Earlier this month, a rather confused rumor popped up claiming that AMD would be moving away from its partnership with Asmedia and would begin building its own chipsets once again. Some sites have suggested that AMD was moving away from Asmedia altogether, while others have implied AMD would design its own X570 chipset, but that lower-end boards would continue to rely on Asmedia parts.
We now have confirmation that at least some of AMD’s 500-series chipsets will use Asmedia hardware. DigiTimes reports that “Taiwan’s ASMedia Technology is expected to land contract design orders for all mainstream PCIe chips from AMD even after the US chipmaker rolls out X570 motherboard chipsets that support PCIe 4.0 in mid-2019.”
First, one point of clarification: Because modern CPUs integrate a significant amount of I/O on-die, PCIe support is no longer strictly a question of what the motherboard provides. AMD’s 7nm Ryzen CPU (Matisse) already has PCIe 4.0 baked in and on-die. It’s possible that some 300 or 400-series motherboards may be capable of PCIe 4.0 speeds already, at least on the first GPU slot. Even if AMD chooses not to support PCIe 4.0 in anything but a 500-series board, however, Matisse has that functionality on-die.
This image of the X370 chipset shows which chipset connectivity is provided by which hardware block. The forest green links correspond with the PCIe 4.0 connectivity Matisse can offer without requiring any kind of corresponding chipset support on the motherboard. The question is whether AMD wants to field PCIe 4.0 support off what we used to call the southbridge. Boosting the interconnect bandwidth from PCIe 3.0 x4 to PCIe 4.0 x4 would improve maximum theoretical bandwidth.
It’s not clear, at the moment, if the X570 is a standard Asmedia design or if AMD took a hand in developing it. We’d expect the X570 to contain some Asmedia technology no matter what — it’s highly unlikely that AMD would develop its own USB 3.1 controller, for example. There have also been rumors that the X570 might be a relatively high-power chipset, with a TDP of ~15W, compared with a typical 6-8W for a desktop chipset. If true, at least some of that impact would likely be the PCIe 4.0 bus support.
The bigger question is whether PCIe 4.0 support is going to be a significant practical benefit. We’re starting to see NVMe SSDs approaching the theoretical maximum transfer limit of PCIe 3.0, which means, yes, some truly cutting edge storage arrays and applications will be able to use the additional bandwidth. The impact on the typical user, however, is likely to be much smaller. Most users don’t put enough of an I/O load on their system to benefit from the jump in PCIe bandwidth.
The jump in GPU bandwidth will be marginally useful eventually but isn’t likely to make a major difference in overall performance. Desktop GPU tests have demonstrated that single GPU configurations aren’t particularly sensitive to PCIe bandwidth unless you’re running multiple GPUs. Even then, it’s typically been more important to hang both GPUs off the CPU / northbridge in an equal x8/x8 configuration as opposed to running in a lopsided x16 (CPU) / x4 (Southbridge).
One place where PCIe 4.0 might come in handy is with eGPU configurations. Tests have shown that high-end GPUs lose significant amounts of performance when run in an eGPU enclosure, though it looks as if NV GPUs lose more performance than AMD cards do. Whether this reflects the intrinsically lower performance the RX 580 offers compared to the 1080 Ti or the work AMD did to optimize eGPU configurations in the first place is unclear. Either way, the places where PCIe 4.0 will matter most are the areas where relatively few PCIe lanes are offered. Boosting an x4 PCIe 3.0 slot to PCIe 4.0 is more likely to impact overall performance than upgrading a single GPU configuration in the same fashion.
Let’s block ads! (Why?)