Blade-Runner
Supreme [H]ardness
- Joined
- Feb 25, 2013
- Messages
- 5,670
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
-
Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.
Intel Fights Back | Arc Battlemage, Xe2 GPUs, & Changing Hyper-Threading
- Thread starter Blade-Runner
- Start date
Blade-Runner
Supreme [H]ardness
- Joined
- Feb 25, 2013
- Messages
- 5,670
- Joined
- Sep 7, 2011
- Messages
- 10,180
So no SMT on the energy-focused mobile chips (4+4 core max). Wondering if they're going to cut SMT for the big-boy CPUs?
Intel is smart enough to know that x86 is running out of time and GPU’s are going to make or break them in the near future.
ARM is growing in both datacenter, for laptops it’s creeping in, for mobile it’s completely dominant. GPU’s for datacenter’s now outsell CPU’s by almost 4 to 1 at this stage and now Nvidia is coming for ARM in the desktop and is supposedly partnered with Intel for their 3nm fab to build it.
Blackhawk+Blackwell coming soon to a desktop near you.
The info says yes, they are cutting it entirely from all chips. More or less with e-cores, it doesn't make sense to do SMT anymore, save p-cores for heavy tasks, offload the easier stuff of e-cores.
Both things can be true. They haven't actually talked about any battlemage desktop GPUs as I understand it. All the XE2 stuff is integrated. They may well skip releasing much for the desktop till their Celestial arch is ready.
- Joined
- Jun 15, 2005
- Messages
- 2,764
I like the idea of no HT on a P-core. Less power ,heat and more streamlined architecturally. Will be interesting to see how gaming performance is affected by the changes.
DukenukemX
[H]F Junkie
- Joined
- Jan 30, 2005
- Messages
- 11,466
Why would x86 be running out of time? AMD's Zen5 is amazing, and Intel's Lunar Lake looks to be a serious competitor. We're praising ARM because Apple keeps losing Steam users to x86 Linux? Maybe it's their decline with web traffic? Clearly Microsoft's 3rd attempt to bring Windows to ARM will work this time?
View: https://youtu.be/hx15hJ87k_0?si=Gx-UFsZz4L0-H_fv
- Joined
- Sep 7, 2011
- Messages
- 10,180
Intel reminds me a bit of the Phenom I and II era AMD. Good chips, just a step behind their competition, and using way too much power.
Maybe Lunar Lake is Intel's "Bulldozer Moment"
But seriously, If this brings interesting competition and a change in paradigm, I'm all about it.
Interesting how this processor is fabbed on "Intel 7" process ....
except for the CPU P Cores, That's on TSMC
.... and except for the E-cores, That's on TSMC
...Or the GPU cores, That's on TSMC
...or the AI tensor cores, That's on TSMC
... or the memory controllers, That's on TSMC
But the rest is fabbed on Intel 7™
So basically EVERYTHING except the interposer is on TSMC.
Maybe Lunar Lake is Intel's "Bulldozer Moment"
But seriously, If this brings interesting competition and a change in paradigm, I'm all about it.
Interesting how this processor is fabbed on "Intel 7" process ....
except for the CPU P Cores, That's on TSMC
.... and except for the E-cores, That's on TSMC
...Or the GPU cores, That's on TSMC
...or the AI tensor cores, That's on TSMC
... or the memory controllers, That's on TSMC
But the rest is fabbed on Intel 7™
So basically EVERYTHING except the interposer is on TSMC.
Why is that?
SMT still has die and power costs. Sure you get a bit more performance 20-30% with hyper threading but you eat up die space and draw more power as well.
Intel is saying they are at a point where if you need performance your better off engaging the P cores. If you don't need the performance E cores can more then handle the job, and saving the die space and power draw is acceptable. When designing any CPU or GPU its always a trade between die space power envelope and performance. Intel is claiming they are at a point where its an advantage to putting extra E cores on the die over using space and power envelope space for SMT.
I like this change. Its probably time for SMT to die off. I would expect Zen6 will probably drop SMT as well, or at least remove SMT from their full cores. AMD is a little different in that their c cores that are optimiazed for efficency are more aimed at server parts. I could see them leaving SMT on those parts... but removing it from their higher clock higher performance cores.
AMD is going to sell more GPUs and accelerators this year than CPUs.
Nvidia is gobbling up datacenter contracts left and right.
Budget systems like Chromebooks and sub $500 laptops are transitioning to ARM.
Intel keeps loosing datacenter contracts to AMD and Nvidia because they more easily integrate accelerators.
And both Amazon and Facebook are expecting to replace another one of their major stacks which are currently Xeons with ARM chips.
Intels bread and butter is big bulk contracts for mostly ok processors and they are gradually loosing them.
Intel not having a strong GPUs and such is hampering their overall portfolio making alternative solutions from AMD, Nvidia, and Ampere far more attractive.
Sales of ARM based servers have increased to a point where Gigabyte, Asus, and Supermicro now offer them, they aren’t custom jobs any more and Linux options for them are better than ever before.
Regarding Windows arm edition it’s good, it isn’t the limiting factor of any of the hardware I’ve played with it on, the GPU was, but that’s something being strongly addressed by everyone because we’re at a point now where storage, ram, and the GPU are the most noticeable things when they have been skimped out on, because for 60% of the users out there even the cheapest of the Intel or AMD options is still more than they will ever need. It’s why Intel moves so very many machines with the N95 and N4500 and those are just barely usable.
As it currently stands for any machine that isn’t a gaming rig ARM is a strong contender IF you have one with a decent GPU. And with the advancements on the ARM platform as a whole it’s not at all guaranteed that gaming won’t become viable in the next 3-5 years there either.
Not only that but as cache quantity goes up and task schedulers get better, it opens options for out of order execution that (on paper at least) looks to do a far better job than SMT can accomplish. Intel looks to be laying a foundation for the next few generations of hardware.
This latest windows arm play will be the strongest we have seen. Software is so much easier to deal with now... and the arm hardware is starting to get very attractive.
I know we have been going on about arm it seems for ages... but really in the grand scheme its only been a decade or so since the first serious ARM servers shipped.
I think your probably right on with the 3-5 years for ARM gaming going mainstream. A few big things are about to happen. First is the Switch 2 or whatever Nintendo decides to call it. Its going to be rocking much more powerful Nvidia ARM hardware. Game developers already have dev kits, its going to mean a lot of "AAA" and "AA" backend work is going to be in the works. So many of the major game engines like Unreal have already made ARM builds reasonable undertakings. If the new gen ARM laptops sell. We are probably 2-3 years away from someone going all in with ARM and producing a ARM MOBO chipset. Socktable ARM? Why not. We know AMD was toying with the idea with the first version of Zen... they had ARM AM4 chipsets up and running internal before Lisa killed the project. I think its very possible someone revives the idea.
Very true... if they pull this generation off without giving up too much in things where SMT did well before. It is a logical change, your right it shouldn't really be needed anymore 99% of the time the CPU has more real cores then it needs anyway. Better faster cache should only make that more true. I'll be interested to see how it plays out with games that do better without SMT on. As long as the production workloads aren't effected this could be a big boon for gaming.
Well a prime example is the 5800x3d and 7800x3d, game performance is either identical or slightly better with SMT off.
It’s likely only going to get better with it off as time goes by.
NVidia currently sells slotable ARM SoCs and they sell very well, so we’re close.
Lisa was right to kill their ARM offerings though, AMD was not in a position to fund it at that point and while their financial position is far better than it was, AMD still doesn’t have the software team to head up ARM Linux development they are far better off collaborating with Microsoft as they work with everybody to develop ARM chips then simply launch one themselves when the groundwork is laid out.
Agreed. They would have been biting off way more then they could chew at that time. Just an interesting side note... knowing AMD has had a ARM socket product basically ready to go at one point.
If the new Qualcomm chips do well I could see AMD releasing a ARM chip aimed at mobile... they wouldn't have to do a socket for a play like that, but it could be interesting if they did. Could lay the ground work for an eventual Desktop ARM chipset and socket. It seems like AMD is going to keep pushing x86 and that is probably the smart play for now. If Qcom does get some market share though, AMD entering the market could make things interesting for Intel.
DukenukemX
[H]F Junkie
- Joined
- Jan 30, 2005
- Messages
- 11,466
It didn't so far.
https://www.pcgamesn.com/amd/radeon-gpu-sales-nosedived
For their GPU's and not ARM.
Which somehow has less market presence than GNU/Linux on x86.
Yea but AMD is still x86.
OK...
Didn't Intel have a problem where they couldn't keep up with the demand for their Core Ultra chips?
https://www.pcworld.com/article/231...blems-are-hindering-hot-core-ultra-sales.html
I get AMD and Nvidia for GPU's, but Ampere? Ampere doesn't have a GPU.
Again servers. How does this apply to the desktop?
Ok but Intel's new GPU's aren't bad. Qualcomm's new Snapdragon though is probably not going to be good. Also, which Windows ARM edition because we've had ARM for Windows long before Apple switched to ARM for their Macbooks?
Or like 90% of the software and hardware on Windows. For what? A different CPU architecture? ARM has no advantages over x86 right now.
DukenukemX
[H]F Junkie
- Joined
- Jan 30, 2005
- Messages
- 11,466
Just to remind you guys but the 3DO ran ARM. Yes, the 90's 3DO. The Game Boy Advance ran ARM. Why would the Switch 2 have an impact on ARM's adoption for PC gaming? Especially when the original Switch also had an Nvidia ARM chip.
Its about the engines engines not needing a ton of tweaks to work on either.
Sure we have had ARM gaming devices as long as mobile gaming has been around. The difference is the games on those devices had to be built from the ground up to work on ARM. The versions those machines got were not the same as versions of the same game for x86.
Even with the Switch... there was a ton of heavy lifting that developers had to do to get their game to run on the switch. I mean less then ever before but still it often requires repacking textures, and making a lot of engine level tweaks and adjustments. So sure you get a harry potter game on switch as an example... but that wasn't a simple rebuild of the source they had to spend resources making a switch(arm) version. Switch 2 should give them a lot more parity feature wise. AAA games won't need a ton of development time to create Switch2 versions. Those shame builds should essentially run on any Windows ARM machine with a GPU at least on par with the switch 2 hardware. If a AAA studio makes a switch2 version of a game... they have done 99% of the work of making a windows ARM version as well.
Last edited:
Lisa has been bragging for months about how the MI300 series is dominating their sales figures. Desktop means nothing right now, those numbers reach their new lowest figures ever each quarter.
Bundled with their Grace CPU’s there’s no reason to pair any of the H100, H200, or BK100 devices with anything else unless you are planning on having those machine for dual purpose work, and that’s not really what they are designed for.
That had more to do with manufacturing limitations than demand but it was the first serious contender to the Apple M series and the world took notice.
The more products change to online based it trickles down, it helps that anything .Net is basically seamless and that Unreal has a native ARM variant that shares identical scripting with x86 so you can switch between them. Arm laptop sales go up each year not down. Desktop everything seems to go down at each report, desktop is becoming far less relevant and the industry is taking notice.
The Intel GPU’s aren’t bad and the new architecture looks awesome, I’m excited for it. But the fact it’s still not on the desktop as an iGPU is an issue Iris no longer holds up and it needed to be gone 2 generations ago.
Other than it constantly does better at power usage and thermals at sub 65w which is where the bulk of device sales are.
There was a time when Intel would get projects for 80,000 Xeons.
Now those same projects are for 12,000 Xeons and 72,000 GPU accelerators. And thanks to Nvidia’s bundling of their enterprise cards and the Grace CPUs Intel isn’t even likely to be getting those sales right now. Their accounting teams have noticed, and that’s why Intel can’t afford to ditch their GPU lineup, it relegates them to 3’rd in the integrated hierarchy because Nvidia and AMD offer better balanced bundles in the case of the Datacenter Altera doesn’t have a GPU solution but usually neither do the Xeons they use bare bones VGA adapter over an M.2. It’s a dead heat there but depending on the workload ARM has significant advantages and the more things become a web app the more ARM servers make sense.
I'm confused here, I hate to admit it. Doesn't SMT mean Surface Mount Technology? What does chip mounting have to do with core optimization?
P-core only Desktop chips would still have SMT
Mobile chips focusing on both area efficiency & power efficiency will not have it
(Future consoles with power limiting scenarios might ditch SMT)
https://overclock3d.net/news/cpu_ma...hyperthreading-with-lion-cove-and-lunar-lake/
SMT meaning Simultaneous Multi Threading or Hyperthreading for Intel.
Lot of this (like the texture) seem more because of how little ram-power the switch has than x86 vs arm. what it took to run WoW on arm or death stranding, doom 3, Tomb Raider etc... when they had Mac in mind could give a better idea, was it a big deal ? did dev ever got a feedback, the running on 4GB of total memory (vram+regular) and a 4 cortez a 1ghz would be probably such the bigger challenge than compiling for ARM. And like porting a PS5 games to PC is still a bit of work, would the GPU difference be a bigger factor than the cpu, if the switch is not regular Vulkan (they use some custon NX API call I think), we can imagine the switch 2 like the PS5 and Switch 1 will have a bit its own language and particularity.
i think what would jump start an PC gaming on ARM, would be an really good x86->arm translator that let you run older title (or new low requirement one).
That said a lot of the most popular title have already an arm version, World of Warcraft, fornite, Baldur gates 3, Among Us, Minecraft, League of Legends.
True, but ARM is way more open (and like you often say open always win versus fully close like x86), if you have Nvidia and other entering and competing, what come out of an Apple-Nvidia-Intel-AMD-Qualcoom-Samsung-Microsoft-Amazon-Huawei battling out could beat the Intel-AMD x86 chips, that a lot of R&D dollars to beat.
Or at least push CPU margin down and be popular until Intel-AMD cut-price, x86 cpu price by transistor/mm seem really high versus GPU right now and GPU are not that cheap.
Last edited:
DukenukemX
[H]F Junkie
- Joined
- Jan 30, 2005
- Messages
- 11,466
Do you see Nintendo porting their games over to Windows, ARM or otherwise?
You think games built for ARM on Windows doesn't need to built from the ground up? Just because it's the same OS, doesn't mean a different CPU doesn't matter. Whomever is working on ARM right now for Windows has to rearchitect everything.
When the Switch originally came out it was weak, but not too weak compared to the PS4 and XB1. We know enough about the Switch 2 that it'll be the same situation. The benefit here is that Switch 2 games can be made for DLSS and some form of frame generation that'll get it much closer to the PS5 and XSeries consoles. On the other hand the Steam Deck has started a revolution in portable PC handheld devices, and if the Switch 2 costs as much or more than the Steam Deck, then you won't see the same sales boom as the original Switch did. The Steam Deck clones aren't exactly running ARM either. Why you think Nintendo shut down Yuzu when it allowed Switch games to run on the Steam Deck? The Yuzu developers were getting ready to emulate the Switch 2.
The MI300X is doing very well, but The MI300A comes with Zen4 cores so... These are meant for servers and particularly AI stuff.
Yea but because sales are so good that Intel is going forward with Lunar Lake, which is said to be even better at everything Apple M series. It's not like Intel plans to dump x86 for ARM.
Where do you see ARM laptop sales go up when Apple last year had a 34% sales decrease year over year. Microsoft's attempt at ARM failed, and what Chromebooks? Assuming most Chromebooks even run ARM, their sale declined by 20% last year. Microsoft Windows is gaining market share on x86 so I don't know where you get your numbers from?
If you think Intel's GPU isn't on the desktop then wait and see what Qualcomm's Adreno is going to be like.
Except not against AMD. Intel is like 80% there, but that's not too bad. Also what sales?
It is would more be about third party porting their game on the Switch (from The witcher, doom, The Outer Worlds, Diablo 3, MOnster Hunter Minecraft, Skyrim, etc...), now that we have a Arm version launch it has well on Apple-Windows arm silicon native, to test this idea, is there a larger percentage of game with a Switch port that also have a Arm Mac port ? seem to be the case, but obviously there a bias, a studio that port their game to the switch will have porting game to most platform already more in their plan-habit-goal regardless of if that mattered at all.
WIth x86 MacOs slowly dying in the field, that seem normal.
in actual relevant test, not in test that push x watt cpu to their max usage ?
Nintendo porting games over... of course not they are still Nintendo. I was more referring to the AAA titles from third party developers on Switch.
As for the CPU mattering to games... ya it really doesn't matter much anymore. 95% of all games are built on the same 2 or 3 development platforms. Yes any game made with unreal as an example can be compiled for ARM in an afternoon. Having said that their are "best practices" for using unreal with ARM. My point is with systems like Switch 2, its just more reason for AAA studios to ensure they are using the best practice methods from the start. I would imagine pretty much any game released in 2025-26 will be able to have a ARM version compiled in a few hours of work. That is a big deal yes... games on ARM devices up until now have required pretty massive rebuilds and weeks or months of dev debug time.
As for the steam deck outselling the Switch 2. That is laughable. No PC gaming is not going to over take Nintendo. No matter what you may thing of Nintendo games, their games sell their systems. That isn't going to change. Over 140M switches sold so far, I know people with 3 of them. (I mean the deck is wicked cool and all lets be realistic Valve doesn't release numbers but its probably 3-4m total units sold) When the switch 2 hits I'm sure Nintendo will have a handful of first party Nintendo games that will move units. If they have something like a follow up to Zelda tears of the kingdom ready to go at launch they will move units and outsell the steam deck in a matter of months maybe more like weeks.
Last edited:
That was never going to happen. The shift to GPUs is clear, especially at the datacenter level, and Intel knows it’s necessary for them to remain relevant, unless they want to become a pure fab play.
Also it hurts single threaded performance on the P-core. If you have it cranking hard on a thread, but then another thread is saying "can I please to have time on the core too?" it hurts that single thread performance. Now when CPUs only had 2 or 4 or 6 cores, maybe that was worth it. The decrease in performance was less than if you had to do context switching to have that same thing happen without SMT. But now we are seeing CPUs with 24 cores, and it could grow further. No reason to split the time on the P-cores when there's a bunch of E-cores just sitting there waiting for stuff to do.
Good summary. To add to that list, hyperthreading has security implications as well.
I saw some performance numbers today on Intels all E core Xeons and, it was reasonably impressive. I still don't see the value over AMD right now in that space. It proves the point though... those Xeons have insane numbers of E cores. If they can make processors with 8 P cores and 16 E cores at that point hyper threading is just holding you back, assuming the caching system is up to the task.
Ya but Intel never worries much about that...
I would even say if the scheduler is good enough, no reason to not have e-core, p-core, HT p-core if one want, they seem to have made both type of core, maybe will see a 4x1 thread Pcore, 4x2 thread PCORE,8x ecore, with some logic; application-scenario that use more than 4 thread to their max will use a lot of them.
If you have more active thread than core this will still happen but much worse now too right ?
These sales
https://www.statista.com/statistics/273495/global-shipments-of-personal-computers-since-2006/
Based on those in 2023 Intel and AMD combined moved some 242 million x86 units.
Apple sold 130 million iPads alone in that time frame.
Of those 242 million units less than 60 million are desktops.
But your shifting the argument which was about how Intel is not at all in any position to give up on their graphics departments, and their GPU division will make or break Intel.
I’m not at all going to be surprised if The foundry services and GPU become the more valuable aspects of Intel over the next 5-10 years.
Ya but continually saying "The schedular needs to rock," doesn't make sense if there's an easier way to solve the problem. If HT is not useful performance wise for a desktop these days, take it out. Particularly since, performance aside, it is NOT free as others noted. It costs transistors, which is why E-cores don't have it. So if taking it out doesn't have a negative performance impact overall, and it has a slight positive impact for single threaded, AND it lets you save transistors for other things... why not do it?
The idea would be that for all those little threads, which your system can have literally thousands of (mine has about 4700 total threads right now and all I'm really doing is web, e-mail, teams, and the Hyper-V manager) that don't use much time you toss those on the E-cores. Now yes, because they don't have SMT it will have to execute a context switch each time, but that isn't a big deal, the cores aren't doing much anyways.