Just Some Other Guy
kiwifarms.net
- Joined
- Mar 25, 2018
Three times, to be correct. 13th gen has stability issues too.
3 gens of bs from Intel in a row. Amazing.
GPUs & CPUs & Enthusiast hardware: Questions, Discussion and fanboy slap-fights - Nvidia & AMD & Intel - Separe but Equal. Intel rides in the back of the bus.
- Thread starter Smaug's Smokey Hole
- Start date
3 gens of bs from Intel in a row. Amazing.
- Joined
- Mar 30, 2023
Not the first time that it happened. Literally the whole thing that pushed Apple to in-house CPUs was Intel practically pushing QA onto it's customers.
- Joined
- Aug 28, 2019
I don't think the NPU or iGPU distracted from the CPU development in any meaningful way. They are on their own tiles, with either the die or the IP blocks plucked straight from Meteor Lake. If anything they did even less work on those because it's mostly a copy+paste. This is an approach they can take with all future tile-based microarchitectures. Teams can work on their tiles, and if a new one isn't ready, an old one can be used.
This is their first use of chiplets/tiles instead of monolithic dies on the desktop, which could explain some of the findings. Arrow Lake is fine, but not attention-grabbing. The rumored 8+32 Arrow Lake Refresh is cancelled, there's no X3D-like Adamantine cache anywhere to be seen, and there's no Royal Core (which may be cancelled), although they did cut out hyperthreading. The massively improved E-cores are doing a lot of the heavy lifting in the absence of hyperthreading, and an E-core-only Arrow Lake-N would be cool to see.
So far it's only broken in the way that Zen 5 was: unoptimized and buggy at launch, with Winblows likely to blame. 13th/14th gen instability was a result of high voltages.
- Joined
- Dec 7, 2020
Why do people even follow him? All of his videos are awful clickbait and he makes literal 30 minute productions out of just unboxing something.
TheRaptorJesus
kiwifarms.net
- Joined
- Mar 14, 2021
Somehow he got popular for being "good" at watercooling when nobody else on youtube was doing it. All his builds are terrible and not at all serviceable nor does he use proper monitoring. In #currentyear he still just blasts fans based on GPU/CPU temp instead of liquid, he doesn't monitor liquid temp at all. He was also a huge EK shill.
Just Some Other Guy
kiwifarms.net
- Joined
- Mar 25, 2018
EK shilling and not measuring liquid delta T? Gross.
- Joined
- Nov 15, 2021
I don't mean it was a "distraction." I mean you've got a certain power & transistor budget for a processor, and the decision was made somewhere that little of it was going to be spent on more, bigger cores. We've been deep in the realm of diminishing returns on both fronts. Bigger & bigger branch predictors buy you less and less. More and more cores benefit fewer and fewer applications. I think everyone is flailing a little, trying to figure out what the next differentiator is going to be. Intel seems to be going hard after "AI PC," maybe in part because they missed the boat so badly on AI server hardware. Gaudi isn't selling at all.
TheRaptorJesus
kiwifarms.net
- Joined
- Mar 14, 2021
- Joined
- Aug 28, 2019
Now that it's not monolithic, increasing the iGPU or NPU doesn't affect the yields of the CPU. It just increases overall die area and cost a little. It should be cheaper than an equivalent monolithic die. They also mixed nodes, saving the best node (TSMC N3B) for the CPU tile while using inferior N5/N6 nodes for everything else.
The 13 TOPS NPU is using barely any power or die area, but in subsequent generations they will make it relatively larger to be Copilot+ ready and beyond (I think 100 TOPS is already on roadmaps). Lisa Su joked on stage about how much die area had to go to theirs in Strix Point. I don't think the iGPU is using much area or power either, think a single watt when idle compared to hundreds overall. Die shots are already available.
Theoretically, they could ditch the iGPU and NPU (instead of disabling them) and add more or larger cores. But they have included an iGPU since 2010, it's nothing new.
BTW, maybe the "F" models can retain Quick Sync and decode video now? It should have been disaggregated from the iGPU.
What's actually wrong with Arrow Lake? They ditched hyperthreading, which they claimed was good for the power and transistor budget. Never mind that AMD is sticking with it for both core types, while giving AVX-512 to everyone. Their IPC projections may have missed the mark, especially for the P-cores. They are being conservative with voltages, although that's probably a good thing. The CPU design teams don't get to collaborate with the fab on this one, since everything but the base tile is made by TSMC.
Maybe Intel just sucks and the internal politics has driven good employees away.
- Joined
- Nov 15, 2021
No, but larger tiles still increase the package size and overall power consumption. The 8+16 compute tile isn't taking up a trivial amount of space or drawing a trivial amount of power. A 25% or even 50% increase in compute would mean either a substantially larger, hotter package (possibly two compute tiles, or a more expensive compute tile), or cutting something that's already there.
It wouldn't do anything; the heat still has to go from the plate to outside somehow, and the thin surface at the edge of this plate wouldn't allow much heat transfer. There's ongoing research in using microfluidics to cool stacked dies.
- Joined
- Jul 18, 2019
What's most impressive about Arrow Lake is how much more performant the e-cores have become - allegedly Arrow Lake Skymont e-cores are supposed to be comparable in performance to Golden Cove p-cores from 2021.
Obviously the launch has its problems but I could very quickly see the core design path that Intel is on evolving into an absolute beast for heavily multithreaded workloads in the next generation or two.
Obviously the launch has its problems but I could very quickly see the core design path that Intel is on evolving into an absolute beast for heavily multithreaded workloads in the next generation or two.
- Joined
- Aug 28, 2019
There were rumors of an 8+32 Arrow Lake Refresh coming next year, then rumors of it being cancelled, and recently rumors of Intel ditching E-cores in the far future for some kind of transforming core reminiscent of Royal Core (hopefully not Bulldozer).
Intel should put 8-core Skymont in its own die (Arrow Lake-N) and sell that. Maybe they can make it at their own fabs for a change.
- Joined
- Dec 31, 2019
indomitable snowman
Can't stop, won't stop (unless it gets above 32°F)
True & Honest Fan
kiwifarms.net
- Joined
- Dec 13, 2022
The EU annulled the 1 billion Euro fine levied against Intel for their bribing of OEMs to not use AMD.
This private information is unavailable to guests due to policies enforced by third-parties.
- Joined
- Aug 28, 2019
Battlemage Xe2 appears in Intel Media Drivers, video encoding yet to be added
Late and gay, but it's presumably still coming in some capacity in 2025.
Pattern Juggler
kiwifarms.net
- Joined
- Apr 11, 2024
The new offering from Intel and AMD both being 'meh' at best is interesting - is it just that node scaling isn't yielding results anymore, are there no more architecture improvements to be made, or have both companies been infected by the incompetent?
- Joined
- Aug 28, 2019
The AMD cope: Zen 4 (desktop, not mobile) was on TSMC N5. Zen 5 client may have been intended to be on a TSMC N3 node years ago, but the node was delayed at TSMC and a decision was made to switch products to N4X. Zen 5c used in Epyc Turin Dense (coming later) is on N3E. Big changes that could have been in Zen 5 have probably been shifted over to Zen 6. While Zen 5 does make a lot of changes under the hood, they tend to be more useful for enterprise workloads (particularly AVX-512) than consoomers. Zen 5 is laying the groundwork for bigger change in the future. For starters, Zen 6 may not be using the familiar chiplet layout that has been copied since Zen 2, and could introduce something like the "Infinity Links" from RDNA3 for much higher bandwidth than Infinity Fabric.
- Joined
- Jul 18, 2019
Both Zen 5 and Arrow Lake seem to have delivered on their promises for everyone except gamers. Zen 5 is quite strong in productivity workloads on Linux and really only seems to be held back on desktop by a weak I/O die and Windows weirdness - the enterprise counterpart Turin chips with a beefier I/O die seem to be absolute beasts. Arrow Lake is near Raptor Lake performance at lower power draw and support for faster memory, along with the central design being much more expandable in the future thanks to a move to tiles.
Gaming workloads, for a variety of reasons, don't really scale nicely like everything else and even if we weren't facing pressure from semiconductor production scaling problems we shouldn't really expect significant performance gains in gaming every gen.
Architectural improvements are being made right now but it's not really a linear process. Historical example: There was a period where Intel was absolutely certain the future was going to be pushing netburst to ever-higher clockspeeds with ever-deeper pipelines but that design didn't pan out and instead they shifted to Core 2. People remember core 2 winning out but they memory hole how underwhelming it was initially, and that's kinda where we're at right now - completely new architectures from both companies that are underwhelming compared to what they're laterally replacing.
Of course this could also all be cope. Maybe there is no actual additional performance to wring out. Maybe x86 is dead. A day will eventually come when this is true.
Just Some Other Guy
kiwifarms.net
- Joined
- Mar 25, 2018