Comment by fc417fc802
8 hours ago
> keep in mind that modern integrated graphics can be really good
Sort of but not really? Without the wider bus (and various drawbacks) of GDDR any iGPU is very severely memory bandwidth constrained compared to a dGPU from an equivalently recent generation. It's just that a 3 or 4 generation old dGPU is sufficient for an ever increasing number of titles.
TBF it entirely depends on what is meant by "really good".
> I’d have a hard time recommending a new laptop purchase with 16GB if your intention is to keep it many years.
One of my devices still has only 8GB and I use it daily. It's really not an issue for most things.
Don’t forget that all MacBook models are “integrated graphics.” If you start doing graphical stuff or gaming you’ll need more RAM. If you’ve got a Windows laptop with an RTX 5050/5060 with its own dedicated memory, then yeah, you don’t really need more than 16GB unless you’re doing something specific.
Integrated graphics are constrained but often sufficient to play AAA games.
Panther Lake laptops are roughly equivalent to an RTX 4050, which is only a single generation old. Obviously not the strongest GPU to compare against but you’re also pulling a lot less wattage doing it.
I’m sure your 8GB device is working great, but someone buying a brand new device really should avoid that since the implication is that they’re going to keep that device for somewhere between 3-7 years.
Do those laptops not have on the order of ~64 GB/s bandwidth? Meanwhile your example of an RTX 4050 is a mobile SKU with 192 GB/s while the lowest entry level dGPU from that generation, the RTX 4060, boasts 272 GB/s. The RTX 4070 range ups that to 500 GB/s and the high end breaks 1 TB/s.
tl;dr iGPU vs dGPU memory bandwidth is a >10 year difference.
So now to double check my guestimate of 3 to 4 generations from earlier. Entry level RTX 30 448 GB/s in 2020, eventually downgraded to 168 GB/s in 2024. Entry level RTX 20 448 GB/s in 2018 eventually downgraded to 336 GB/s in 2019. You have to go all the way back to the GeForce 10 series to dip below 100 GB/s and even then that's due to some highly unusual configurations introduced many years after the initial launch. At launch in 2016 the lowest end GTX 1050 pulled 112 GB/s which is nearly 2x Panther Lake IIUC.
Ah, I understand where you’re coming from now. You’re really heavily focused on memory transfer speed. That’s really not the only measure of GPU performance and certainly not the one I’m interested in for my workloads. (I do not care for local LLM work for one thing)
For example, the ThinkPad X9 15p with Panther Lake integrated graphics will do 80FPS in Cyberpunk 2077 while the RTX5060 in a Legion 7a will do 109FPS on the same settings (1920x1200, quality up scaling, no frame gen).
So what that means is you can get a super quiet thin and light laptop that gets double digit hours of productivity battery life and has roughly the same tier of gaming experience as a thicker, heavier, power-hungry gaming PC that is pumping 115 watts into the GPU.
To answer your question, LPDDR5x should be more close to 100GB/s so I think you are underestimating the memory bandwidth of Panther Lake laptops. Not sure though.
That all said, overall a RTX5060 or higher laptop would be preferable, I just find it hard to find one that checks a lot of the “thin and light 13-14” premium long battery life productivity laptop that can game on the side” boxes.
> Do those laptops not have on the order of ~64 GB/s bandwidth?
The Macs? No, they go up to 614 GB/s, nearly an order of magnitude more.
Are we talking about gaming or GPGPU? Because AFAIK gaming is more likely to be constrained by SSD performance than GPU memory bandwidth.