Will APUs benefit from PCI-e 4.0?

[JustMe21]

I was wondering if APUs will benefit from the extra bandwidth provided by PCI-e 4.0?

 

[whm1974]

Not really. PCIe-e 3.0 is more then enough for high end video cards.

 

[Arkaign]

I was wondering if APUs will benefit from the extra bandwidth provided by PCI-e 4.0?

Absolutely not, unfortunately.

Now DDR5, yes. That is the big bottleneck for APU performance. Shared system memory.

 

[whm1974]

Absolutely not, unfortunately.

Now DDR5, yes. That is the big bottleneck for APU performance. Shared system memory.

Let's hope for the APUs sake that Intel and AMD don't decide to go single channel for their budget processors.

 

[Arkaign]

I am hopeful that someone will make async SLI type tech in a future DX variant. Theoretically PCIe 4 would be fast enough to put a couple of different GPUs together and make for a faster GPU (vGPU?). Various attempts previously have met with only moderate success. I remember adding a 7570 or something like that to a Mobo that supported their weird hybrid crossfire, and it actually did work alright.

Being able to mix say an RTX 2250ti and an RTX 2260 together in a couple of years that would roughly equal an RTX 2280 would be cool.

Aside from the random hybrid crossfire with the old Radeon that I only did because I happened to run across matching parts at the time, the last multiple GPU thing I think I did was add a 9600GT for physx back in the day.

 

[Arkaign]

Let's hope for the APUs sake that Intel and AMD don't decide to go single channel for their budget processors.

Ah man, don't get me down lol. That would be really disappointing for sure.

It's really impressive how well multichannel memory scales. I have worked with a ton of x58/79/99 builds, and triple and quad channel is definitely awesome. QC 1600 DDR3 is a pretty competitive match to DC 3000/3200 DDR4.

Conversely, the same is true for running single channel, ESPECIALLY with iGPU/APU. It's simply dogshit to be kind lol.

 

[whm1974]

Ah man, don't get me down lol. That would be really disappointing for sure.

It's really impressive how well multichannel memory scales. I have worked with a ton of x58/79/99 builds, and triple and quad channel is definitely awesome. QC 1600 DDR3 is a pretty competitive match to DC 3000/3200 DDR4.

Conversely, the same is true for running single channel, ESPECIALLY with iGPU/APU. It's simply dogshit to be kind lol.

For what it's worth, I doubt AMD will go single channel even for "budget" laptops, but Intel might depending on what OEMs are asking for. Atom core with single channel DDR5 Atom I can see going in this direction.

 

[IntelUser2000]

QC 1600 DDR3 is a pretty competitive match to DC 3000/3200 DDR4.

Well, it should. 4x 1600 = 2x 3200.

Conversely, the same is true for running single channel, ESPECIALLY with iGPU/APU. It's simply dogshit to be kind lol.

The laptop space, especially the value end of the market is extremely cost sensitive, and 2x DIMMs cost more than 1x. The way for AMD to get out of that rut is by being in more systems, because then there will be more choices and more configurations.

 

[LightningZ71]

I see PCIe-4 (and 5) being a boon to APUs for several reasons:
1) The link to the I/O chip gains a lot of bandwidth for situations where those are used.
2) NVME drives can run on 2 lanes instead of 4 for the same performance, giving the possibility of either running extra NVME drives, or reducing the cost of implementing them
3) The 8 PCIe lanes that the APUs currently expose to the motherboard slots will be able to be effective for higher end GPUs (that are configured for PCIe4 and 5) without diminishing their performance.
4) With lane bifurcation, the APU can directly run a 4x PCIe 4 slot for a mainstream card and still have 4 more lanes for additional slots.
5) Cheaper NVME drives can operate with better performance by using HMB instead of on IC DRAM for scratch memory. The faster NVME/PCI connection means more available bandiwidth to the host to access the DRAM, and with the expected capacity increases coming with DDR-5, the host will have sufficient RAM for the buffering to not make a significant impact on system performance. The higher DDR speeds will also enhance this functionality.
6) With a potential chipset driven Thunderbolt implementation in the future, having the higher PCIe bandwidth between the CPU and the chipset will allow that to perform more closely to its expected levels.

All in all, I think that APUs will definitely enjoy a lot of benefits from PCIe 4 and 5, though those won't all be evident in every implementation.