poke01
Diamond Member
- Mar 8, 2022
- 4,395
- 5,718
- 106
This iGPU branding seems incoherent
MTL-H : Arc Graphics(in Dual ch. config), (Xe-LPG)
LNL: Arc 140V/130V, (Xe2)
ARL-H: Arc 140T/130T, (Xe-LPG+)
An option could have been to continue similarly to ARL-H for PTL-H with something like the following:
12Xe3: Arc 240T ; 10Xe3: Arc 230T ; 4Xe3: Arc 210T
Or some other thing.
But some marketing felt the need to "brand/indicate" this as a separate thing probably to imply "discrete level" and somehow we would get a somewhat confusing "Arc B- series" for PTL. Might not matter much for general public though.
MTL-H : Arc Graphics(in Dual ch. config), (Xe-LPG)
LNL: Arc 140V/130V, (Xe2)
ARL-H: Arc 140T/130T, (Xe-LPG+)
An option could have been to continue similarly to ARL-H for PTL-H with something like the following:
12Xe3: Arc 240T ; 10Xe3: Arc 230T ; 4Xe3: Arc 210T
Or some other thing.
But some marketing felt the need to "brand/indicate" this as a separate thing probably to imply "discrete level" and somehow we would get a somewhat confusing "Arc B- series" for PTL. Might not matter much for general public though.
Looks analogous in some ways, yes.
I agree, 2.5 would have been better when Xe3p with a more substantially different architecture didn't get a full name change.
I don't think iGPU names matter.
Almost no one is going around comparing the 890M to the 8060S or the 140V to 140T.
The names people are using to compare GPU performances are Z2 Extreme, Ryzen AI MAX 395, or when they get a bit more technical it's Strix Point vs. Lunar Lake vs. Arrow Lake.
I bet most of the people commenting on Strix Halo's gaming performance don't even know how its iGPU is called.
Yeah, like I said it doesn't matter to general public much at least in short run. Still having a more coherent stuff is better when you are trying to make a name and also among the enthusiasts, but yeah performance and other parameters are more important than some naming.
Naming shenanigans:
Cause yea Xe3 is significant. 60% better with 50% more compute means 20-30% better per xe cores. For Xe3P to be more significant, it will have to deliver more than 20-30% better per core. I doubt that's happening.
The last time I asked him, he didn’t know about the Xe3p changes. Do you trust him more than Petersen? Xe3 indeed looks more like Xe2.5 based on the changes Intel showcased. Of course, it can still result in bigger gains in some workloads where Xe2 was bottlenecked, but overall the architectural changes aren’t that big on Xe3. And obviously Intel agrees with that, otherwise they would have branded PTL iGPU as C-series and not just Xe3p.
Neither and both, but the reasons are more complicated.
Engineering wise Xe3P might be a bigger deal, but it doesn't mean that it'll turn out into greater performance gains. Regardless of what you think, 20-30% gain per Xe core is substantial, especially because GPU usually doesn't rely on uarch gains, because focusing elsewhere such as uncore, caches bring you the gains, plus outright more cores.
Even if the Xe3P changes require more changes, it doesn't mean it'll result in more gains, because Xe3's changes could have extracted lot of the gains, thus it's harder to extract more. 20-30% gains over B580 is already in comparable RDNA4 territory, or really good. End users don't care about tech details really. In the end performance/watt, performance/$ matters. I don't care if your engine is a warp drive, if it results in 10% gain over a rocket engine 1.0. If a rocket engine 2.0 offers 30% gains, then that is a bigger gain for me.
Intel hasn't been known for being forthright for over a decade now. Also naming is partly about hiding details from the consumer to persuade you to buy higher end. That's why 3/5/7/9 naming exists. You'd think a 15W CPU should be named 3, but Intel calls it 9, because they know people pay premium for ultraportables and low power. Intel's revenues were stagnant for quite a few years until the Core ix marketing scheme, and the first year their revenue increased by 20%. Sandy Bridge increased revenue by 20% again. Core 2 in all it's glory didn't do that. Because in 2006 it already was in replacement mode for computers, so a 1.5x faster chip at 30% lower power is nice, but that's for when the current one breaks.
I wonder if G31 is Xe3 B Series just like Panther Lake has.
Really wonder what this means, Shader Model 7.0? Feature additions to make it equal to RTX 60 Rubin & RDNA 5?
Intel just make every pieces of G31 into ARC Pro, presumably ARC Pro B70 32GB.That's one of the movement of latest ARC management. So yep, carelessly left some traces or evidences in kernel.
For Xe3 iGPUs unsurprisingly there's no FP8 compute in hardware, only dequantization support.
No, G31 is Xe2-HPG(same as in B580).
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 23K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter