Able to upgrade not just the speed but manufacturer of CPU?


For anyone wondering more about the i.MX8 GPU the series is outlined here;

http://www.cnx-software.com/2014/04/19/vivante-gc7000/

It should be the plain GC7000, but i.MX8Quad series seems to have two of them. There's more detailed information from NXP here: https://community.nxp.com/servlet/JiveServlet/downloadBody/331530-102-4-267278/DES-N1940 i.MX 8 Graphics Architecture.pdf This clarifies that the high end part has 256 medium precision GFLOPs/s or 128 high precision.

The only GC7000 series device I can find is the UltraLite deployed in a Marvell SoC that's used in some Samsung phones, like this one: https://gfxbench.com/device.jsp?benchmark=gfx31&D=Samsung+Galaxy+J1+Ace+(GC7000,+SM-J110F)&testgroup=overall

The GC7000 in i.MX8QuadMax should be about 4-8x faster in theory, going off of how the FLOP, fillrate, and triangle rate specs have scaled.

I've been using iPad 2 as a rough proxy for Pyra's GPU performance (SGX543MP2). I can only find GFXBench 2.7 T-Rex scores for comparison: https://gfxbench.com/compare.jsp?benchmark=gfx40&did1=2623&os1=iOS&api1=gl&hwtype1=GPU&hwname1=Imagination+Technologies+PowerVR+SGX+543MP2+(dual+core)&D2=Samsung+Galaxy+J1+Ace+(GC7000,+SM-J110F)

But if this is any accurate indication at alli.MX8QuadMax should offer several times the GPU performance OMAP5 does.
 
  • Like
Reactions: rSl
oo, two GPUs, why i would need two, nobody knows, but that sounds cool :)

They did it that way to allow running two OSes on the same SoC without having to virtualize the GPU (which is tricky business and not totally viable on all of them)

But I wonder if their hardware/drivers/whatever is good at making the two GPUs perform as well as one GPU when it needs to.
 
They did it that way to allow running two OSes on the same SoC without having to virtualize the GPU (which is tricky business and not totally viable on all of them)
In other words: GPU passthrough. I'm doing that on my desktop machine - difficult to get it to work, but working out really well. It might even be possible to share the screen with both GPUs given how independent the involved components are being driven on such embedded hardware...

But I wonder if their hardware/drivers/whatever is good at making the two GPUs perform as well as one GPU when it needs to.
The OMAP5's SGX544 has 2 cores as well, after all. I'm generally surprised how similar those SoCs are, the i.MX8 has 2x M4s, a DSP and a SATA controller as well - must be some kind of common automotive requirements.
 
The OMAP5's SGX544 has 2 cores as well, after all.

SGX544MP2 is built up from two units but I doubt it can really work like two independent GPUs with separate registers and command interfaces and everything. This was standard scaling for IMG but I don't think it works that way for Vivante's "Vega" GPUs - there are much higher end offerings than GC7000 that don't involve coupling multiple separate GPUs together.

This is why I have some question about performance/efficiency. NXP says that the two GPUs can appear as one, but I'd be a little surprised if this was a hardware feature they engineered and not a driver trick.
 
Back
Top