AMD's Ryzen 4000H APUs limited to only 8 PCIe 3.0 lanes for dGPUs? A non-issue mostly, even for dGPUs above an RTX 2060

[Redaktion]

Many users have been asking why OEMs are not providing dGPU options above the RTX 2060 for the Ryzen 4000H laptops. Would this latest reveal regarding the limited PCIe 3.0 lane access add to the conspiracy theories? They really should not, as the issue barely affects performance even for more powerful dGPUs. Tinfoil hats aside, AMD is most likely imposing these restrictions in order to keep laptop prices in the US$1,000 range.


https://www.notebookcheck.net/AMD-s-Ryzen-4000H-APUs-limited-to-only-8-PCIe-3-0-lanes-for-dGPUs-A-non-issue-mostly-even-for-dGPUs-above-an-RTX-2060.482722.0.html

[opelit]

But OEM will hear Nvidia and will not put higher cards to PCIE3x8.
Its funny, as they see no issue by installing single channel DDR....

[_MT_]

If you intended to install a dGPU in a desktop, why would you choose APU instead of a normal Ryzen CPU? Especially since APUs lag behind. In laptops, it's done for efficiency. Not as big a deal in a desktop. It's good that it offers more lanes but I'm not convinced they're going to be used for dGPUs.

[deksman2]

If you intended to install a dGPU in a desktop, why would you choose APU instead of a normal Ryzen CPU? Especially since APUs lag behind. In laptops, it's done for efficiency. Not as big a deal in a desktop. It's good that it offers more lanes but I'm not convinced they're going to be used for dGPUs.

An APU is useful in a desktop for various people who might not see the need in getting a standard CPU only.
The 4000 series APU's are comparable to their desktop counterparts, AND they come with an iGP. (which seem quite potent).
Think of all in one solutions. but for desktops.

Apart from that, not using 5700M, 2070 and above in laptops just because of 8 PCIe lanes is not a good reason - and also, I don't think its the ACTUAL reason.

For example, the DELL G5 15 SE is the ONLY laptop on the market to come with 4800H and 5600M... but despite the fact AMD also has 5700M (both 5600M and 5700M are highly efficient and powerful GPU's and comparable if not better than NV counterparts), no other OEM's will be using them - but DELL's cooling implementation is not really good to begin with (its not even allowing either 4800H or 5600M to reach their maximum potentials - whereas other laptop chases do).

Most Zen 2 laptops aren't being paired with anything above 2060... and I doubt the lanes is the reason why.
If they would put up a 'limit'... why not do it with 5700M and 2070 instead?

There's a much larger performance loss the higher above that you go and an unnecessary increase in power consumption... not to mention cost.
But then again, other OEM's don't really care and they still use RTX 2070, 2080 and 2080ti in laptops.

[_MT_]

An APU is useful in a desktop for various people who might not see the need in getting a standard CPU only.
The 4000 series APU's are comparable to their desktop counterparts, AND they come with an iGP. (which seem quite potent).
Think of all in one solutions. but for desktops.

We're talking about a situation where you have a dedicated GPU. You haven't provided even a single reason for using APU instead of CPU if you have a dGPU. I know that on Intel platforms, some people use iGPU along a powerful dGPU - for example for video encoding, as I understand it. And it still seems like a niche use. Since big Ryzens don't have an iGPU, this is even less common with AMD (although I'm sure there are oddballs with, say, a 3400G and a dGPU).

[jeremy]

w.r.t. the article: Sure, RTX2080Ti Desktop with 11GB of its own VRAM, outputting to a monitor using its own video outputs is not as affected by PCIe3x8. However, we've already seen 5500XT 4GB is affected by it, sometimes severely. Furthermore, most laptops with Nvidia dGPUs use Optimus (AMD does something similar nowadays), which feeds every single frame back over the PCie link, further impacting the limited bandwidth.

@_MT_ it seems the monolithic APU does have a bit of a slight performance edge in some cases vs the 3700/3800 systems, inspite of the much smaller L3 cache. For me, it's having the IGP for my VM host - since only Intel supports GPU sharing on consumer HW, AMD and Nvidia only support it on their pro cards. This means one of the two GPUs have to be integrated onto the CPU. I could chose a weaker Intel system or a more powerful AMD... hard choice /s. My system is ideally mITX, so I don't quite have the space for 2 GPUs.

Of course, neither company's approach is ideal, IMO.

AMD gives their desktop platform 20 lanes from the CPU (more from the chipset), but starves their upper range APUs (lower range gets the same layout, but fare better against Intel U series).
Intel gives all of their CPUs an IGP and 16 lanes, but what's good for an H series laptop is not always the greatest for a desktop. This isn't even speaking of the cheapness of Intel, where their U/Y series laptops have always gotten 4 lanes from the CPU (the 10, 12, and 16 lane configurations all come from the chipset - NO EXCEPTIONS). Intel makes an even greater mess by allowing OEMs to permanently downclock the 4 PCIe3 lanes between the CPU and chipset to PCIe2 speeds, further starving what little bandwidth there is.

In the end, no decent full stack approach, IMO.

[cb88]

AMD has officially stated that there are 24 lanes on desktop renoir.... which means the GPU gets 16 lanes. And there is an addition a 8 lanes probably divided to chipset and 1 NVMe.