Discussion in 'AMD x86 CPUs and chipsets' started by RnR, Apr 5, 2021.
you heard it here first.. Sept 21 for Zen 4 !!
and here's why
But then what is this war going to do to prices of EVERYTHING... maybe we should buy now instead of waiting???
sell everything and invest in crypto and NFT
I have to wonder if they've brought some of the E core branch prediction and things into the P cores.
Much like modulation schemes, anything that's good for wifi is 10x better for fibre.
Given that the E are so damn good for their transistor budget, any advances that damn good should also be good on the P cores - without any clock boost.
I thought the branch prediction improvment was around which core to run down ?
can it go E vs P type logic ?
Nah, IIRC it's the magic that makes our processors seem faster than they really are.
Instead of doing nothing and waiting, they process ahead of our actions, and store the result. Then whatever you do, if the prediction is good, it usually already has the answer.
I get the branch prediction and out of order stuff, I mean they've improved selection of what instructions run through each core as they're fundamentally different in operation.
You can't add bits of one core to the other core and expect similar results
Are you talking about the scheduler?
The E cores are Atoms with a bunch of optimisations like a new and better branch prediction that make them as good as a skylake with far fewer transistors.
The P cores are evolved Skylake++
So if you can optimise the shit out of the Atoms, it stands to reason that some of those optimisations can lead to better performance on the P cores.
It's an interesting era, where two separate design teams are competing in a sense, albeit with different underlying targets.
I'm saying the exact opposite.
The design choices for one a core type don't translate as improvements in another core type.
A deficiency in one core may not be deficiency in another core type.
If you targeted high frequencies then a longer pipeline is usefull, but that choice doesn't translate to a core that targets different requirements and may be a massive disadvantage/liability.
Sorry for poor word choice and ability to explain myself. I'm doing some long days on power restoration after the flood up here. So the old noggin ain't firing cleanly
bunch more roo-moors over here
Am i going blind or did we have a 7000 thread ??
anyway, here will do
If rumors are correct, the mainstream Ryzen 7000 desktop processors will be based on the "Raphael" design. Aside from Zen 4 CPU cores, Raphael is supposed to include a small Radeon GPU in each processor, meaning that for the first time, we'll have mainstream desktop Ryzen processors that don't require a discrete graphics card to be used. These upcoming chips have a lot of things going for them that justify the increased power limit, including higher clock rates, the aforementioned integrated GPU, and brand-new AVX-512 support.
But used for what?
probably not good enough for VR, which i personally set as my minimum...
its on onboard iGPU .. used for those that only require a desktop gui and basic decode of multimedia.
VR isnt widely used, it's not the target market
I'm talking the performance level though, not the use of that performance..
That's just the level I would want as minimum
Well for you there's a $4k nvidia card just released
this is iGPU for corporate workstations and those general purposes boxes in every office reception around the world
If only HSA was still a thing. Could use those extra iGPU cores for something useful. Imagine a RDNA2 iGPU that does the Ray Tracing for you.
I read somewhere that and were looking to have raytracing on the APU at not embarrassing levels so that it could be used for offload ...
Not sure if it was a rumour or a roadmap though
Well I just remember the days that AMD was touting HSA as the perfect solution of their first generation APU's on Socket FM1. It was going to be a new type of crossfire.
They dumped it when Lisa Su took over. Which is a shame as they could use it with good effect when the RDNA2 APU's come out. Having those 8 iGPU cores working on Ray Tracing, while most of the discrete GPU just does its usual Rasterization would be a clear victory for AMD.