Any reason NOT to buy the most-souped-up 13 AMD? Upside of lower tiers?

The 7840U also has 2 more (33% more) CPU cores, which is a pretty meaningful difference. It also has slightly higher clock speeds.

And Framework only offers the larger battery with the 7840U variant of the laptop.

In total the Framework Laptop with 7840U has (compared to 7640U):

  • 50% more iGPU cores
  • 33% more CPU cores
  • 11% more battery capacity
  • 9% more CPU total cache
  • 4% higher CPU boost clock speed
  • 4% higher iGPU clock speed

At the $320 price difference I think the 7640u is probably the better value (performance relative to price), but the 7840u is a reasonable choice if more performance is desired.

8 Likes

If you don’t need more than 16 GB of ram the mid-tier might have slightly better battery life than the top-tier, however that difference is likely very small. (More ram can consume more power, although if you are running out of ram the system will start paging which consumes even more power)

Both the top and mid tier should have significantly better battery life than the low tier. This is because the low tier has a smaller battery (55 Wh vs 61 Wh) and uses a weaker CPU.

A common misconception is that more powerful CPUs inherently consume more battery. But that isn’t automatically the case.

The 7640U is the same CPU design as the 7840U, however the 7640U has 2 CPU cores and 4 GPU cores permanently deactivated (either due to being defective or just cause AMD needed to meet demand for 7640U so they crippled some 7840U chips).

Ryzen processors tend to be pretty good at automatically putting most of the cores to sleep under low load (where they consume negligible power). So under low load the 7840U deactivates its extra unnecessary capabilities and behaves exactly like the 7640U.

Under heavy load both CPUs will consume 28w of power (what Framework’s cooling system is rated to handle), however the 7840U has more cores each operating at lower power (same total power as the 7640U). More cores each at lower power results in better performance than fewer cores each at higher power.

So both CPUs will consume the same amount of power under both light and heavy loads, however the greater performance of the 7840U will allow it to complete heavy loads quicker and overall consume less power.

IMO all but the base model Intel models are way overpriced.

Framework offers 3 models of Intel CPU: 1340P, 1360P, and 1370P.

The 1340P and the 7640U are competing CPUs and priced the same amount (in the Framework Laptop).

However the 1360P and 1370P are where AMD absolutely destroys Intel in value. The 7840U is competitive in price with the 1360P but competitive in performance with the much more expensive 1370P.

IMO the 1360P is such a small improvement over the 1340P that it just shouldn’t exist (and the 1370P should be priced at the price the 1360P is currently at).

The 1370P is just overpriced IMO. One of the selling points of the 1370P is that it supports Intel vPro Enterprise. vPro Enterprise is a suite of tools that allow a company to remotely manage large fleets of computers. It doesn’t matter to 99% of end users, however to the massive corporations buying thousands of laptops it is a feature they are willing to pay a premium for, so Intel prices vPro Enterprise capable CPUs at a premium.

The AMD CPUs tend to be more efficient and have much more powerful iGPUs.

Part of the efficiency comes from the lithography (manufacturing process). AMD is using a TSMC 4 nm process (the number loosely corresponds to the size of the transistors) whereas Intel is using their own process (which they used to call 10 nm but now call Intel 7 because they claim it is competitive with TSMC 7 nm).

So AMD is using a more modern and efficient manufacturing process, which helps a lot with efficiency.

Furthermore AMD’s cores seem to be generally better at providing high performance (at high power consumption) which under heavy load but efficiently dropping down to low power when high performance isn’t needed. Intel on the other hand has resorted to having some cores optimized specifically for high performance/high power and other cores optimized specifically for low power.

2 Likes

With other laptop vendors, it’s common (and good) advice to buy the most ram you might ever need, and then maybe some more, because it’s soldered on and you can’t upgrade it later. But not so with the Framework! You can upgrade later, so don’t worry so much about it.

I have the 7840U, and really … if all the CPU cores are busy, it thermal throttles in seconds. I’ve never really seen them all going (even at below-turbo speeds). But maybe it’s worth it for the GPU.

1 Like

I have the 7640U and honestly the only reason I’d upgrade for the 7840 is if I needed the extra graphics power. The 7640U has playable graphics imo but I also have low standards.

1 Like

AFAIK all modern laptop CPUs are limited by thermals. So if by “throttles” you mean that it clocks up to the turbo frequency, then quickly (seconds) clocks down so as to keep below 100 Celsius, getting to around the nominal 35W power draw, then I believe you can’t buy a laptop today that doesn’t do at least that.

But you shouldn’t see disabled cores or vector units, or frequencies dramatically lower than the base. (According to Notebookcheck and other reviewers, on the AMD Framework you shouldn’t see noticeable performance degradation on the scale of minutes, either, while other laptops do suffer from this to a varying extent.) So if you are seeing something dramatic—there have been some reports of bad thermal paste on the forum, maybe that’s your problem?

1 Like

Thanks everyone for your information and suggestions!

A couple of people mentioned looking at the DIY option, and it sounded like fun. I found all the alternative parts I wanted for a bit lower cost than the ones packaged with the Framework, but ultimately, even if I downgraded my RAM to half a gig, it was still less expensive to get the pre-built because I don’t have any parts lying around to reuse.

For reference, Framework just came out with a list of compatible RAM that they have tested:

and there’s a Poll in the forums that has a recent list of other options that people are trying from Crucial, g.skill, etc.

Since the RAM options I was considering would only save me a few bucks and weren’t on the compatibility list, I decided I’d rather be able to stay eligible for support, no questions asked. (I’m guessing the other RAM options that are truly compatible will eventually show up on the list, so maybe I’m being overly cautious.)

I may still switch out the SSD on the pre-built. I’ve read that the SK Hynix Gold P31 is good for minimizing power usage.

I am definitely going to buy a second power cord to use as the one I carry with me. I saw some recommendations in other posts.

1 Like

Thanks for your answer. I was also interested in this from a technical perspective, so that’s helpful.

1 Like

I have zero thermal throttling on mine after a PTM7950 repaste. I never got past 85C, even with a 45W CPU+GPU load ! I’m TPD limited on 7840U.

Edit: I had really bad temps out of the box.

I regret not saying that this is a choice you should make based on your own experience/comfort level. I think this forum is predominately occupied by DIY’ers, so you should take that into consideration when getting advice here.

That said, there seems to be some RAM that is consistently compatible, even though its not on the compatibility list. Kingston KF556S40IBK2, for example (which I have, but haven’t installed yet), and some Crucial and G.Skill RAM.

SSD-wise, SK Hynix P31 is popular for it’s power profile, but most NVMe drives seem to be compatible (and I think the P31 might not be as good as reviews imply due to laptop power profiles, but I digress).

So, I will summarize by saying that I still think DIY is the way to go, if you’re comfortable with this route, especially because you won’t buy a new component then immediately replace it… which goes against Framework’s ethos.

1 Like

Indeed, that’s what I expect; this laptop is still pretty snappy under full load, it’s not badly throttling, it’s just that there’s only so many watts a thing this size can dissipate. I also understand that more cores running at lower speed and power each can be more efficient than fewer cores running faster, because as voltage and speed go up, power use goes up even faster (squared, kinda). Anyway I figured 6 cores running a bit faster compared to 8 cores running a bit slower, at roughly the same total TDP, is close enough.

Hmm, my laptop seems to have reasonable temperatures under normal usage (50C?), but under load it definitely seemed worse than that. This inspired me to do some slightly more rigorous testing. I’m running linux, using turbostat to monitor freq/power and stress -cN to load N cpu cores, and ectool to monitor temp and fan, and control the fan.

What drove my impression is the default fan control, it’s really slow to react - it keeps fan below 3000 rpm for like 15 seconds, while the temperature can make it up to 94C, then the fan starts slowly ramping up. It makes sense to not be instant, and spin up and down annoyingly, but it is really slow to adapt.

But I can force the fan to 80% (6500 rpm), and then it’s mostly like yours, around 85C at 33W all-core load at about 4 GHz. Nifty.

So far my 7840U amd seems to be more limited by the hard wall power limit than thermals even without any fancy paste. I can’t really set the long term power limit high enough for it to thermal throttle once the fan kicks in (which does take a bit which I like, try to just tank it for a bit and then go into beast mode XD) but that may change in the future.

With stock paste I am barely touching 90 during an extended 35W load. Over all I am very pleased with the cooling performance of the thing. Turning a fan below 100% is a lot easier than turning it above so I am glad they went with a more powerful one than you’d find on similar laptops even if it does make them look somewhat worse in reviews.

Is yours actually thermal throttling or just pl?

It probably will be, those cores are freaking beastly. I did do some cinebench performance per watt testing, you may be able to just shift it downwards by 25% and get a rough idea.

How did you manage to keep a load above 35W? Mine pl throttles down to that and I can’t seem to set it any higher.

Appart from money, no.

Market segmentation. At least you actually get more cores on the gpu and cpu, look at the intel model. For an extra 320$ you get a whole extra 400mhz of theoretical boost clock and for an extra 720$ you even get 2 more pcores XD.

2 Likes

I’m not sure. Reported by powertop on battery, running the unpatched 6.7 Archlinux kernel and steam games. CPU only loads do not go above 35W, but GPU ones do.

I find it too fast actually. So many ramp up and down during constant loads, it’s annoying. It does wait until 94C exactly before turning on, even with PTM applied, then drops 15-20C and keeps the fan running.
I would rather they settle on a higher fan speed than going up-down up-down every 10 seconds on a perfectly constant load.

Ohhh you mean whole system power not cpu power. Yeah that sounds more plausible.

I was at the lowest screen brightness with no keyboard backlight and no audio. It’s less than a 5W difference, I think.

The fan at max alone draws 2W then there’s vrm inefficiencies and stuff like that. Enough for me to assume you didn’t bypass the 35W long-term power limit.

1 Like

This is difficult to answer since the worth of a dollar/euro/whatever is personal. I can afford to buy the top model every 3 years but still don’t since in my estimation the step up in price from the second best CPU to the very best CPU is quite ridiculous if you do not run your system at full power frequently. As a programmer i’m pretty much doing text processing and browsing 80% of the time. I’ve picked 32GB of ram so I can run a few VMs easily, but the top CPU did not make sense for me.

You are probably right. I just found ryzenadj and went to a 43W (hard limit) constant load. See [TRACKING] PPD changes have no effect after suspend on AMD until reboot - #16 by Justin_Weiss

TLDR, cinebench R23 went from 13818 to 14874 with unlimited cooling. There is serious throttling due to the chassis temperature triggering skin burn protection. The laptop body needs more room to breathe to sustain the TPD. CPU temperatures are fine.
I suspect GPU bound scenarios might benefit more from this.

You managed to go above 35 on the long term limit with ryzenadj? Mine only let the short term one go that high, lasts long enough for a couple cb15 runs for my power/performance map but pl throttles after that.

Point a fan at the underside of the laptop. Even a small one.

Turns out I was wrong too, I can indeed set the long term limit to 43 but it thermal throttles to 35, I was just looking at the wrong sensor. My bad

This little exercise did get me in a testing mood though so I am currently running tests with different thermal limits and fixed fan speeds so I can compare with ptm7950 (not doing lm jet and if the ptm performs well maybe not even at all)