You missed the part of "semi-custom", it can even be monolithic for all we know. Only thing for certain is that CPU cores will be based on Zen 2 and GPU on Navi-architecture, which can have any amount of CU they choose
Next Generation Hardware Speculation with a Technical Spin [pre E3 2019]
- Thread starter AlNom
- Start date
-
- Tags
- 5 bytes non sequitur
- Status
Do you have sources for those imaginary quotes ?Dr Lisa Su at AMD CES: (Microsoft) is really looking at revolutionizing the console era for the next decade
Dr Lisa Su at GDC: (Google) is really looking at revolutionizing the console era for the next decade
Dr Lisa Su at AMD XXX: (Your favorite company who is partnering with us) is really looking at revolutionizing the console era for the next decade
Jen Sen Huang at whatever: Nvidia invented everything
Amazing how this PR thing works..
So basically so don’t know anything?
D
Deleted member 11852
Guest
Technically these are all ENIAC based.Technically these are all "Fermi based..."
Not true
PS5
Zen 2 based
Navi based
Solid State storage
But anything else is just assumptions and guessing. We've not had any, what I would deem credible leaks on anyone's system.
MS system we know that Xbox had some input from azure team, but that's it.
Love_In_Rio
Veteran
And that is abaco based.Technically these are all ENIAC based.
D
Deleted member 13524
Guest
If he had sources they wouldn't be imaginary..?
It was just a joke.
And some form of ray tracing, and 8K output.
Yea, fair point, also PS4 BC.
On the surface that may not sound like a lot, but it could mean a lot of customization to Navi as AMD are saying its not just GCN.
I doubt that the software stack is as well placed to emulate it as Xbox.
I'd say that there was a significant change in architectural philosophy between Fermi and Kepler that included the removal of certain concepts like half-warps that wouldn't start to show again in a somewhat different way with Volta/Turing's concurrent issue of FP and INT instructions. The ISA changed significantly, per Nvidia's statements. For example, it appears Fermi had example of mem-op instructions combining accesses with operations, which was abandoned in future generations. The way it handled ALU pipeline latency shifted from run-time scoreboarding to static software scheduling.Technically these are all "Fermi based..."
From Maxwell through Volta, there's been some changes to the ISA and changes in the architecture as presented to software, although it's not clear the shift was as far-ranging as changing the ISA from one category of CISC mem-op to something more streamlined.
Volta included a change in threading model, which at least when enabled is a change to a part of the foundation of the architecture. There's been some wild swings in instruction encodings as well in the post-Kepler era, but since GPUs often rely on run-time translation and Nvidia doesn't commit itself to the internal ISA like GCN does it may not be as clear a distinction. From reverse-engineering, it seems like post-Volta instructions have gone to 128 bits and I recall that somewhere the encoding's bit order was reversed.
AMD has made (then unmade) some significant dislocations within its ISA encodings, and more publicly documented them as part of GCN. Despite that, the instructions themselves have generally the same way of doing things since GCN 1.0. GFX10 does change at least some things, though most of the instructions and their overall structure and behavior are familiar or even a throwback to Sea Islands.
The previous GCN used GloFo's 14nm. "2.56x power efficient" does not mean 2.56x the performance. I think the process to 7nm gave Radeon VII around 1.3x the performance. It's difficult to make a precise estimate. That "1.5x power efficiency" claim for Navi considers architectural improvements and the performance gain for switching to 7nm. Also it is probably misleading because it might be comparing cards of different TDPs. They said compared to Radeon VII we can expect 1.25x the performance, but that doesn't add up because 1.25*1.3=1.625, not 1.5. Maybe around 10% more performance for this new architecture would be a reasonable expectation, I'm not sure. I would be happy with 10%, HDMI 2.1 and not charging much more than what everyone used to charge for previous generations releases just because nvidia now does because of the RTX meme.
Last edited:
Just a quick thought/ question. One of the benefits of RT is supposed to be development cost in art asset creation. Yes? If Anaconda has RT, but Lockhart does not, then this benefit would be lost, yes?
(Apologies for the post from a couple weeks ago - Didn't even know baseless-next-generation-rumors-with-no-technical-merits-spawn existed.)
(Apologies for the post from a couple weeks ago - Didn't even know baseless-next-generation-rumors-with-no-technical-merits-spawn existed.)
Ike Turner
Veteran
Not really.. if at all... Only in the head of people who don't know much about asset/game development. It's one more tool (a good one) in the artist/dev toolbox.
That's mostly for offline rendering. Approximating correct lighting through rasterisation involved lots of hacks and trial-and-error. Once a scene was set up with correct shadows and lighting, and then needed changing, the lighting work would need to be revisit to fix errors. Raytracing purifies the lighting process, so the artists just create the scene and the renderer creates optically correct light and shadows, eliminating the scene-creation overheads such that, even though tracing is a lot slower than rasterising, it ends up a considerable net win.
In an ideal ray-tracing engine, yes, costs should be a bit reduced as you have zero light baking to do. With next-gen hybrid renderers, there will still be plenty of precomputing to bake into scenes, I expect.
Dang. Could have sworn I read something, something reputable, about real time RT reducing artist time. Just couldn't remember why. Ah well. My mistake.
I don't believe that any studio would do a drastic jump like this, it will take time before RT get's mainstrean even with consoles supporting it, graphics options will still need to exist and RT will be on present only as Special option for a while, so under high, medium, and low, there's no way devs would just ignore assets, etc...
RT could reduce development costs in many areas beyond art asset creation. Worries me that everytime one makes it easier to one side to do their job, what end's up is worse technical skills applied and overall diminished visual quality of games and worse frame times. RT in my opinion now under development will allow artistis to pré-visualize faster and with more accuracy how light will behave under different lighting conditions so a faster workflow.
My thanks to all. Just figured if there was a serious cost reduction to be had, then you would want your full lineup (high, low, cloud) to be able to use it. Otherwise any advantage would be lost. Well, one day closer to that perfect world Shifty. (I am feeling more optimistic at the moment. Give it 5 and I will be back to believing in nothing.)
Also worth noting we have no idea what RTRT actually amounts to in regards to AMD when they have it, or how it's implemented in either next gen console.
I'm sure it's improved by now, but 2080 was running RTRT at 1080p down from 4k when it was enabled.
Consoles may perform better, worse, the same, or accelerate it differently so different choices are made when using it.
Do hope MS at E3 says a bit more than just a couple top level features like Sony did.
Discussing all these rumors has been nice, but I'm pretty much over it now.
Be nice if people been sitting on specs etc until after reveal, out of consideration.
That and some base navi details from AMD E3 would go down nicely.
Last edited:
- Status
Similar threads
