50TF closed system, I'd better start saving because I'm going to want one of those. If any developer manages to lower the fps of 60 in this thing it's just going to be a matter of pure dedication and sadism. Looking at you, China.
Unreal Engine 6:
That's a lot of GHz...
Such a weak hardware. That's like only 0.05 Petaflops
No I'm not, and dual-issue has nothing to do with half-precision.
Didn't you say the clocks were already finalized though?
No, we pretty much always use FP32 when talking about Floating Point performance on gaming GPUs. You're confusing the terms. FP16 and FP64 have been around for decades, but we pretty much never use them when generally talking about graphics performance.
winjer
said, FP16 has nothing to do with what we're talking about.
Any insight on how it will be used? And is it any different relative to Nvidia?
FP32 Dual Issue, it seems. But 30TFlops is still "a lot of GHz".
You're a sign of two things:
Clocks are only final after B0 hits, which is about a year from now
The problem with current dual-issue in RDNA3/4 is that there a lot of restrictions regarding VGPR bank usage which makes it almost impossible for VOPD to be used outside of code specifically optimized for it. The new implementation in RDNA5 removes almost all restrictions which means it can be used quite easily.
Interesting. How does that compare to Nvidia? Is dual issue unique to AMD?
Not really.
Xbox lives in his head 24/7
They call them different things, but what they achieve in similar. In name, they are different, but in practice, not that much.
That sounds purty good. 3.4 ghz would be just the GPU right? The CPU in Series Consoles was roughly 3.6 ghz, that will likely go up to 4.5?Not really.
For example,
70CU × 4SIMD32 (Dual-Issue) × 2MA × 3.4Ghz = 60.92 TFlops or 30.46 TFlops without Dual-Issue.
Here, the Ryzen AI Max+ Pro 395 with 40 RDNA3.5 CUs run @ 2.9 GHz with a 120W TDP.
So Xbox Magnus clocked @ 3.4GHz with the reported 250-350W TDP is possible.
If they don't find some way to stop Sarah Bond from opening her trap, nobody.
Actually they are very different in implementation, AMD's is limited by the complexity of VGPR operand fetch while NVIDIA's is limited by the number of execution lanes.
Uncertain about the implementation, but they aim to achieve the same or something similar, unless I misunderstood. For gaming at least, but so far, I don't think many if any game did anything with that.
Difficult to estimate, but I think Magnus could be slightly faster than 5080.
Yeah, it's a cheap (in terms of silicon area) way to get more performance.
Maybe. Specified numbers doesn't have to be close to real numbers.Not really.
For example,
70CU × 4SIMD32 (Dual-Issue) × 2MA × 3.4Ghz = 60.92 TFlops or 30.46 TFlops without Dual-Issue.
Here, the Ryzen AI Max+ Pro 395 with 40 RDNA3.5 CUs run @ 2.9 GHz with a 120W TDP.
So Xbox Magnus clocked @ 3.4GHz with the reported 250-350W TDP is possible.
From the Amethyst revealed tech AFAIK the full (RC) Radiance Core coordination logic can't be accelerated efficiently on current Nvidia GPUs, only the (NRC) Neural Radiance Cache algorithm can be accelerated on the tensor cores, so when making that estimate are you just looking at relative current spec efficiency and spec uplift in a AMD to Nvidia comparison, or are you going by AMD in house expected Amethyst performance savings too on PT?
No, we pretty much always use FP32 when talking about Floating Point performance on gaming GPUs. You're confusing the terms. FP16 and FP64 have been around for decades, but we pretty much never use them when generally talking about graphics performance.
What you're thinking about is VOPD which is a number that still refers to FP32 performance, but for gaming, it has been mostly meaningless, so we just ignore it and give the "regular" FP32 number that is more insightful in terms of what to expect.
You're mixing them up because using VOPD, the FP32 throughput is doubled, similar to how FP32 performance is double that of FP16 performance.
As
Do you know if this will be full pc? As in I could change the OS and everything if I wanted?
All the information we have so far says videogame console so probably no.
Where do the pc rumors come from then
news.xbox.com
Maybe a maguns console and magnus pc then? Or completely difference specs between each system, i figured magnus was the pc since it costs 1200Xbox and AMD: Advancing the Next Generation of Gaming Together - Xbox Wire
This week, Xbox announced it is actively building its next-generation lineup across console, handheld, PC, cloud, and accessories. Find out more inside.
"This week, Xbox announced it is actively building its next-generation lineup across console, handheld, PC, cloud, and accessories. As part of this, Xbox unveiled that it has entered into a strategic, multi-year partnership with AMD to co-engineer silicon across a portfolio of devices, including future first-party consoles and cloud."
They're doing BOTH Xbox PCs and Xbox Consoles.
Consoles will 100% be more locked down than the PCs.
All the information is actually all over the place. Best guess is it is Windows based, but things like Bios access etc or dual boot are total unknowns. It may well be that the 1st party version is locked down and OEM's are the only path to Linux on Magnus for instance. But perhaps in doing so you give up back compat. We will see. Actually a really good third party device without MS margin is pretty attractive to me if it can access the console like UI/UX/cutdown OS.
Sony may clock higher than normal again.Very interesting talk you guys. Thanks Kepler as always. We have to remember he doesn't need to share anything with us, yet he's doing it
Magnus sounds incredible. Since it's in the 50-60TF range, I guess PS6 will be around 30 to 40.
Magnus is dual purpose yes. But Magnus is just the CPU SOC with the media die and NPU. Theoretically it should be able to be paired up with any of the 5 AMD GPU dies.
Notice I said "for example" right?
