Im planning on buying the AMD Ryzen 7 7840HS version of the FW16. How badly would that bottleneck a RX 6950 XT. Assuming that Im either using an oculink or thunderbolt 4 as these are plans for the far future. And if the CPU is not strong enough will framework plan on selling stronger CPU’s?
The bottleneck is likely going to be bandwidth not the CPU. I use an eGPU albeit with an Intel 1240P. The experience is adequate for what I play so I’m not that concerned. There are various tricks you can do to decrease the penalty but there is always a penalty.
I’d go oculink over TB4 if you can manage it.
so the 7 7840HS is going to be able to keep up with the RX 6950 XT assuming I use oculink
Your cpu should not be bottleneck. It’s a pretty decent cpu for gaming.
Good luck on using Thunderbolt 4.
The eGPU enclosure is the most likely thing to be a bottleneck. Unless you are doing some home made method most of the commercially available enclosures only support transfer speeds equivalent to four Gen 3 PCIE lanes, so about a 4GB/s transfer rate.
Keep in mind the expansion bay with the 7700S has eight Gen 4 PCIE lanes which is rated for about 16GB/s, so four times the bandwidth.
wait so even if framework makes a oculink expansion bay, it will be capped at 16GB/s ? Please excuse my ignorance when it comes to this. Is 16GB/s enough for high proformance cards like the RX 6950 XT? Is does that card draw over that amount?
I am not the most familiar with occulink but from what I am finding it looks like the most common way that it is being implemented for things like eGPUs is to use an m.2 slot for the PCIE connection since m.2 slots have 4 PCIE lanes running to them. It won’t scale 100% perfectly, but from this article that examines the performance of an RTX 3080 in PCIE 4.0 X16, PCIE 3.0 X16(Equivalent bandwidth to PCIE 4.0 X8 like the expansion bay connection) and PCIE 3.0 X8(Equivalent bandwidth to PCIE 4.0 X4 like a PCIE 4.0 m.2 slot) you can see that you are going to lag behind with occulink if it is running on 4 PCiE lanes but it shouldn’t be too bad. If we get an expansion bay which utilizes all 8 lanes it should be just a bit behind a full X16 connection. Tested with a much weaker GPU, the RX 5500 XT you can see that things like a traditional eGPU with PCIE 3.0 X4 starts to fall far behind.
Nice! I really appreciate you going out of your way to help me better understand what all this meant. I do really hope that it will be able to utilize all 8 lanes. That is a defining factor for me purchasing a framework.
Josh is working on an 8i interface, so that would be 8 lanes.
Then framework indeed would be a product for me Also is josh like one of the people on the framework team? CEO? thirdparty? I would like to follow up on this.
Josh is a 3rd-party.
Framework Founder / CEO is Nirav Patel. His username is “nrp”.