The 2 big reasons for this single core thing. 1. E core and P core is having trouble talking to each other. And as a single thread it behaves "better". 2. They are more stable at higher frequency with a single thread. Because, Marketing.
Wrong. It's performance per watt that they were trying to optimize. Not overall all core performance. This CPU is meant for thin and light laptops. Raw performance matters a lot less than performance per watt in this application.
i am also on my 9900k that has been overclocked to 5.0ghz for nearly 6 years now. it still gets the job done. dropping hyperthreading makes sense to me, because they could just add more of the e-cores that are rumored to be as powerful as raptor lake. i can also see the ai portion of the cpu being utilized with the direction that tech is going. it would be one of those "i would rather have it and not need it, than need it and not have it" things. if the performance of arrow lake or zen5 x3d blows my socks off, i may upgrade.
I watched, by chance, a lecture about wasted processing power due to optimizations like hyperthreading. Did not understand it very well but I'll link the lecture here. th-cam.com/video/wGSSUSeaLgA/w-d-xo.html The short of it: there's the von Neuman bottleneck (DRAM is much slower than CPU) and things like multithreading and prefetching were created. Researchers - like the lecturer on that video - have shown all this adds nothing in term of computational power. All they do is waste electricity and silcone area. BTW, the lecturer is talking about Servers and, you know, that's the big market. We're just beta testers. EDIT: the lecture is from C++Now 2024. Energy efficiency has nothing to do with laptops - it's all about servers. We, of course, love that. Intel deserves applauses for how efficient they've done their CPUs and deserves a stern talk for how long they took to make it.
Lunar Lake looks very interesting. I'm purely a Linux user, and have been on AMD for about 5 years now (and have been very happy with AMD, have had laptops with Ryzen 7 4800H, Ryzen 7 6800H, Ryzen 7 8840U and one with a Ryzen 7 5850U). If the MacBooks with the M series chips were more "open" and Linux friendly, I would have purchased one in a heartbeat (great performance while sipping battery power). Really hoping Lunar Lake lives up to the hype.
TIMESTAMPS BELOW! Thanks to @Intel for providing the opportunity and to @ASUS , @MSI and @Dell for their hands-on experiences that aided this video. If you found my videos useful you can always buy me a coffee! 👉www.buymeacoffee.com/crazytechlab Check out my Amazon shop with all the best products! 👉 www.amazon.com/shop/antonyleather www.paypal.com/donate/?hosted_button_id=P8PHVM6V3UAQS Subscribe: bit.ly/2QYAZK9 Instagram: instagram.com/crazytechlab Facebook: facebook.com/CrazyTechLab Twitter: twitter.com/antonyleather Timestamps 0:00 intro 1:17 What is hyper-threading? 4:30 What benefits does Lunar Lake if you have an old laptop? 6:28 Lunar Lake's gaming performance 7:00 Should you upgrade from a Meteor Lake Core Ultra Series 1 laptop? 8:46 Why is Lunar Lake launching so soon after Meteor Lake? 10:10 What's the difference between Meteor Lake and Lunar Lake? 11:31 What's new with Intel Evo? 15:32 Why is Intel removing hyper threading?
About time. Hyperthreading only made sense when you only had one or two cores. In that case if you had 3 or 4 or 5 threads that wanted to run, you could get a little more work done with a few "half cores" that could do something while the other half was momentarily stalled. These days, you still typically only have 3 or 4 or 5 threads, but 6 or 8 or 12 cores, so there's just no need for those half cores thrown in the mix.
@clint_254 No, hyperthreading does not help the latency of a single thread. It takes advantage of small stalls in the pipeline of that single thread where another thread might be able to get some work done. It will LOWER the throughput and INCREASE the latency of the first thread in the hopes that the combined throughput of the two will be higher. It is only of any benefit ( to throughput, not latency ) when you have more threads that you want to run than you have cores on which to run them.
@clint_254 How do you figure that SMT does anything to lower the latency for a single thread? Do you not understand that what SMT does is take hardware resources *away* from one thread, and give them to another? That can NEVER help the first thread. The ideal is that it negatively impacts the first thread little to none, while benefiting another thread.
@clint_254 Are you referring to this? "The cores with hyperthreading-support are SP-enabled. SP technology leverages the idle hardware threads to carry out long-range cache prefetches which offsets performance degradation due to cache misses.". If so, you seem to be misunderstanding. The whole reason one half of an SMT core core goes idle is because it is waiting for a fetch from RAM. Latency means the time that thread spends waiting for that RAM access. During that time, the core can run another thread that that is trying to do some math on data that is already in the L1 cache ( or registers ). Even in that case, that does nothing to lower the latency of the first thread, and you keep saying you are talking about a single threaded load, which means when that one thread stalls waiting for ram, there is no other thread for the other half of the SMT core do any work on either. You just have two idle SMT cores instead of one. Arguing that SMT helps a single thread makes even less sense than arguing that having more cores in general helps a single threaded load; If there is only one thread, then all of the other cores ( SMT or not ) won't have anything to do and so will be totally wasted.
@clint_254 "No, an SMT thread goes idle when there is no micro-op for it to process. In such cases, the resources are dynamically dedicated to servicing the single thread that has micro-ops to process.". You forgot a word there. When ONE thread can't make any progress, THE OTHER thread is switched to. If you don't *have* another thread, then there is nothing else to switch to. How are you not getting this?
I kinda understand removal of multithreadng from High Performance cores, removal of IPC penalty and sync issues. But E-cores? Stupid. If something is easily parallelizable (faq, what a word) even e-cores can contribute. If not, they will be able to handle many more lightweight / occassionally acting backgroudn processes.
Hyper-threading only works if the core is bottlenecked and has poor scheduling and branch prediction. Meteor Lake has such as good core that Hyper-Threading only gained 4% integer and 17% floating point performance from Hyper-Threading and Lunar/ArrowLake's cores are even better and Hyper-Threading would result in 3-10% performance at best.
serious and good article. I will buy some more intel stock... and new desktop (may be first of year 2025, I thinks). I have may notebook but my desktop too old, that 17% over clocked.6700K
Dropping hyperthreading makes sense if you dont have the manpower and resources to design both laptop and desktop chips. It makes no sense if you're trying to make powerful industry leading desktop chips.
I understand dropping HT on mobile 100%, On Arrow Lake sure its just 8 cores HT but that made up a lot of the MT perf. I personally don't care too much about efficiency as long as its not absard and dangerous/annoying for coolers to tame. dropping HT will hurt their gains but it'll help normalize what was becoming a meme, I hope Arrow OLake has some head room for me to get at least an extra 100-300mhz mentaining at least 5.8ghz on the P-cores. I'm also super curious to see how the E-cores scales with power.
Meteor Lake got 4% integer performance gains and 17% floating point performance gains from HT, Lunar/ArrowLake cores are even less bottlenecked and with even better scheduling and branch prediction, so at best you'd gain 3-10% performance with Hyper-threading while increasing power and die size area by 15-20%.
@@rattlehead999the few extra percentages do add up but I guess it wasn’t worth it. If they were just disabling it in software I’d be confused but they apparently made two separate version of lion cove with and without the HT logic. I’m fine with it, just make faster cores and memory ecosystem
@@rattlehead999 Yeah, in their opinion atleast. Its still a pretty good P-core but latency not portionately improved and we got some regressions with the focus on eco. It was gonna happen regardless, still thinking of a 285K build no less.
They were working on royal core project (Jim Keller) which uses rentable core idea … a p core can become multiple e cores if required Panther Lake might see this come into fruition but recent cost cuts might impact this
I smell some marketing bs in certain answers. “Crap, we couldn’t get Hyperthreading to work in this crappy new CPU… let’s tell people this was planned and is actually a good thing!” Remember kids - Intel knew their 8th Gen CPUs were all affected by the Spectre/Meltdown Hardware bugs, yet they still chose to sell them anyway. Intel is NOT acting in favor of their customers! I wouldn’t be shocked if in a few years reports surface that this Lunar Lake was actually an unfinished prototype Gen, that Intel had to rush to not loose market share to Qualcomms Elite CPUs and AMDs power efficient CPUs.
Never say "forever". Intel could release a CPU geared towards WS or server and they need to boost core counts a LOT AND run p cores and they would once again have to maximize die space which is why HT or SMT-2 came about in the first place.
No the lower multi-threaded performance on lunar lake is mainly due to the low power consumption. You have 8W, 17W and 30W to feed 8 cores, NPU, massive iGPU and a lot of cache.
IMO Hyper threading was removed because memory access is so fast that there is no time to run (hyper) multiple threads in one cpu. Its a logical step for more efficient instruction execution.
HT was working around memory latency, not bandwidth. And latency hasn't improved significantly. But maybe larger caches and better data prefetchers are good enough to mitigate it.
According to Wccftech, latency has improved: "One of the key reasons why latency has improved so much happens to be aligned with the overall bandwidth uplift which is up to 2.8x for the Lunar Lake low-power cores versus Meteor Lake. Intel has also shared the cross-cluster latency coherency chart, showcasing the Lunar Lake P-Cores with an average latency of around 25ns and 55ns for the LP-E cores. This is a 3x improvement over the Meteor Lake CPUs."
It was removed because when running HT you don't get 2X the performance of a single thread running on a core. Memory access has nothing to do with it. It's faster for 2 single thread cores to run those 2 threads than it is one core running HT. And this is because there isn't duplication of the instruction logic. So why now? Because die space is now sufficiently small (transistor density) with the current process nodes, especially when e cores are used more often. HT or SMT has ALWAYS been about maximizing die space, giving more thread processing capability with a core that's slightly bigger. It has NEVER been about a core being more performant. You could say it's made cores more efficient but it depends on the metrics you use to measure efficiency. If anything faster memory access would lend itself to SMT-2, because the cores can get the instructions/data faster which is more important if you're trying to double the thread count on a single core.
On Meteor Lake hyper-threading in the best case scenario in synthetic benchmarks gave 4% higher Integer performance and 17% higher floating point performance, Lunar/ArrowLake cores are even less bottlenecked, with better schedulers and branch prediction, so at best you are looking at 10% better performance with Hyper-Threading when the thread is at 100%, while adding 15-25% power draw and area size. For reference Zen4 gains 21% integer performance and 28% floating point performance from hyper-threading. The core is more bottlenecked, even though the schedulers and branch prediction is as good as intel's. For another reference the 12th gen intel CPUs gain about 18% integer performance and 26% floating point performance from hyper-threading.
@@rattlehead999 Thanks for sharing your insights on Meteor Lake's hyper-threading performance. You're absolutely right that Intel has positioned Meteor Lake as a power-efficient platform. The addition of a virtual core, while potentially boosting performance in certain scenarios, does introduce overhead, which can offset some of the gains. Given Meteor Lake's focus on efficiency and the potential for lower power consumption, it might not be the most ideal choice for workloads that heavily rely on hyper-threading. Additionally, the tile-based architecture of Meteor Lake could potentially offer some cost advantages, making it a more attractive option for certain applications. It's interesting to compare Meteor Lake's hyper-threading benefits to those of Zen4 and 12th-gen Intel CPUs. While Zen4's gains seem more substantial, it's important to consider the specific use cases and priorities for each platform.
@@moonwatcher6594 I would wager that ArrowLake will be faster than the i9 14900k in multi-threaded tasks due to its 68% improved E-cores. And in single-core tasks it should be on par or very slightly faster, considering it's clocked much lower on the P-core => 5.5Ghz Arrow vs 6.2Ghz Raptor Lake. Lunar/ArrowLake E-cores have the same or slightly higher IPC than 14th gen P-Cores, but lower clock speeds(4.6Ghz). At the same time I expect the 285K to run at 125W with 90%+ performance of 254W 14900k.
@@rattlehead999 That's a move towards the right direction IMO because my 13900k/RTX 4090 build is literally a room heater that runs up the air-conditioning bill! Nvidia also should keep in mind that we are already planning to go "Up and Atom!" to power the AI data centers!
@@moonwatcher6594 There was an engineering standard. Desktop GPUs max out at 250W TDP(220-230W power draw) and CPUs amx out at 140W TDP(120-140W power draw). Sadly the RTX 5090 is rumored to be 600W and the 5080 400W, which is a massive increase yet again. The 5070 should be under 250W, but that won't be faster than a 4090, the 5080 is said to be 10% faster than the 4090. Otherwise I'm happy that this time Zen5 has lower power consumption and can actually be cooled, unlike Zen4/Ryzen 7000 "X" CPUs. the non-X Ryzen 7000 CPUs are excellent though.
I'm still rocking an i9 9900k all cores overclocked to 4.75ghz and it's been rock solid since day 1, I have no issues with anything speed wise. The 13900k and 14900k were an epic fail, even more so with their chips burning out and hoping nobody noticed till Wendell kicked up a fuss. Intel really have a long way to go to get back, my next CPU will be AMD when I deem the i9 9900k obsolete, but as is, it's fine, it's the GTX 1080 of CPUs. Workloads I use, docker, creative cloud (photoshop, illustrator), affinity photo & designer (adobe replacements), blender, bitwig studio, studio one 5 pro, etc.
@@MegaPixel404 I agree. But remember the 9700K didn’t have hyperthreading and it was the best 9000 series CPU in a lot of games as a result. So st the very least we could see more performance and not just a refresh like 13-14. Of course it needs to restore faith too. No argument there.
@@CrazyTechLab True, unified non core tech seems to / well used to always perform better speed wise (has been too long for me to remember how they used to perform when I had a chip with no HT). I do probably bottleneck having an RTX 4090 with my cpu however, but for running ML stable workloads with ComfyUI, A1111 and LLM studio the cpu never gets to 100% on all cores (seems it's more of a ram thing). I expect a couple of generations from now we might have a new dedicated slot for ML (AI) processing (TPU / Intel's Neural Spiking chip or other) and a very different architecture on mobos for handling ML (fake AI) workloads (DDR7+ and distance to CPU/GPU/TPU to RAM etc), which should also inherently improve gaming performance.
what will happen with older games? might they run better with disabled HT? i remember with my old ivy bridge cpu, where disablind HT lowered the performance in games
Many PC enthusiasts are bitching about core type simply because they branded as eCore They not even a developer, they don’t understand Core are doing They just don’t like get something different
I think to some extent that’s true. What Luna Lake shows is that it’s not just about power savings. It is about taking more load off the P-cores so they in turn can put effort towards games and content creation. Plus hyper-threading can actually result in lower performance in some situations. If you look at task manager and your taskbar there are an awful lot of programs running in your PC. The more of those basic programs you can run on e-cores the more you free up the P-cores. Plus no one liked the power consumption of Intels recent desktop CPUs. This could drastically improve that.
@@CrazyTechLab The main idea behind E cores on desktop isn't power efficiency, but rather performance per area, though this also leads to better efficiency (because you can run more of those cores at lower clocks while having the same performance and using less power). E cores take up less die space and IIRC you can fit 4 E cores in place of 1 P core, but in an app like cinebench those 4 E cores score roughly 50% more than a single P core. Also on Lunar Lake those E cores sit on the "low power island" so they can just run while the P cores are completely turned off. On the topic of SMT/Hyperthreading - removing it will hit multithreaded performance, but it allows Intel to get somewhat better single-thread performance. And removing hyperthreading might reduce the need for some spectre/meltdown mitigations which could also bring performance uplifts, though this is part is questionable.
Nah, they are completely right. E-cores, in their current implementation are trash. Literally worthless for anything, but multithreaded performance without manually forcing all other side apps to run on them. They arent even that efficient to begin with.
Didn't learn a thing. Just a rehash of several points discussed at the beginning and needlessly hashed out over a longer format. Same as everyone else. Anyway, some of these changes might be to the capabilities of the new fab that ASML and Intel have been working on in the last decade for which no other fabs such as TSMC or Samsung purchased.
Only intel would remove the only thing that made there x86 platform back compatible , legacy software rakes a massive hit , equivalent to removing 2 performance cores , e cores just rebranded crleron die modded to one chip no matter how intel spin this your getting worse performing chip than last generation and half the chip it should be , intel see stick only giving us 4 core systems while the compition has moved on, this does nothing but bad things to there stock ,and sales figures
@@LouisDuran Well tell that to the software engineers. Plus it's not always that AVX would be better on the GPU if there is some boolean logic tied to it(dependency).
I am not so sure what is the point of Intel's emphasis on gaming. In case you have not noticed, many gaming companies are closing. Ubisoft is about going bankrupt. Twitch has laid off employees, Microsoft laid off developers at the gaming division. The gaming era might be reaching a plateau, if not a decline. NVDA is actaully focusing more on AI chips than GPUs for gaming and AMD stated the same.
Sorry Intel but I don't care about my desktop power consumption or AI. Still using the 9900k OCed to 5.1Ghz pulling 250W under full load but it's never at full load so most of the time it seeps under 20W while surfing the net or watching TH-cam, I don't work on my PC or do content creation so for my needs the 9900k is still great and I have no plans to upgrade any time soon.
seems like ya getting paid by some cpu manuf... but whoecer has q 10th gen cpu , they have no reason whatsoever to upgrade to a newest cpu , they are just fine , ...my opinion
You’re entitled to your opinion of course. But if you pair a modern graphics card with a 10th gen it will be much slower in games than a 14th gen CPU or Ryzen 7000. That’s a fact. And this video is entirely unsponsored 😊
@@Scudmaster11 hyperthreading is 95% stupid marketing and 5% (in particular apps) performance gain. It represents the lack of innovation of intel in '10s which ended in '20s fall...
@@Scudmaster11 when you find out using less is not an good idea it makes gpu slower a bit as it not speedly in task to deal with task as the cpu get slower the list is gets need to be more cores over it to keep it close
Hyperthreading should of stayed... mobile performance... really???.. its faster at the expense of some extra power consumption.... PCs arent a mobile phone... and is why ARM doesnt belong in a PC because its not a PC then
Who needs fxxxing e-cores on a desktop? There is no battery to save. Intel should better learn to stand the heat and learn how to propper glue a good cpu together. This kind of woreshipping "single core" peformance is good for times of c64 or MSDos.
To be fair though when they’ve removed hyper threading in the past for example on the Core i7 9700k it actually had a net benefit in gaming performance and in some games it was faster than the 9900k as a result. If e-cores enable that without a drop in multi thread performance it could be a good thing, irrespective of whether it lowers power consumption or not.
@@EnochGitongaKimathi You are a kid right? I've heard somewhere that in 2024 majority of internet users globally are children and tenagers. Limited screen space is mobile's ultimate limitation.
Lack of hyperthreading is not 'Apple concept' but consequence and limitation of ARM architecture. As cor integrated memory - again, not Apple concept and stupid one at that. Just look at yields and Apple RAM pricing.
The 2 big reasons for this single core thing.
1. E core and P core is having trouble talking to each other. And as a single thread it behaves "better".
2. They are more stable at higher frequency with a single thread. Because, Marketing.
Wrong. It's performance per watt that they were trying to optimize. Not overall all core performance. This CPU is meant for thin and light laptops. Raw performance matters a lot less than performance per watt in this application.
They are no longer making hyper threaded cpu's. When one engineer was pressed. It was for the reasons I stated above.
then pack more cores, guess what i will stick to muh 13900k
i am also on my 9900k that has been overclocked to 5.0ghz for nearly 6 years now. it still gets the job done. dropping hyperthreading makes sense to me, because they could just add more of the e-cores that are rumored to be as powerful as raptor lake. i can also see the ai portion of the cpu being utilized with the direction that tech is going. it would be one of those "i would rather have it and not need it, than need it and not have it" things. if the performance of arrow lake or zen5 x3d blows my socks off, i may upgrade.
In 2 or 3 years time the NPU will get used a lot more than you expect. Microsoft will build it into the OS and office and teams.
I watched, by chance, a lecture about wasted processing power due to optimizations like hyperthreading. Did not understand it very well but I'll link the lecture here.
th-cam.com/video/wGSSUSeaLgA/w-d-xo.html
The short of it: there's the von Neuman bottleneck (DRAM is much slower than CPU) and things like multithreading and prefetching were created. Researchers - like the lecturer on that video - have shown all this adds nothing in term of computational power. All they do is waste electricity and silcone area.
BTW, the lecturer is talking about Servers and, you know, that's the big market. We're just beta testers.
EDIT: the lecture is from C++Now 2024. Energy efficiency has nothing to do with laptops - it's all about servers. We, of course, love that. Intel deserves applauses for how efficient they've done their CPUs and deserves a stern talk for how long they took to make it.
Lunar Lake looks very interesting. I'm purely a Linux user, and have been on AMD for about 5 years now (and have been very happy with AMD, have had laptops with Ryzen 7 4800H, Ryzen 7 6800H, Ryzen 7 8840U and one with a Ryzen 7 5850U). If the MacBooks with the M series chips were more "open" and Linux friendly, I would have purchased one in a heartbeat (great performance while sipping battery power). Really hoping Lunar Lake lives up to the hype.
TIMESTAMPS BELOW!
Thanks to @Intel for providing the opportunity and to @ASUS , @MSI and @Dell for their hands-on experiences that aided this video.
If you found my videos useful you can always buy me a coffee! 👉www.buymeacoffee.com/crazytechlab
Check out my Amazon shop with all the best products! 👉 www.amazon.com/shop/antonyleather
www.paypal.com/donate/?hosted_button_id=P8PHVM6V3UAQS
Subscribe: bit.ly/2QYAZK9
Instagram: instagram.com/crazytechlab
Facebook: facebook.com/CrazyTechLab
Twitter: twitter.com/antonyleather
Timestamps
0:00 intro
1:17 What is hyper-threading?
4:30 What benefits does Lunar Lake if you have an old laptop?
6:28 Lunar Lake's gaming performance
7:00 Should you upgrade from a Meteor Lake Core Ultra Series 1 laptop?
8:46 Why is Lunar Lake launching so soon after Meteor Lake?
10:10 What's the difference between Meteor Lake and Lunar Lake?
11:31 What's new with Intel Evo?
15:32 Why is Intel removing hyper threading?
About time. Hyperthreading only made sense when you only had one or two cores. In that case if you had 3 or 4 or 5 threads that wanted to run, you could get a little more work done with a few "half cores" that could do something while the other half was momentarily stalled. These days, you still typically only have 3 or 4 or 5 threads, but 6 or 8 or 12 cores, so there's just no need for those half cores thrown in the mix.
@clint_254 No, hyperthreading does not help the latency of a single thread. It takes advantage of small stalls in the pipeline of that single thread where another thread might be able to get some work done. It will LOWER the throughput and INCREASE the latency of the first thread in the hopes that the combined throughput of the two will be higher. It is only of any benefit ( to throughput, not latency ) when you have more threads that you want to run than you have cores on which to run them.
@clint_254 Yea, and I said that SMT does not give a single threaded program lower latency or higher throughput. Was I clear enough this time?
@clint_254 How do you figure that SMT does anything to lower the latency for a single thread? Do you not understand that what SMT does is take hardware resources *away* from one thread, and give them to another? That can NEVER help the first thread. The ideal is that it negatively impacts the first thread little to none, while benefiting another thread.
@clint_254 Are you referring to this? "The cores with hyperthreading-support are SP-enabled. SP technology leverages the idle hardware threads to carry out long-range cache prefetches which offsets performance degradation due to cache misses.". If so, you seem to be misunderstanding. The whole reason one half of an SMT core core goes idle is because it is waiting for a fetch from RAM. Latency means the time that thread spends waiting for that RAM access. During that time, the core can run another thread that that is trying to do some math on data that is already in the L1 cache ( or registers ). Even in that case, that does nothing to lower the latency of the first thread, and you keep saying you are talking about a single threaded load, which means when that one thread stalls waiting for ram, there is no other thread for the other half of the SMT core do any work on either. You just have two idle SMT cores instead of one. Arguing that SMT helps a single thread makes even less sense than arguing that having more cores in general helps a single threaded load; If there is only one thread, then all of the other cores ( SMT or not ) won't have anything to do and so will be totally wasted.
@clint_254 "No, an SMT thread goes idle when there is no micro-op for it to process. In such cases, the resources are dynamically dedicated to servicing the single thread that has micro-ops to process.". You forgot a word there. When ONE thread can't make any progress, THE OTHER thread is switched to. If you don't *have* another thread, then there is nothing else to switch to. How are you not getting this?
I kinda understand removal of multithreadng from High Performance cores, removal of IPC penalty and sync issues. But E-cores? Stupid. If something is easily parallelizable (faq, what a word) even e-cores can contribute. If not, they will be able to handle many more lightweight / occassionally acting backgroudn processes.
Hyper-threading only works if the core is bottlenecked and has poor scheduling and branch prediction. Meteor Lake has such as good core that Hyper-Threading only gained 4% integer and 17% floating point performance from Hyper-Threading and Lunar/ArrowLake's cores are even better and Hyper-Threading would result in 3-10% performance at best.
serious and good article.
I will buy some more intel stock...
and new desktop (may be first of year 2025, I thinks).
I have may notebook but my desktop too old, that 17% over clocked.6700K
i stop using Intel since 2021 because of lack of threads, lack of cores and stucked in milimeters
Dropping hyperthreading makes sense if you dont have the manpower and resources to design both laptop and desktop chips. It makes no sense if you're trying to make powerful industry leading desktop chips.
I understand dropping HT on mobile 100%, On Arrow Lake sure its just 8 cores HT but that made up a lot of the MT perf. I personally don't care too much about efficiency as long as its not absard and dangerous/annoying for coolers to tame. dropping HT will hurt their gains but it'll help normalize what was becoming a meme, I hope Arrow OLake has some head room for me to get at least an extra 100-300mhz mentaining at least 5.8ghz on the P-cores. I'm also super curious to see how the E-cores scales with power.
Meteor Lake got 4% integer performance gains and 17% floating point performance gains from HT, Lunar/ArrowLake cores are even less bottlenecked and with even better scheduling and branch prediction, so at best you'd gain 3-10% performance with Hyper-threading while increasing power and die size area by 15-20%.
@@rattlehead999the few extra percentages do add up but I guess it wasn’t worth it. If they were just disabling it in software I’d be confused but they apparently made two separate version of lion cove with and without the HT logic. I’m fine with it, just make faster cores and memory ecosystem
@@ItsAkile they didn't use the Lion Cove with the HT in it, it turned out to not be worth it.
@@rattlehead999 Yeah, in their opinion atleast. Its still a pretty good P-core but latency not portionately improved and we got some regressions with the focus on eco. It was gonna happen regardless, still thinking of a 285K build no less.
@@ItsAkile I'm absolutely happy that they focused on efficiency a desktop CPU should top out at 140W and a desktop GPU at 250W.
They were working on royal core project (Jim Keller) which uses rentable core idea … a p core can become multiple e cores if required
Panther Lake might see this come into fruition but recent cost cuts might impact this
I smell some marketing bs in certain answers.
“Crap, we couldn’t get Hyperthreading to work in this crappy new CPU… let’s tell people this was planned and is actually a good thing!”
Remember kids - Intel knew their 8th Gen CPUs were all affected by the Spectre/Meltdown Hardware bugs, yet they still chose to sell them anyway.
Intel is NOT acting in favor of their customers! I wouldn’t be shocked if in a few years reports surface that this Lunar Lake was actually an unfinished prototype Gen, that Intel had to rush to not loose market share to Qualcomms Elite CPUs and AMDs power efficient CPUs.
Never say "forever". Intel could release a CPU geared towards WS or server and they need to boost core counts a LOT AND run p cores and they would once again have to maximize die space which is why HT or SMT-2 came about in the first place.
Explains why they have such poor multi-threaded performance.
No the lower multi-threaded performance on lunar lake is mainly due to the low power consumption. You have 8W, 17W and 30W to feed 8 cores, NPU, massive iGPU and a lot of cache.
IMO Hyper threading was removed because memory access is so fast that there is no time to run (hyper) multiple threads in one cpu. Its a logical step for more efficient instruction execution.
Not to mention that HT was a quick hack around a problem to begin with, not a solution.
@@alexturnbackthearmy1907😅
HT was working around memory latency, not bandwidth. And latency hasn't improved significantly. But maybe larger caches and better data prefetchers are good enough to mitigate it.
According to Wccftech, latency has improved: "One of the key reasons why latency has improved so much happens to be aligned with the overall bandwidth uplift which is up to 2.8x for the Lunar Lake low-power cores versus Meteor Lake. Intel has also shared the cross-cluster latency coherency chart, showcasing the Lunar Lake P-Cores with an average latency of around 25ns and 55ns for the LP-E cores. This is a 3x improvement over the Meteor Lake CPUs."
It was removed because when running HT you don't get 2X the performance of a single thread running on a core. Memory access has nothing to do with it. It's faster for 2 single thread cores to run those 2 threads than it is one core running HT.
And this is because there isn't duplication of the instruction logic.
So why now? Because die space is now sufficiently small (transistor density) with the current process nodes, especially when e cores are used more often. HT or SMT has ALWAYS been about maximizing die space, giving more thread processing capability with a core that's slightly bigger. It has NEVER been about a core being more performant. You could say it's made cores more efficient but it depends on the metrics you use to measure efficiency.
If anything faster memory access would lend itself to SMT-2, because the cores can get the instructions/data faster which is more important if you're trying to double the thread count on a single core.
Who was it recently that turned off HT and found hardly any difference in power consumption - lol -.
Not the same architecture and there is a 15% or so difference. Also depends if the chip has a hard cap on power draw.
Intel might be gone in 2-3 years ... how odd...
"they are not always doing that" implies that they may still be doing that a lot of the times ;)
On Meteor Lake hyper-threading in the best case scenario in synthetic benchmarks gave 4% higher Integer performance and 17% higher floating point performance, Lunar/ArrowLake cores are even less bottlenecked, with better schedulers and branch prediction, so at best you are looking at 10% better performance with Hyper-Threading when the thread is at 100%, while adding 15-25% power draw and area size.
For reference Zen4 gains 21% integer performance and 28% floating point performance from hyper-threading. The core is more bottlenecked, even though the schedulers and branch prediction is as good as intel's.
For another reference the 12th gen intel CPUs gain about 18% integer performance and 26% floating point performance from hyper-threading.
@@rattlehead999 Thanks for sharing your insights on Meteor Lake's hyper-threading performance. You're absolutely right that Intel has positioned Meteor Lake as a power-efficient platform. The addition of a virtual core, while potentially boosting performance in certain scenarios, does introduce overhead, which can offset some of the gains. Given Meteor Lake's focus on efficiency and the potential for lower power consumption, it might not be the most ideal choice for workloads that heavily rely on hyper-threading.
Additionally, the tile-based architecture of Meteor Lake could potentially offer some cost advantages, making it a more attractive option for certain applications. It's interesting to compare Meteor Lake's hyper-threading benefits to those of Zen4 and 12th-gen Intel CPUs. While Zen4's gains seem more substantial, it's important to consider the specific use cases and priorities for each platform.
@@moonwatcher6594 I would wager that ArrowLake will be faster than the i9 14900k in multi-threaded tasks due to its 68% improved E-cores. And in single-core tasks it should be on par or very slightly faster, considering it's clocked much lower on the P-core => 5.5Ghz Arrow vs 6.2Ghz Raptor Lake.
Lunar/ArrowLake E-cores have the same or slightly higher IPC than 14th gen P-Cores, but lower clock speeds(4.6Ghz).
At the same time I expect the 285K to run at 125W with 90%+ performance of 254W 14900k.
@@rattlehead999 That's a move towards the right direction IMO because my 13900k/RTX 4090 build is literally a room heater that runs up the air-conditioning bill! Nvidia also should keep in mind that we are already planning to go "Up and Atom!" to power the AI data centers!
@@moonwatcher6594 There was an engineering standard. Desktop GPUs max out at 250W TDP(220-230W power draw) and CPUs amx out at 140W TDP(120-140W power draw).
Sadly the RTX 5090 is rumored to be 600W and the 5080 400W, which is a massive increase yet again.
The 5070 should be under 250W, but that won't be faster than a 4090, the 5080 is said to be 10% faster than the 4090.
Otherwise I'm happy that this time Zen5 has lower power consumption and can actually be cooled, unlike Zen4/Ryzen 7000 "X" CPUs. the non-X Ryzen 7000 CPUs are excellent though.
I'm still rocking an i9 9900k all cores overclocked to 4.75ghz and it's been rock solid since day 1, I have no issues with anything speed wise. The 13900k and 14900k were an epic fail, even more so with their chips burning out and hoping nobody noticed till Wendell kicked up a fuss. Intel really have a long way to go to get back, my next CPU will be AMD when I deem the i9 9900k obsolete, but as is, it's fine, it's the GTX 1080 of CPUs. Workloads I use, docker, creative cloud (photoshop, illustrator), affinity photo & designer (adobe replacements), blender, bitwig studio, studio one 5 pro, etc.
@@MegaPixel404 I agree. But remember the 9700K didn’t have hyperthreading and it was the best 9000 series CPU in a lot of games as a result. So st the very least we could see more performance and not just a refresh like 13-14. Of course it needs to restore faith too. No argument there.
@@CrazyTechLab True, unified non core tech seems to / well used to always perform better speed wise (has been too long for me to remember how they used to perform when I had a chip with no HT). I do probably bottleneck having an RTX 4090 with my cpu however, but for running ML stable workloads with ComfyUI, A1111 and LLM studio the cpu never gets to 100% on all cores (seems it's more of a ram thing). I expect a couple of generations from now we might have a new dedicated slot for ML (AI) processing (TPU / Intel's Neural Spiking chip or other) and a very different architecture on mobos for handling ML (fake AI) workloads (DDR7+ and distance to CPU/GPU/TPU to RAM etc), which should also inherently improve gaming performance.
what will happen with older games? might they run better with disabled HT? i remember with my old ivy bridge cpu, where disablind HT lowered the performance in games
Interesting to see what happens with this chip. Chip is manufactured by their main competitor in Taiwan
TSMC is competing with IFS.
This is a product from Intel Product group (fabless)
Many PC enthusiasts are bitching about core type simply because they branded as eCore
They not even a developer, they don’t understand Core are doing
They just don’t like get something different
I think to some extent that’s true. What Luna Lake shows is that it’s not just about power savings. It is about taking more load off the P-cores so they in turn can put effort towards games and content creation. Plus hyper-threading can actually result in lower performance in some situations. If you look at task manager and your taskbar there are an awful lot of programs running in your PC. The more of those basic programs you can run on e-cores the more you free up the P-cores. Plus no one liked the power consumption of Intels recent desktop CPUs. This could drastically improve that.
@@CrazyTechLab The main idea behind E cores on desktop isn't power efficiency, but rather performance per area, though this also leads to better efficiency (because you can run more of those cores at lower clocks while having the same performance and using less power). E cores take up less die space and IIRC you can fit 4 E cores in place of 1 P core, but in an app like cinebench those 4 E cores score roughly 50% more than a single P core. Also on Lunar Lake those E cores sit on the "low power island" so they can just run while the P cores are completely turned off.
On the topic of SMT/Hyperthreading - removing it will hit multithreaded performance, but it allows Intel to get somewhat better single-thread performance. And removing hyperthreading might reduce the need for some spectre/meltdown mitigations which could also bring performance uplifts, though this is part is questionable.
Nah, they are completely right. E-cores, in their current implementation are trash. Literally worthless for anything, but multithreaded performance without manually forcing all other side apps to run on them. They arent even that efficient to begin with.
Didn't learn a thing. Just a rehash of several points discussed at the beginning and needlessly hashed out over a longer format. Same as everyone else. Anyway, some of these changes might be to the capabilities of the new fab that ASML and Intel have been working on in the last decade for which no other fabs such as TSMC or Samsung purchased.
I don’t think software was using that feature anyway.
a lot is nowadays, but the less bottlenecked a core is the less performance you gain for the same increase in power draw.
All of this is great and interesting, but Intel is 4 years late. 🤣🤣🤣
What will be the first CPU with x86-S ?
Panther Lake in 2025
Only intel would remove the only thing that made there x86 platform back compatible , legacy software rakes a massive hit , equivalent to removing 2 performance cores , e cores just rebranded crleron die modded to one chip no matter how intel spin this your getting worse performing chip than last generation and half the chip it should be , intel see stick only giving us 4 core systems while the compition has moved on, this does nothing but bad things to there stock ,and sales figures
x86 is long overdue to dropping legacy support. E-cores in Lunar Lake have the same or slightly higher IPC than P-Cores in 13th/14th gen CPUs...
No importa cuántos anuncios como este publiquen, ya no pueden convencer a nadie para que compren está basura.
Intel ya debe cerrar para siempre.
im so exited for future pc handhelds
Intel focusing on power... yeah they weren't with the 14900k that was sucking up to 350 watts with AVX workloads.
If you were relying on AVX you probably would have been better off running on the GPU
@@LouisDuran Sure if you have access to the source code and go and re-write the software.
@@rattlehead999 sounds like maybe someone's using a hammer when they should be using a screwdriver.
@@LouisDuran Well tell that to the software engineers. Plus it's not always that AVX would be better on the GPU if there is some boolean logic tied to it(dependency).
I am not so sure what is the point of Intel's emphasis on gaming. In case you have not noticed, many gaming companies are closing. Ubisoft is about going bankrupt. Twitch has laid off employees, Microsoft laid off developers at the gaming division. The gaming era might be reaching a plateau, if not a decline. NVDA is actaully focusing more on AI chips than GPUs for gaming and AMD stated the same.
Very soon Intel is going to be a subsidiary of AMD.
or even Apple Chips design just not Locked to the board like M1 is
Sorry Intel but I don't care about my desktop power consumption or AI. Still using the 9900k OCed to 5.1Ghz pulling 250W under full load but it's never at full load so most of the time it seeps under 20W while surfing the net or watching TH-cam, I don't work on my PC or do content creation so for my needs the 9900k is still great and I have no plans to upgrade any time soon.
This product is targeted at power efficiency market. If it’s not for you, it’s not for you.
Lunar lake would quite literally never make it to desktop. It’s only for laptop.
@@whosehandle did he not talk about Arrow lake?
Their CPUs literally started dying due to high power consumption...
seems like ya getting paid by some cpu manuf... but whoecer has q 10th gen cpu , they have no reason whatsoever to upgrade to a newest cpu , they are just fine , ...my opinion
You’re entitled to your opinion of course. But if you pair a modern graphics card with a 10th gen it will be much slower in games than a 14th gen CPU or Ryzen 7000. That’s a fact. And this video is entirely unsponsored 😊
hyperthreading was a joke... At last they removed it?? Thanks god!!
Yes removed from desktops to laptops
If you dont know how it works.. then you arnt bright as its actually better in keeping it busy
@@Scudmaster11 hyperthreading is 95% stupid marketing and 5% (in particular apps) performance gain. It represents the lack of innovation of intel in '10s which ended in '20s fall...
@retrogaminga1072 say all you want... but its actually better than you complain about
@@Scudmaster11 when you find out using less is not an good idea
it makes gpu slower a bit as it not speedly in task to deal with task as the cpu get slower the list is gets need to be more cores over it to keep it close
Hyperthreading should of stayed... mobile performance... really???.. its faster at the expense of some extra power consumption.... PCs arent a mobile phone... and is why ARM doesnt belong in a PC because its not a PC then
Wait till Arm processors get hyperthreading lol😅
@v5hr1ke they wont... and they will never take over for PC as because we want real PCs
then pack more cores, guess what i will stick to muh 13900k...
Who needs fxxxing e-cores on a desktop? There is no battery to save.
Intel should better learn to stand the heat and learn how to propper glue a good cpu together.
This kind of woreshipping "single core" peformance is good for times of c64 or MSDos.
To be fair though when they’ve removed hyper threading in the past for example on the Core i7 9700k it actually had a net benefit in gaming performance and in some games it was faster than the 9900k as a result. If e-cores enable that without a drop in multi thread performance it could be a good thing, irrespective of whether it lowers power consumption or not.
Are you still using dial-up Internet? Noisy desktops are so outdated in 2024.
@@EnochGitongaKimathi "so outdated" - literally the most future-proof platform to ever be.
@@alexturnbackthearmy1907 that's what IBM said about the mainframe. Talk about beating a dying horse and call it future proof. The future is mobile.
@@EnochGitongaKimathi You are a kid right? I've heard somewhere that in 2024 majority of internet users globally are children and tenagers. Limited screen space is mobile's ultimate limitation.
Anderson Paul Anderson Deborah Martinez Jeffrey
Intel is copying Apple silicon design concepts: as least integrated memory and no hyper-threading.
Lack of hyperthreading is not 'Apple concept' but consequence and limitation of ARM architecture. As cor integrated memory - again, not Apple concept and stupid one at that. Just look at yields and Apple RAM pricing.
just with removable cpu over locked cpu
no ht no buy
HT in Arrow lake would have provided a 3-10% performance increase. The core is not bottlenecked almost at all.
@@rattlehead999 ultra 9 is for big boy professionals, they need threads... so my 13900k will work for a couple more years...
"budget cores and actual cores" get it right fanboi
What? You do realize the (SKYMONT) Efficiency core in ARL will be just as powerful as 14th gens performance core?
@@thetheoryguy5544 you do realize intel can no longer keep up dont you?> fanboys are stupid
@@CommanderBeefDev Bro your liking your own comments, just sad lmao.
The new E-cores have the same or slightly higher IPC than 13th/14th gen P-Cores.
@@rattlehead999 benchmark or stfu
my i5 14600 burned off after 2 days.never ever no more intel...im going amd
lol, you're joking then it was DOA my GF's 13600K going strong since day 1
amd adrenalin software not so good
@@cuongtang9539 maybe u Got lucky with your one.
@@abhignaconscience358melting connectors not good
My 13900kf going strong more than a year in use