s a 28W TDP mobile U-class CPU. It’s an extremely fast mobile CPU and even holds its own against much higher power desktop CPUs.

It is very impressive at single-core tasks. Multicore it cannot compete with 8C/16T or 12C/24T CPUs, but that should be obvious without benchmarking. However it beats other 4C/8T CPUs in most benchmarks so it is quite competitive.

This is no mean feat considering it’s a 10 nm thermally constrained mobile CPU competing against 7 nm desktop CPUs with much higher power limits.

9 Likes

The CPU got beaten by a 7 inch GPD Win Max running i5-1035G7. Booo

gpd image

That thing above scored 4841 multi core and 1470 single core.

Really?
I mean, yes, it’s a 7 inch brick with screaming dual fan at 5500 rpm providing the 25W of cooling when you give it the beans. But … but the 1165 is supposed a …
oh. they are the same gen. okay. but still.
Anyhow, you can enjoy the teardown as I dive in to see exactly how this works.

I wouldn’t say so! I guess I’m comparing it to my MSI laptop with its even-thinner chassis and a 45W TDP performance CPU though.

But the fan is not noisy or whiny and does definitely move the heat.

Devices like the one you link to and like the thin yet powerful MSI I mentioned are not comfortable to be around. Noisy and hot. The fan spins ridiculously fast and on my MSI sounds “hysterical”, rapidly spinning up and down, making lots of noise as it goes.

The fan on the Framework is much better behaved and the CPU performs better as my benchmarks show.

I mean, if you want the other device rather than the Framework no one’s stopping you. But I think you’ll find performance-at-all-costs devices like that aren’t meant to be everyday devices. The Framework strikes a much better balance and is really very fast when you measure it.

1 Like

Yes. I see that they have a 65mm unit. But 65mm of …?
I am assuming it’s the entire width of the fan, but …
well, I don’t have a ruler, but even if I do its meaningless. Framework do however claim that it provide 28W of cooling on the replacement part description. And the thermo seem pretty beefy.

yes but no. I have the GPD, as you can see because how I wrote my picture-missing-teardown.

How much TDP does framework give it? because the TDP is adjustable, up to 25W. If it hit that score on a 20W tdp budget, that’s very nice.

I was measuring exactly 28W on “package-0” in s-tui in Linux when at least one core was at 100% load.

1 Like

hm.
Can you see the core temps? because if it indeed is using 28W it make less sense that a smaller chip with less wattage beat it. unless it throttle

Yes. As I recall, on a single-core load I once saw a brief spike to 95°C for a second on one core before the fan kicked in and the CPU frequencies adjusted. This is in line with what @nrp suggested would happen on a heavy load, a brief spike to about 100°C before it settled in.

On 100% full load all cores it would settle down to 82°C exactly. At first it revved up to all cores 4100 MHz like the specs state, but it did slow down to all cores 2800 MHz, so maybe there is throttling going on.

Here’s a sample s-tui output at idle.

Scrolling down to see the power data on the lower left:

2 Likes

either it’s a fan curve tuned for quiet rather than cool, or the thermal is indeed insufficient.
on a sufficiently cooled system like GPD I don’t think you will ever see temps past 90. Like, ever. And during performance-demanding tasks the cores will be around 70, if I remember correctly.
I’m kinda disappointed. However, I have to say that it’s at least better than XPS 9360, which is a plus.


Did you do this plugged in?

Plugged in with battery at 100%.

1 Like

Huh. I guess there is, I’m way down the list?

:confused:

The only thing I did that might impact performance was install TLP, but that should get out of the way on line power.

I’ll delete the Geekbench results from this thread then. I’m inclined to delete the rest since they may not be valid, but they seem to make sense and I have no way of knowing.

Sorry for this.

True, and I’m using a really weird kernel, “5.14.0-1004-oem” which is so new that it’s untested, untuned and unproven.

Momentarily spiking to 100 before settling in to the low 80s is expected design for this chip set.

1 Like

nm is a meaningless number I’m afraid

The transistor density between the 2 processes actually favors 10nm

I’ve really screwed up this thread and reached all sorts of incorrect conclusions.

I’ve requested it be deleted.

I mean, based upon the other benchmarks I remember seeing at tiger like launch, 1165g7 is the king in single threaded performance compared to any laptop sku

So that’s on the money, and given the clocks and IPC are higher than 4th gen, I don’t see why it would be unreasonable to suggest it’s a more performant CPU than 4790k

I really just skimmed the thread, where did you screw up?

https://community.frame.work/t/framework-intel-core-i7-1165g7-benchmarks/9629/10?u=user1209

Some of those geekbench results you see here were from when I was testing an issue I was having with the cpu throttling.

The post in question with links to many of those geekbench scores. What I found out was that the performance of the CPU is heavily dependent on having a powerful enough power supply plugged into the laptop. Those 1650 single core, 5700 multicore scores were from when I had a 90W power adapter plugged into the laptop.

I would be interested in the results of your test if you redid them with a stronger power supply.

There is some anomalous performance behavior going on when the laptop is plugged in and charging with a less powerful power supply. In fact, my on-battery performance was better than the geekbench results you have above there, and using the framework power adapter my scores were inline with what you got. It’s all detailed in that thread I linked above. This might explain your benchmarking results.

1 Like

There is actually a meaning here.
The smaller the transistors, the less amount of power it uses (at least, to my knowledge), and the less heat it generates.

The density is completely irrelevant. You can stack transistors, even layers. It just doesnt make a lot of sense in the end.

And thus, the 7nm will beat 10nm because it generate less heat and use less power, and therefore pound for pound, 7nm wins. And yes, if you put a refrigerator on top of the 10nm, it also wins.

The nm in the process name don’t relate to any real world measurements anymore. Its just a marketing name at this point. There is nothing in a 7nm transistor that is 7nm long. A 10nm process can be better than a 7nm process from another manufacturer. Its just a name and says nothing about how good or efficient the process is. The fact that tsmc 7nm is more efficient than intel 10nm is nothing to do with its name. It could also be the other way around

Btw smaller transitors a higher heat density. If you decrease the size of your transitor by lets say 50% and double the efficieny you also increase the transitor density by a factor of 4 (size of the transitor squared). In other words you chip only consumes half the power but has double the heat density as before. Things would only get worse if you add more layers and build 3 dimensional. As a result your chip with a higher transitor density is way more difficult to cool at the same total power consumption. Thats one of the reasons why the heat density and total power consumption of modern processors is now way higher than lets say 10 years ago despite being more efficient

2 Likes

Excuse me lol, how exactly do you think we have managed to get such giant core counts in such little packages

You think a modern CPU core count can be constructed on a process node from the early 2000’s and still be the same size package? Nope, not gonna happen, regardless of power draw or heat output(although that also would balloon) the physical size of the chip would be much larger

To demonstrate how meaningless the “nm” is a meaningless number

Do you know the difference between TSMC’s 16nm and 12nm process node? Not much-it’s more akin to intel adding another + to the end of their 14nm process than a true evolution or die shrink

As @_Asic was saying, you cannot compare process technologies between manufacturers

The number can be manipulated in any number of ways and is more akin to a guesstimate than a true measurement

Tons of articles to read on this subject if you care to look

I suggest AnandTech or AdoredTV if you’d like to learn more-TechTechPotato is another good resource(Dr. Ian Cutress of AnandTech runs that channel)

I’m not going to suggest that Intel is god tier or that AMD is god tier, both are great imho-I’m glad Intel is finally moving off of 14nm tho, it was a workhorse but it was time to move on

1 Like

Another benchmarking point regarding graphics, Unigine Heaven run on the Phoronix Test Suite. Same systems as the glmark2 benchmark but different results…

The Iris Xe defeats even discrete cards here.

1 Like