Monday, January 22nd 2024
Intel 15th-Generation Arrow Lake-S Could Abandon Hyper-Threading Technology
A leaked Intel documentation we reported on a few days ago covered the Arrow Lake-S platform and some implementation details. However, there was an interesting catch in the file. The leaked document indicates that the upcoming 15th-Generation Arrow Lake desktop CPUs could lack Hyper-Threading (HT) support. The technical memo lists Arrow Lake's expected eight performance cores without any threads enabled via SMT. This aligns with previous rumors of Hyper-Threading removal. Losing Hyper-Threading could significantly impact Arrow Lake's multi-threaded application performance versus its Raptor Lake predecessors. Estimates suggest HT provides a 10-15% speedup across heavily-threaded workloads by enabling logical cores. However, for gaming, disabling HT has negligible impact and can even boost FPS in some titles. So Arrow Lake may still hit Intel's rumored 30% gaming performance targets through architectural improvements alone.
However, a replacement for the traditional HT is likely to come in the form of Rentable Units. This new approach is a response to the adoption of a hybrid core architecture, which has seen an increase in applications leveraging low-power E-cores for enhanced performance and efficiency. Rentable Units are a more efficient pseudo-multi-threaded solution that splits the first thread of incoming instructions into two partitions, assigning them to different cores based on complexity. Rentable Units will use timers and counters to measure P/E core utilization and send parts of the thread to each core for processing. This inherently requires larger cache sizes, where Arrow Lake is rumored to have 3 MB of L2 cache per core. Arrow Lake is also noted to support faster DDR5-6400 memory. But between higher clocks, more E-cores, and various core architecture updates, raw throughput metrics may not change much without Hyper-Threading.
Source:
3DCenter.org
However, a replacement for the traditional HT is likely to come in the form of Rentable Units. This new approach is a response to the adoption of a hybrid core architecture, which has seen an increase in applications leveraging low-power E-cores for enhanced performance and efficiency. Rentable Units are a more efficient pseudo-multi-threaded solution that splits the first thread of incoming instructions into two partitions, assigning them to different cores based on complexity. Rentable Units will use timers and counters to measure P/E core utilization and send parts of the thread to each core for processing. This inherently requires larger cache sizes, where Arrow Lake is rumored to have 3 MB of L2 cache per core. Arrow Lake is also noted to support faster DDR5-6400 memory. But between higher clocks, more E-cores, and various core architecture updates, raw throughput metrics may not change much without Hyper-Threading.
100 Comments on Intel 15th-Generation Arrow Lake-S Could Abandon Hyper-Threading Technology
76%72%if you actually tried to find numbers. Linux is closer to 4% and MacOS at just over 16%For the server market it's more like 84% for *nix like systems as seen here.
You seem to have an impoverished idea of what the market actually looks like. Next time, do some research before spouting baseless claims.
If you're going to challenge that, then produce some sources like I did because I can't find anything that supports your claim.
Conversely, AMD already has the "compact core" concept down with the Zen 4c, while still having considerable performance due to not needing to rewrite schedulers for. So if SMT was disabled on those, they'd still have a reasonably competitive edge. Moreso, if they tweak the design further to omit SMT in favor of more speed.
Windows is terrible for audio work the dpc latency is way too high, the OS is too bloated and way too janky. Linux is far lower latency on even the same systems.
My laptop (specs in sig) has iffy bitwig performance in windows compared to Linux. Same projects even.
Most creatives are on mac or linux cuz microsoft won't get thier shit together and stop making it so bloated and laggy!
oh and another thing to add about intel.....
When I was hunting for a dell latitude or precision laptop I noticed tons of 4 core only systems and these were 7th and 8th and 9th gen i5/7s. So they've been nudging hyperthreading away for a while now, Same with Lenovo too... Also a ton of xeon processors are 4c or 8c only too... no HT
A 16 core CPU with no HT would be good for everything. HT in some cases is only to make a lower core CPU seem better as the avg non tech competent user will not understand what HT is and just see the numbers.
Now for most home users there is really no need to have a CPU with more than 16 cores(no HT) for gaming or most other general home use requirements. If you need something for any work or productivity related use then you should be using a more productivity aimed CPU/PC with a xeon or threadripper.
Unfortunately, it doesn't seem to be a top priority.
For reference, with Alder Lake arriving November 2021, and W680 launching Q1 2022, the first motherboards still weren't available by early March 2022 (Supermicro and ASRock), and the excellent Asus Pro WS W680-ACE was only announced November 7th 2022, that's over 1 year after the CPUs launched. :(Well for gaming and specific applications, sure.
But as a desktop, Linux isn't lacking at all. Linux is much more responsive, stable and not to mention configurable. For a development setup it's outstanding. After over 16 years of doing most of my development on Linux, having to use Windows at work is a huge step backwards.
Windows is certainly dominating the consumer market, but not so much the enterprise market. I don't think there is a precise metric for this, and those web traffic analyzers most certainly under-sample enterprises heavily, many of which can't even browse the web freely, or are hidden behind proxys etc. But given the fact that brands like HP, Lenovo and even Dell have offered more and more Linux options, so clearly there is demand. I wouldn't be surprised if the real market share for Linux in the enterprise desktop space is >10%, and Mac even higher. Especially workstations for CAD/modelling, simulations, sysadmin and increasingly development Linux is a serious contender.
From time to time I still hear about companies where thousands of computers got infected from someone like a secretary opening an email, I think we all know what kind of OS they are running… :rolleyes:
and the rest of thier systems like 98% is ubuntu certified to run linux so you know everything works, my dells are all running linux nicely.same for me with audio production! Linux has way lower latency and is way more responsive and the audio is definitely better too...
I go the other way, gnome classic with as little bloat as possible to make everything snappy.
But I was actually thinking much wider in terms of customization, as Linux allows you to customize virtually everything; to fit your workflow.
People haven't really tried Linux until they've embraced the terminal, preferably a dropdown one (and possibly a terminal multiplexer), and all kinds of goodies to speed up their workflow. This is especially relevant for developers and sysadmins.
I've trained a bunch of developers, both fresh and "seniors", in using Linux, terminal, git, grep, etc., and those who dive into it really get super efficient, even though they all end up with their own "quirky" workflow. (I of course use a lot of GUI tools to, but the terminal is a large part of the workflow.)
It's fascinating to see how much faster and efficient people get when they get out of the typical "Visual Studio" or "Eclipse" way of doing things through heavy slow GUIs. I try to use whatever gives me the control I want and the information I want, so I can stay in "the zone" and not be distracted. :cool:For sure, not to mention faster IO in general. What annoys me is the typing lag on Windows, I have to slow down to avoid a lot of typos. It even sometimes enters the characters in the wrong order if I do it fast enough.
-----
Luckily, Intel systems have generally worked well at launch with Linux. I know Bulldozer, Zen 1 and Zen 2 had some issues, but I haven't seen any on my Zen 3. Granted I didn't buy that at launch, but I haven't heard anything regarding Zen 3 and 4 like with the previous iterations at launch.
it's so easy to do that too...
had someone from another forum say intel is partaking in shrinkflation for CPUS.... that's a thought,
It also has more execution pipelines, period. Haswell (and probably Skylake as well) has three scheduler ports for ALU & FPU operations. Each of those ports has dedicated pathways for different instructions.
On Haswell, the unified scheduler can actually hold 60 entries which are pending execution (waiting for data), Skylake upgraded that to 97 entries. Ryzen holds 14 dedicated entries per ALU/AGU pipeline. Floating point is treated like a co-processor in Ryzen and is largely decoupled via a large 96-entry (IIRC) scheduler feeding four flexible pipelines. Ryzen can have 180 entries in the scheduler queues - almost double Skylake. This was clearly done to mitigate Intel's patented unified scheduler advantage (which allows getting results back to dependent operations very quickly) while allowing the FPU to work freely.
Sometimes this situation benefits Intel, sometimes it benefits Ryzen. For Intel, if code is running heavily on a small set of instructions (with relevant memory ops), then you might have 60~97 ops in the scheduler all pertaining to the same task, each finding its way into an execution pipeline efficiently as data is found (a definite IPC advantage in these scenarios is seen, but these are relatively rare situations in natural code - this happens with processor-specific optimizations). On Ryzen, you have to have pipelines with the necessary support and use that specific pipeline's scheduling queue, which can only hold 14 entries. This is why Ryzen's ALUs and AGUs are made to be extremely flexible while Intel's pipelines are very purpose-built.
Ryzen can do four adds, shifts, LEAs, and many many other common operations in each and every cycle while Intel's best can only do three - and frequently fewer. Intel focuses on getting into and out of the core - a full half of Skylake's scheduler ports are dedicated for memory operations, whereas AMD only has two pipelines dedicated for memory operations.
www.anandtech.com/show/10591/amd-zen-microarchiture-part-2-extracting-instructionlevel-parallelism/6
The decoder, the portion of the CPU that performs instruction scheduling, cares about the SMT vs Hyperthread vs (whatever) details.
Note that POWER9 is SMT-4 (4-threads per core) and SMT-8 (8-threads per core), depending on model. GPUs tend to be extremely SMT as well. The gist is that memory is so slow today that any memory-throttled multithreaded application will prefer SMT (ie: let all those pipelines and execution units operate on a 2nd thread entirely, while waiting on RAM).