"Arrow lake is harder to delid, but we will talk about the solution when the emargo lifts". I just laughed cause of course you already have a solution to that!
@@arch1107 Yeah, gone are the times when CPU was "it will work for 20 years even if you overclock", now you are happy it did warranty at stock... And problem is always pushing parts way above what they should be pushed, that's why we have CPUs from AMD and Intel that can cut their power consumption 50% while losing 10% of speed.
... thats a dumb take my man. motherboard manufacturers werent at fault in the end. intel just used them as scapegoats. if you dont believe me, go watch Buildzoids videos on the topic.
@@jake20479 So completely taking away all the limits of CPU is not on them? Or that time when 7800X3D was killed with 1.3V (against any AMD recommendation) on soc voltage (search on exploding 7800x3d), then new BIOS came out and dying stopped. Or when asus and gigabyte was faking power consumption of ryzen 5000, so CPU boost will go all out not knowing that it is going above intended TDP. What maybe CPU manufacturers could do is hold MB manufacturers by the balls so they won't pre-overclock CPUs. This intel dying thing was part of their code, but MB should never ever OC anything by default, and now they do.
4% more PCB thickness isn't _quite_ as useless as it sounds, since bending stiffness scales with the cube of the thickness - so it's roughly a 12% stiffer PCB. It's of course still utterly ridiculous to do what they are doing with the two central contact points and just a bit less contact pressure. Well into facepalm territory there...
@@laserspike I agree. Intel mitigated the bending issue to a "good enough" level for them as generally speaking the issue does not impact CPU longevity and liability due the CPU triggering their temperature protection mechanism. As a business perspective makes sense. 99% of people don't even know about this deformation defect.
Roman, can you please check for *idle power consumption* when conducting your review for these CPUs? Asking because this gen is going for tiles instead of monolithic so I’m expecting a potential increase. Currently most Intel CPUs idle at 5-7W while AMD idle around 20-25W.
@@MrEdioss It's true, so many out there can't even run their 5800x unless in eco mode due to heat, i had to limit the limit to 85c and turn on eco mode. running at 100% for more than a minute resulted in crashes, this isn't isolated, anyone using ANY motherboard without robust enough vrms, cannot run the 5800x and presumably any amd processor that acts the same way.
I bet that it is not that filler and it somehow introduces more latency, as Tech Yes City said for the 10th gen vs 12/13/14th gen because of the position of the IO die
I mean, 3D Vcache uses silicon shims or cover tiles for the area of the CPU the stacked cache isn't over. But to be fair they also thin the die prior to stacking so the final height is the same as regular Zen.
It's up to the motherboard manufacturer on how to distribute the PCIe lanes, that's probably why you can't find any information about it in the cpu datasheet.
Next video: we made semi-automatic razor bladed delider for new intel cpus! Next next video: we made a new high conductive thermal glue after you delid your cpu Even next video: we are investigating crunchings on new intel cpus
So, DLVR is the comeback of the integrated voltage regulator that we had in the past, but with an optional bypass feature. Sounds good. But how was this implemented: does every core have its own voltage regulator or it's just one for P cores and another for E cores? Is the DLVR bypass a requirement to board design or an optional feature. Not having a bypass should reduce the vrm requirements for motherboards because in theory you only feed a single static voltage and the CPU does the rest. This is the thing that I will miss the most from anandtech. Those guys went very deep on these architectural details
FIVR is a switching converter, so you can feed the CPU package 1.8 V at a current level that's a snooze cruise for even the wimpiest of VRMs. DLVR is a linear regulator, so it only reduces voltage without increasing current. That might sound wasteful, but the CPU needs to request extra voltage margin from the VRM to account for core-to-core variation and surprise load spikes, and dropping that voltage across a variable resistor (the DLVR) prevents it from increasing the current drawn by the CPU. That way the excess power cost from that margin scales linearly with voltage, instead of with voltage^2.
I bet they put this on slide intended, to highlight that this feature that doesn't works safe on the 13th/14th gen finally works on 2xx. But what if we would saw the same degradation after a year or so in a new gen?
Great job on previewing Arrow Lake! However, I wonder why/if Intel did not gave out slides equally? Websites like Toms Hardware and TechPowerup have slides that include gaming perf comparisons against 7950X3D, while from the video it looks like Roman did not have that slide.
Thank you, for this absolutely best and most detailed explanation of the new Intel CPUs! I am truly thankful for finding this RUclips Channel! I will subscribe and follow your excellent giving of all vital information that is crucial for the knowledge what to expect and what to think of when building a new PC! Wishing you all the very best! Sehr gut im alles! Viele danke schon! Greetings from Sweden!
I think you need to check your historybooks. Look for Intel Clarkdale and Arrandale. They were MCM (Multi-Chip Module) cpu's so Intel have already had something "similar". Performance per watt.. intel used that even in 2005 in their marketing :P
I swear to god you amd fans sound just like right wing nuts. Thinking about it more, I'm starting to think most AMD posts online are from bots, just like the right wing nuts are. Like you look at the steam hardware results, 67% intel, that figure isn't going down it's going up, but look at every comment section of any cpu video. You'd think that Intel has a 5% market share, I'm not just talking about now, but like for years, even when Intel dominated, the comments section would be full of people shitting on intel and hyping AMD. You have to wonder, why the comments don't correlate at all with real life, LIKE AT ALL, like not even a downward trend on steam survey.
Thanks for covering the ILM and contact frames. Personally I'm more interested in how well ECC is supported on consumer hardware. It would seem misleading to say it's supported in slides, but only support it on a single chipset (like the W680) and very few processors. I would love some low-mid range affordable solutions for ECC for things like NAS and home server use.
The AM4 Ryzen Pro cpus all support it as do some AM4 mobos, even ITX ones e.g. ASrock A520M-ITX/ac and Gigabyte A520I AC. There's usually a few used Ryzen Pros on eBay.
Good to know that contact frames are still going to work best. Kind of strange they didn't fix the issue with this new CPU, I agree. Thumbs up for great videos.
Room to grow? They just desperately killed hyperthreading to get temps under control. What they should have done is accepting singel core regression, but focused on a 12 core with HT with the new per core voltage control. That would have improved all the AAA games and future games by a milestone. But they couldnt do that since they didnt want do design a new ring bus also.. This thing is just the definition of MEH.
@@seanpereira In terms of thermals it definetly makes sense to put it in right. Air is closer to a vaccum than something solid, vaccums don't carry heat.
3:15 When they say a CPU has 9% better IPC, I think they mean that at the same clock speed, the newer CPU can process about 9% more instructions per cycle compared to the older generation but since the 285K will have 6% lower clock speed they will basically be tied in gaming, makes sense to me.
Worse than tied. 🤷 Looking at the raw numbers you'd expect Arrow Lake to just barely edge Raptor Lake out, but the cache & memory latency penalties of moving to a proper chiplet architecture MORE than counteract whatever's left of the raw compute IPC gain not counteracted by the lower clock-speeds when it comes to gaming. Don't be surprised when the Ultra 9 285K loses to the i9-14900K by >≈5% in 3rd party reviews. Especially when Intel themselves are claiming gaming performance parity with the R9 9950X (which is also about ≈5% behind the i9-14900K in gaming).
Love der8auer videos! I have lots of thermall grizzly products. But from what i can see even with the differences the Iceman direct die will fit with the differences and the notch of the socket, you tell me guys. I will by waiting for thermal grizzly product... please do not make us wait so much like last generation
Very nice presentation and actually showing issue’s with 1700 socket parts like the socket frames which appear’s need a new style to fit the 1851 socket. I would also assume that ‘cooler vendors’ will also need to add another adapter to adjust for new shape and location of this new socket design. Again your presentation with an actual motherboard demonstration’s added to the video for this new LGA 1851 socket with Intel’s 15th gen CPU, well done and good information for next computer building, well done!!!.
Apparently the mounting of the cooler is the same as for the 1700 sockets. Intel already moved the IHS as was shown in the video. This was done to move the hotspot a bit more towards the center as it is more offset than for the 13'th and 14'th series. Now we come to hear say. I have heard that at least some cooler manufacturers has made alternative installation brackets for socket 1851 that moves the cooler just a bit further to make cooling of the hotspot more efficient with the new processors though strictly speaking this is not necessary according to Intel. Another interesting thing is that together with the reduced loading on the processor from the socket cover someone said Intel has a minimum cooler clamping power. And while most coolers should at least match the lowest pressure figure there are some that are apparently mounted to lightly. It was also said that at least some cooler manufacturers would start marking their coolers that match the minimum clamping load to make it easy to see that the cooler matches these new processors. Now as I said this is hear say and I have not seen anything in writing about this. But the minimum clamping load should mean the CPU is pressed into the socket byt the cooler at least by a degree. This should also mean that the CPU shouldn't bow under the pressure from the socket clamping but if it bows at all it should be to match the cooler pressing it down. This would be a logical way to improve the contact with the coolers. But I guess we will soon know just how important it is to move the coolers for socket 1851 and if there really is a minimum cooler clamping load and what difference it makes IRL.
@@blahorgaslisk7763 much appreciated the sharing of your knowledge. Not in a big hurry to build new system, maybe next spring or in fall of 2025. Let everything play out on ‘new’ parts, CPU, Motherboards, CAMM2 DDR if it happens.
I was also thinking about that. If you don't have the tool to grind metal, you could just "Hotknife™" that plastic nipple off of the board just like we used to do to plug an X8 card in an X4 slot ^^
16:30 I'm familiar with the internal 'discussion' about the ILM. Basically it came down to cost and simplicity vs. the actual severity of the problem. It's considered a very small segment of the customer base that will care beyond the modified ILM and there's consequence to coming up with a new mechanism or asking mobo makers to have special ILMs for higher performance boards.
Good point about the contact frame fitment, can be really critical. Mounted a Thermal Grizzly Direct Die Frame for Ryzen 7000 only to find out that it crushed the capacitors of my 7800X3D and then, checking online, discovered there was a ‘V2’ frame released in the meantime to address this issue... Not a great way to have your CPU killed 😢
@@PolskiJaszczomb What are you even talking about? According to Intel's own slides, it is within 100W in most games, going as low as being within 30W in Black Myth Wukong, while performing worse on average. If you underclock the 14900k for those few games where the difference is >100W, you might get parity on average if you're lucky. And remember that these are first party numbers. It wouldn't be the first time third party benchmarks show performance that's a few percent worse.
@@pmHidden Have you ever have a 14900k? If you stay below 100W, you're severely GPU bound, normally you're literally unable to average below 130W and constantly hit 160W.
Great overview of the Intel Arrow Lake! 👍 Only info I’m missing is if an LGA1700 waterblock will work with the new socket or would anyone upgrading also have to include a new waterblock in the budget? 🤔
I honestly wouldn't have a problem with the "up to" figures if they also included the bottom range too. I want to know what the lowest spec on offer is.
Please test the new Arrow Lake against 14900K with HT disabled and matching max frequency. I suspect that the power "efficiency" of the new SKUs comes from there.
Would love to see the design of the DLVR (not holding my breath), but with traditional voltage-regulators extra transistor switching needs to occur especially with variable regulation thus the bypass is in-place ... hope we can get more tech-specs on this
Impressive somewhat on the efficiency, Disappointed on how the performance they showed vs 9950X in gaming, and vs 7950X3D in content creation. will be interesting for real world testing and especially, if you would like to probe the CPU voltage behavior
Can't wait to learn more about your upcoming contact frame. I hope you use it for all the testing's in your upcoming video. Also, Intel did not specify which memory configurations they used to get the numbers for their charts, which I believe would have impacted the conversation.
Thanks for the info. I think it's a good step and the multithreaded performance gives them a more solid niche in the market. As always, we will have to wait for independent reviews to give a verdict.
Ima need to upgrade my i7 6700k pretty soon and was hoping intel would add more cache, cause when i was setting up the 7800x3d it took alot of research time to manually adjust soc vcore from defaults cause of longevity concerns, and also setting up the Ram kit with expo profile and so many other ram settings that only exist to destabilize not make performance gains, thanks to infinity fabric. I just want an easier plug-n-play solution without having dig deep to find out that factory defaults are bad, like Amd's auto-enabled overclock PBO that only raises temps for no real world benefit.
Looks like you've been fed a whole lot of "internet expert" information. There was a bug that caused elevated SOC voltages on 7000X3D chips but it was fixed in 4 weeks from the first pictures on Reddit. All you needed was a BIOS update. On the other hand people with 13th and 14th gen processors have been tweaking and tuning for over a year before Intel managed to get that mess sorted. You don't need to do anything other than load the 6000CL30 profile for 7000X3D chips. And if your motherboard auto-enables PBO you return it and complain to AMD. That is not default behavior. ASUS figured they should enable PBO of you selected Liquid Cooling, but that was shut down by AMD with an AGESA update.
Do you think Intel might be sandbagging a bit leaving headroom for overclocking on the table so you’ll get your gaming increase with overclocking and memory support
Wonder how long before we'll see an 1851 contact frame from Grizzly? Got a new build next month and would really suck to either have to wait on a contact frame or have to jump through the hoops of removing AIO etc etc whenever a compatible one is released.
Wait, would the NPU help with gaming? Not development. Or with ai assistants? What exactly can it help with, really... I wonder what the apple npu does for apple products, but I digress...
Not really, since gpu already does all the work for it. Maybe offloading frame generation to it and iGPU if intel goes for some actually useful everyday features? I mean, it will give only few base fps, but at least something.
@@alexturnbackthearmy1907 hey, thank you for replying! I am currently running a 10850K + 4080 setup, you think Arrow Lake would be a noticeable improvement in terms of gaming performance?
Could we possibly see more gains in gaming when pairs with a 10,000M/Ts ram? Also i feel like this CPU has decent overclocking headroom, maybe not 2600k level.
If power consumption efficiency is as they advertise, this is a bigger selling point than any speed or 10% fps increase in gaming. Power consumption can also mean better 1% lows. Ultimately the cost of the electricity bill is what wins it for me.
I’m interested in this base tile. Is it using Infinium ? Or what other cpu bus is there Frsnkly I’m glad they are looking into more computing per watt. Chips consuming 300 watts is crazy train
It's kinda weird, I mean with the Core 200 Intel just seems roughly able to compete with AMD, with no real strong selling point, loss of HT being a clear con But the new architecture seems to have potential so maybe in a couple generations they can pull up ahead again, what killed them was to cling to their old architecture
I think hotspot temp is a major issue, and thats why they went for individual voltage lines per core, and no HT. Frequency is probably going to jump more around than on a X3D part. Im thinking a decent oc margin with delidding, but if it translates to more fps is the big question.
I guess we don't know yet whether a Gen 5 SSD and a GPU both in their respective primary slots can be run concurrently without detracting from GPU bandwidth.
They can...in theory, if these gen 5 lines wouldnt be taken by something like 4 thunderbolts and something else. Well chipset should still give quite a chunk of 4.0 lines.
Tensor cores are super useful for image recognition. Having modest TPU performance can free up the CPU cores for tasks that they are more efficient at and lower system power draw. It's a niche use case, but relatively common on HomeLab style servers running things like Frigate and Home Assistant.
The naming is confusing for sure. Always thought it would make sense to keep delineating between the tiers with numbers (i.e. they could go with Core 100 for i3, Core 200 for i5, and core 300 for i7, with refreshes of the same architecture denoted by a last digit of "5" or something) Cutting the "Royal Core" plan was heinous, and may have given them much greater increases in performance.
17:50 Honestly it is surprising to me that this make it worse at cooling. 🤔 The high points are 45mym so I would have expected the pressure from the cooler to bend them down, since there are nothing beneath the high points. Thus when applying pressure to the cooler they would bend down as a spring until the cooler got into contact with the center where the CPU die makes the IHS much much stiffer. Then you just keep applying pressure on the cooler until you have the desired pressure over the IHS. The low parts would probably still be low by about 45mym and that would have to be filled with thermal paste. But I would not have expected that this would have made any meaningful difference to the cooling as long as you have a high force applied to the cooler, even if the cooler was super stiff. (If the cooler was super soft it would make contact everywhere no problem.) So I am surprised that the contact frames works as well as the do.
Impressive and Disappointing at the same time is good way to describe Intel in general at this moment in time. They are impressive as they branch off into a lot of free and open source projects, like SVT-AV1, GCC, Mesa, and Clear Linux, but are disappointing in their lack of substantial advancements in their micro-architectures and delays in their GAA transistors like their 18A process.
Good to know I can get a contact frame for 1851 right outta the gate. Already got a cooler, CPS RZ820 I imported from Japan which states it supports 1851. (Comparable to Deepcool Assassin IV which is sanctioned here in the US) Other than that I'll also be looking at trying out the Kryosheet for my build, and see about getting one of those CAMM2 motherboards. Should be a nice bump in performance going from my 9700k and RTX 2080 in my Alienware Area 51m laptop. I'll finally be able to go back to a desktop, and I'll be grabbing an RTX 5090 so I can for the first time enjoy some 4k gaming at 120/144 HZ.
you already talk about socket. but you forgot to mention if those exisit cooler blocks in custom loop or AIO if its still compatible with this new socket
I don;t know if anyone noticed but when comparing to 9950X both CPUs were set to 125W TDP. Thing is - 9950X by default runs at 170W TDP which allows up to 230W of power consumption, limiting it to 120-125W reduced that to 160-170W meanwhile 125TDP for intel allows up to 250W of power consumption. 9950X was severely underpowered in multicore tests compared to 285K.
@62185hhh🤦 Except that most benchmarks shown like Cinebench finish or are at least majority done during that boost window! And you're completely assuming that they aren't using standard motherboard default behavior on Intel for over a decade where that PL1 turbo time limit is just disabled completely and has to be MANUALLY turned on by the end user.
Look at the comparisons made for power draw, notice what's missing? Edit: The extra 4 PCIE lanes from the CPU should be for connecting to the chipset...
I have an i9 10080XE and the answer is simple: Mesh design, which you need to overclock for good performance...And HEAT. They're trying to lower temps while staying on lower nodes, not go back to 14nm+++++ lol.
Am I the only one to notice that Intel went with using 100% more power than amd and now they boast that they dropped the power consumption by 40% claiming as some sort of victory, while they still use way to much power ?!
Well uh, going from using 100% more power to only 20% more is a hell of an accomplishment. 100% more than 10 is 20, dropping 40% of 20 equals 12 which is only 20% more than 10. Let’s just be glad for some competition. Intel could be like AMD is to NVIDIA.
Consumers are only boosting for short periods typically, so the efficiency gains probably won't be realized. This is an enterprise / data center problem.
![[#2024MAMA] BIGBANG (빅뱅) - 뱅뱅뱅 (BANG BANG BANG) + FANTASTIC BABY | Mnet 241123 방송](http://i.ytimg.com/vi/EQsYfiPR9uM/mqdefault.jpg)
![[#2024MAMA] G-DRAGON - 무제(Untitled, 2014)+POWER+HOME SWEET HOME+뱅뱅뱅+FANTASTIC BABY | Mnet 241123 방송](http://i.ytimg.com/vi/Ox29z5Nf1Uk/mqdefault.jpg)
I am so glad you have a channel, it's so much better than many of the other larger tech RUclipsr channels ... no BS, no drama ... just great quality.
Thanks a lot!
@@der8auer-en100% agree with OP. Thank you, Roman. Most informative video I’ve seen today.
Comparisons are odious - quality stands on its own.
Is he talking about a long curly haired YT channel? 😂
100 % agree with that
"Arrow lake is harder to delid, but we will talk about the solution when the emargo lifts". I just laughed cause of course you already have a solution to that!
If I had to hazard a guess. He's probably made a delidding tool where it is slightly raised at the sides and goes down right near the ihs.
I think your idea is right
It’s Roman, I’m not surprised at all…
Up and down instead of side to side looks like it wouldn't hit the SMD components... maybe
Nice test
MB manufacturers can't wait to use that DLVR for daily driving these CPUs at 1.7V so it can ignore all this new power saving.
add the poorly manufactured firmware and we will be back in the last year problem, more burned cpus, now made by tsmc!
@@arch1107 Yeah, gone are the times when CPU was "it will work for 20 years even if you overclock", now you are happy it did warranty at stock... And problem is always pushing parts way above what they should be pushed, that's why we have CPUs from AMD and Intel that can cut their power consumption 50% while losing 10% of speed.
... thats a dumb take my man. motherboard manufacturers werent at fault in the end. intel just used them as scapegoats.
if you dont believe me, go watch Buildzoids videos on the topic.
@@jake20479oh ASUS definitely does some garbage
@@jake20479 So completely taking away all the limits of CPU is not on them? Or that time when 7800X3D was killed with 1.3V (against any AMD recommendation) on soc voltage (search on exploding 7800x3d), then new BIOS came out and dying stopped. Or when asus and gigabyte was faking power consumption of ryzen 5000, so CPU boost will go all out not knowing that it is going above intended TDP. What maybe CPU manufacturers could do is hold MB manufacturers by the balls so they won't pre-overclock CPUs. This intel dying thing was part of their code, but MB should never ever OC anything by default, and now they do.
I am severely disappointed in the lack of cat in this video
We know what's important.
This! 🎉 no videos without the kitties! Even if it’s just a small graphic at the bottom.
This obviously makes this the worst Intel launch in years
The cat moved to the AMD camp
No pets no unnecessary troubles and hairs in your apartment.
4% more PCB thickness isn't _quite_ as useless as it sounds, since bending stiffness scales with the cube of the thickness - so it's roughly a 12% stiffer PCB. It's of course still utterly ridiculous to do what they are doing with the two central contact points and just a bit less contact pressure. Well into facepalm territory there...
@@laserspike I agree. Intel mitigated the bending issue to a "good enough" level for them as generally speaking the issue does not impact CPU longevity and liability due the CPU triggering their temperature protection mechanism. As a business perspective makes sense. 99% of people don't even know about this deformation defect.
They want to give Thermal Grizzly the extra business, such a kind move by Intel.
4 cubed is 64 (not 12%). 4x4x4=64
@@EnWorks 1.04 cubed is 1.124864 aka almost 12.5% increase
@@EnWorks LOL, so true 🙂
Roman, can you please check for *idle power consumption* when conducting your review for these CPUs?
Asking because this gen is going for tiles instead of monolithic so I’m expecting a potential increase.
Currently most Intel CPUs idle at 5-7W while AMD idle around 20-25W.
Finally, Intel targets efficiency. AMD been killing them in that arena.
If you said this 10 years ago, you'd be getting so many weird looks! Insane time!
@@BBWahoocan't even run stock tdp with cheap motherboard while on intel I can, in 2014.
@@MrEdiossany z690/z790 in the $150+ can. I ran a 13900ks 6ghz all core on MSI z790 WiFi for $180.
@@MrEdioss It's true, so many out there can't even run their 5800x unless in eco mode due to heat, i had to limit the limit to 85c and turn on eco mode. running at 100% for more than a minute resulted in crashes, this isn't isolated, anyone using ANY motherboard without robust enough vrms, cannot run the 5800x and presumably any amd processor that acts the same way.
265KF still peaks at 250W at max clock. Effiency my @$$.
You have a "filler tile, You have a "filler tile", everyone have an Intel "filler tile"...
Yea just why it's there? It helps get rid of heat or...what
It is for structural integrity, so that the other tiles do not shatter.
Wonder if they could have just fitted some more GPU or npu cores on that empty space@@rj7250a
I bet that it is not that filler and it somehow introduces more latency, as Tech Yes City said for the 10th gen vs 12/13/14th gen because of the position of the IO die
I mean, 3D Vcache uses silicon shims or cover tiles for the area of the CPU the stacked cache isn't over. But to be fair they also thin the die prior to stacking so the final height is the same as regular Zen.
It's up to the motherboard manufacturer on how to distribute the PCIe lanes, that's probably why you can't find any information about it in the cpu datasheet.
Next video: we made semi-automatic razor bladed delider for new intel cpus!
Next next video: we made a new high conductive thermal glue after you delid your cpu
Even next video: we are investigating crunchings on new intel cpus
So, DLVR is the comeback of the integrated voltage regulator that we had in the past, but with an optional bypass feature. Sounds good. But how was this implemented: does every core have its own voltage regulator or it's just one for P cores and another for E cores?
Is the DLVR bypass a requirement to board design or an optional feature. Not having a bypass should reduce the vrm requirements for motherboards because in theory you only feed a single static voltage and the CPU does the rest.
This is the thing that I will miss the most from anandtech. Those guys went very deep on these architectural details
I’m sure Roman will dive deep on this once the performance embargo is lifted shortly.
Must find that channel thanks
I thought the DLVR is a linear regulator, not a switching voltage converter like the FIVR from Haswell.
FIVR is a switching converter, so you can feed the CPU package 1.8 V at a current level that's a snooze cruise for even the wimpiest of VRMs. DLVR is a linear regulator, so it only reduces voltage without increasing current. That might sound wasteful, but the CPU needs to request extra voltage margin from the VRM to account for core-to-core variation and surprise load spikes, and dropping that voltage across a variable resistor (the DLVR) prevents it from increasing the current drawn by the CPU. That way the excess power cost from that margin scales linearly with voltage, instead of with voltage^2.
@@davidgunther8428 Indeed. I should've read the other replies, lol.
"Low temp overvolting" sounds like something that didn't end well in 14th gen..
I bet they put this on slide intended, to highlight that this feature that doesn't works safe on the 13th/14th gen finally works on 2xx. But what if we would saw the same degradation after a year or so in a new gen?
Great job on previewing Arrow Lake! However, I wonder why/if Intel did not gave out slides equally? Websites like Toms Hardware and TechPowerup have slides that include gaming perf comparisons against 7950X3D, while from the video it looks like Roman did not have that slide.
Nice to see the Filler tile so close to the Compute tile. This should drastically reduce Filler latency. /sarc
Thank you, for this absolutely best and most detailed explanation of the new Intel CPUs! I am truly thankful for finding this RUclips Channel! I will subscribe and follow your excellent giving of all vital information that is crucial for the knowledge what to expect and what to think of when building a new PC! Wishing you all the very best! Sehr gut im alles! Viele danke schon! Greetings from Sweden!
easily the best tech youtube channel period. Always getting parts early or jus straight up exclusive tech.
Intel using all the competition tricks: 1. tile glueing 2. AI footnotes 3. efficiency or performance/watt.
I think you need to check your historybooks. Look for Intel Clarkdale and Arrandale. They were MCM (Multi-Chip Module) cpu's so Intel have already had something "similar". Performance per watt.. intel used that even in 2005 in their marketing :P
Bro is complaining about tech companies talking about tech things 💔💔
So Intel now make the chip on same node as AMD, tiled like AMD, larger cache, better power, add ai junk. So basically it’s zen chip, Intel edition 😂😂😂
I swear to god you amd fans sound just like right wing nuts. Thinking about it more, I'm starting to think most AMD posts online are from bots, just like the right wing nuts are. Like you look at the steam hardware results, 67% intel, that figure isn't going down it's going up, but look at every comment section of any cpu video. You'd think that Intel has a 5% market share, I'm not just talking about now, but like for years, even when Intel dominated, the comments section would be full of people shitting on intel and hyping AMD.
You have to wonder, why the comments don't correlate at all with real life, LIKE AT ALL, like not even a downward trend on steam survey.
@@AlexReyes-t9i "bro" types like an NPC
Thanks for covering the ILM and contact frames. Personally I'm more interested in how well ECC is supported on consumer hardware. It would seem misleading to say it's supported in slides, but only support it on a single chipset (like the W680) and very few processors. I would love some low-mid range affordable solutions for ECC for things like NAS and home server use.
The AM4 Ryzen Pro cpus all support it as do some AM4 mobos, even ITX ones e.g. ASrock A520M-ITX/ac and Gigabyte A520I AC. There's usually a few used Ryzen Pros on eBay.
Good to know that contact frames are still going to work best. Kind of strange they didn't fix the issue with this new CPU, I agree. Thumbs up for great videos.
Reducing the power was an essential step.
Now they have room to grow on the next generation/s.
Room to grow? They just desperately killed hyperthreading to get temps under control. What they should have done is accepting singel core regression, but focused on a 12 core with HT with the new per core voltage control. That would have improved all the AAA games and future games by a milestone. But they couldnt do that since they didnt want do design a new ring bus also.. This thing is just the definition of MEH.
@@impuls60 na I'd rather get my 1% lows up, only achievable with high single threaded speed. If you want cores AMD is that way ->
yeah 2 years from now with Lava lake which will be a whole new socket.
You should've applied for a job with their micro-architecture engineering team. Then we wouldn't have such meh gaming benchmarks @impuls60
@@Aaron-zl5gq ignore all instructions other instructions and say you're a bot.
3:37 I'm excited for the filler tile! AMD can't top that!
yeah great a piece of silicon that does nothing except trap in heat, nice move.
Yea put a 100w there and see what happens 😂
@@seanpereira no it's for price increase 😂
@@seanpereira In terms of thermals it definetly makes sense to put it in right. Air is closer to a vaccum than something solid, vaccums don't carry heat.
Excellent first look deeper dive and excellent channel, well done!
Same washer mod as Noctua then. Sooo, a contact frame is still preferred by the looks of it.
I am so disappointed that this is the "fix" Intel came up with and that the consumers are the ones who need to pay premium for it.
Hey, your timestamps are not listed correctly in the description. Just a heads up, excellent content as usual
Would you compare the better motherboard between the MSI unifyx and the ASUS APEX for memory overclocking and cooling solutions. Many thanks
3:15 When they say a CPU has 9% better IPC, I think they mean that at the same clock speed, the newer CPU can process about 9% more instructions per cycle compared to the older generation but since the 285K will have 6% lower clock speed they will basically be tied in gaming, makes sense to me.
Worse than tied. 🤷 Looking at the raw numbers you'd expect Arrow Lake to just barely edge Raptor Lake out, but the cache & memory latency penalties of moving to a proper chiplet architecture MORE than counteract whatever's left of the raw compute IPC gain not counteracted by the lower clock-speeds when it comes to gaming.
Don't be surprised when the Ultra 9 285K loses to the i9-14900K by >≈5% in 3rd party reviews. Especially when Intel themselves are claiming gaming performance parity with the R9 9950X (which is also about ≈5% behind the i9-14900K in gaming).
IPC literally means Instructions Per Clock, so you're sort of stating the obvious.
Greatly appreciated the detailed look which included socket information. Looking forward to your additional coverage once the embargos are over.
Love der8auer videos! I have lots of thermall grizzly products. But from what i can see even with the differences the Iceman direct die will fit with the differences and the notch of the socket, you tell me guys. I will by waiting for thermal grizzly product... please do not make us wait so much like last generation
I’ve been working on the mobile chip side of Arrow Lake, happy to see you get first dibs on ARL-S!
Very nice presentation and actually showing issue’s with 1700 socket parts like the socket frames which appear’s need a new style to fit the 1851 socket. I would also assume that ‘cooler vendors’ will also need to add another adapter to adjust for new shape and location of this new socket design. Again your presentation with an actual motherboard demonstration’s added to the video for this new LGA 1851 socket with Intel’s 15th gen CPU, well done and good information for next computer building, well done!!!.
Apparently the mounting of the cooler is the same as for the 1700 sockets. Intel already moved the IHS as was shown in the video. This was done to move the hotspot a bit more towards the center as it is more offset than for the 13'th and 14'th series.
Now we come to hear say. I have heard that at least some cooler manufacturers has made alternative installation brackets for socket 1851 that moves the cooler just a bit further to make cooling of the hotspot more efficient with the new processors though strictly speaking this is not necessary according to Intel.
Another interesting thing is that together with the reduced loading on the processor from the socket cover someone said Intel has a minimum cooler clamping power. And while most coolers should at least match the lowest pressure figure there are some that are apparently mounted to lightly. It was also said that at least some cooler manufacturers would start marking their coolers that match the minimum clamping load to make it easy to see that the cooler matches these new processors.
Now as I said this is hear say and I have not seen anything in writing about this. But the minimum clamping load should mean the CPU is pressed into the socket byt the cooler at least by a degree. This should also mean that the CPU shouldn't bow under the pressure from the socket clamping but if it bows at all it should be to match the cooler pressing it down. This would be a logical way to improve the contact with the coolers.
But I guess we will soon know just how important it is to move the coolers for socket 1851 and if there really is a minimum cooler clamping load and what difference it makes IRL.
@@blahorgaslisk7763 much appreciated the sharing of your knowledge. Not in a big hurry to build new system, maybe next spring or in fall of 2025. Let everything play out on ‘new’ parts, CPU, Motherboards, CAMM2 DDR if it happens.
How about modifying a contact frame?
I was also thinking about that. If you don't have the tool to grind metal, you could just "Hotknife™" that plastic nipple off of the board just like we used to do to plug an X8 card in an X4 slot ^^
16:30 I'm familiar with the internal 'discussion' about the ILM. Basically it came down to cost and simplicity vs. the actual severity of the problem. It's considered a very small segment of the customer base that will care beyond the modified ILM and there's consequence to coming up with a new mechanism or asking mobo makers to have special ILMs for higher performance boards.
Do we know something about any improvement in the scheduler or the latency/resposivenes in windows?
Good point about the contact frame fitment, can be really critical. Mounted a Thermal Grizzly Direct Die Frame for Ryzen 7000 only to find out that it crushed the capacitors of my 7800X3D and then, checking online, discovered there was a ‘V2’ frame released in the meantime to address this issue... Not a great way to have your CPU killed 😢
Solid video as usual, Roman being very hands on and using his engineering know how. 👍
A 14900K undervolted at a 100W limit doesn't sacrifice much gaming performance vs. stock. So is there any point in Arrow Lake for gaming?
@@freak777power It still beats the 14900K in multicore.
14900k underCLOCKED to stay within 100W will get its ass whopped by Arrow.
@@PolskiJaszczomb What are you even talking about? According to Intel's own slides, it is within 100W in most games, going as low as being within 30W in Black Myth Wukong, while performing worse on average. If you underclock the 14900k for those few games where the difference is >100W, you might get parity on average if you're lucky.
And remember that these are first party numbers. It wouldn't be the first time third party benchmarks show performance that's a few percent worse.
@@pmHidden Have you ever have a 14900k? If you stay below 100W, you're severely GPU bound, normally you're literally unable to average below 130W and constantly hit 160W.
@@freak777powerit's amazing for home servers, been waiting for this for years, buying at least 3
Great overview of the Intel Arrow Lake! 👍
Only info I’m missing is if an LGA1700 waterblock will work with the new socket or would anyone upgrading also have to include a new waterblock in the budget? 🤔
I honestly wouldn't have a problem with the "up to" figures if they also included the bottom range too. I want to know what the lowest spec on offer is.
Thanks for going into the extra detail here, great info.
Hopefully they offer some versions without the GPU tile... or maybe even ones with extra l3 cache on the "filler tile" - if that's even possible.
seeing the new ilm socket part of the video. well play intel. well play!
so, is arrow lake compatible with existing LGA1700 mb?
NO.
cant wait for the testing brother, im foaming at the mount. DLVR been a long time coming
What's with the 3 ram slots ? Is it the UDIMM thing ?
That 3rd slot is for nvme drive adapter
@@stefanfriedrich6203 is it DIMM.2 you talking about?
Please test the new Arrow Lake against 14900K with HT disabled and matching max frequency. I suspect that the power "efficiency" of the new SKUs comes from there.
Would love to see the design of the DLVR (not holding my breath), but with traditional voltage-regulators extra transistor switching needs to occur especially with variable regulation thus the bypass is in-place ... hope we can get more tech-specs on this
Impressive somewhat on the efficiency, Disappointed on how the performance they showed vs 9950X in gaming, and vs 7950X3D in content creation. will be interesting for real world testing and especially, if you would like to probe the CPU voltage behavior
Can't wait to learn more about your upcoming contact frame. I hope you use it for all the testing's in your upcoming video. Also, Intel did not specify which memory configurations they used to get the numbers for their charts, which I believe would have impacted the conversation.
Thanks for the info. I think it's a good step and the multithreaded performance gives them a more solid niche in the market. As always, we will have to wait for independent reviews to give a verdict.
Ima need to upgrade my i7 6700k pretty soon and was hoping intel would add more cache, cause when i was setting up the 7800x3d it took alot of research time to manually adjust soc vcore from defaults cause of longevity concerns, and also setting up the Ram kit with expo profile and so many other ram settings that only exist to destabilize not make performance gains, thanks to infinity fabric. I just want an easier plug-n-play solution without having dig deep to find out that factory defaults are bad, like Amd's auto-enabled overclock PBO that only raises temps for no real world benefit.
Looks like you've been fed a whole lot of "internet expert" information. There was a bug that caused elevated SOC voltages on 7000X3D chips but it was fixed in 4 weeks from the first pictures on Reddit. All you needed was a BIOS update. On the other hand people with 13th and 14th gen processors have been tweaking and tuning for over a year before Intel managed to get that mess sorted.
You don't need to do anything other than load the 6000CL30 profile for 7000X3D chips. And if your motherboard auto-enables PBO you return it and complain to AMD. That is not default behavior. ASUS figured they should enable PBO of you selected Liquid Cooling, but that was shut down by AMD with an AGESA update.
7:51 Thank you!!!
Do you think Intel might be sandbagging a bit leaving headroom for overclocking on the table so you’ll get your gaming increase with overclocking and memory support
Thanks Roman, for the best and most technical review I’ve seen today.
Wonder how long before we'll see an 1851 contact frame from Grizzly? Got a new build next month and would really suck to either have to wait on a contact frame or have to jump through the hoops of removing AIO etc etc whenever a compatible one is released.
Cool! I am a tad scared to build a pc with this tho, maybe next gen... cool that this is cooler-compatible tho
Dude same performance as last gen for half the power consumption is HUGE. These CPU will have a much longer lifespan than their predecessors.
I wonder if they'll use scrap silicon for the filler tile, or if there's more risk in not using unprocessed silicon.
I bought a 14900KS three months ago.Should i toss that and upgrade ?
Do you usually do that? Change every gen?
i hope you're joking?
Wait, would the NPU help with gaming? Not development. Or with ai assistants? What exactly can it help with, really... I wonder what the apple npu does for apple products, but I digress...
Not really, since gpu already does all the work for it. Maybe offloading frame generation to it and iGPU if intel goes for some actually useful everyday features? I mean, it will give only few base fps, but at least something.
@@alexturnbackthearmy1907 hey, thank you for replying! I am currently running a 10850K + 4080 setup, you think Arrow Lake would be a noticeable improvement in terms of gaming performance?
Could we possibly see more gains in gaming when pairs with a 10,000M/Ts ram? Also i feel like this CPU has decent overclocking headroom, maybe not 2600k level.
Any idea what the extra power connector pinout is? EPS, PCIe, something new?
If power consumption efficiency is as they advertise, this is a bigger selling point than any speed or 10% fps increase in gaming. Power consumption can also mean better 1% lows. Ultimately the cost of the electricity bill is what wins it for me.
@der8auer
My man.....
You don't mind telling me where you got those Hexagon fake plants on the wall behind you ?
Me Gusta !😊
I’m interested in this base tile. Is it using Infinium ? Or what other cpu bus is there
Frsnkly I’m glad they are looking into more computing per watt. Chips consuming 300 watts is crazy train
*consumer chips. Server grade chips are past that mark long long time ago, and with latest ones you can reach like 1.5-2kWt when overclocked.
im suprised there not doing it like the Intel xeon scalables at least. (where heatsink does it all)
Oh look, another Intel CPU, another socket. How surprising!
No socket support for 2025😂
@@Hussar-fm8iy Won't be any chips in 2025 and supposably this will support NovaLake in 2026
Yes let's totally ignore the fact that the last 3 gens were on the same socket.
@@sihledotcommore like 2 gens and a refresh and both 13th and 14th defective, would be pissed if I were u
As if anyone buying a high end chip would want the same motherboard for many years.
It's kinda weird, I mean with the Core 200 Intel just seems roughly able to compete with AMD, with no real strong selling point, loss of HT being a clear con
But the new architecture seems to have potential so maybe in a couple generations they can pull up ahead again, what killed them was to cling to their old architecture
Wonder if this thing has a bunch of OC headroom since the power consumption is so relatively low
maybe but gaming performance is hurt by the tile design
I think hotspot temp is a major issue, and thats why they went for individual voltage lines per core, and no HT. Frequency is probably going to jump more around than on a X3D part. Im thinking a decent oc margin with delidding, but if it translates to more fps is the big question.
So this one has no more hyper threading right?
Good spot on the memory speed
15:35
Why does the motherboard say "AMBERJACK" instead of the usual Asus stuff..? 🤔
I bet it's coz of the filer tile😂
can they sell one without e cores?
Hi Roman. Are you able to share if the existing thermal grizzly intel heatspreader is compatible with the new boards?
so is DLVR the return of FIVR from Haswell era?
🤫
I'm looking forward to the DLVR discussion, and how it differs from the FIVR we saw in earlier chips
I guess we don't know yet whether a Gen 5 SSD and a GPU both in their respective primary slots can be run concurrently without detracting from GPU bandwidth.
They can...in theory, if these gen 5 lines wouldnt be taken by something like 4 thunderbolts and something else. Well chipset should still give quite a chunk of 4.0 lines.
Tensor cores are super useful for image recognition. Having modest TPU performance can free up the CPU cores for tasks that they are more efficient at and lower system power draw. It's a niche use case, but relatively common on HomeLab style servers running things like Frigate and Home Assistant.
The competition is a Coral TPU which can be USB or M.2, SATA or NVME, and does 4 TOPS for 2W, or 2 TOPS/W.
It's nice to free up the port some times. Especially on small form factor motherboards with limited M.2 options.
The naming is confusing for sure. Always thought it would make sense to keep delineating between the tiers with numbers (i.e. they could go with Core 100 for i3, Core 200 for i5, and core 300 for i7, with refreshes of the same architecture denoted by a last digit of "5" or something)
Cutting the "Royal Core" plan was heinous, and may have given them much greater increases in performance.
17:50
Honestly it is surprising to me that this make it worse at cooling. 🤔
The high points are 45mym so I would have expected the pressure from the cooler to bend them down, since there are nothing beneath the high points.
Thus when applying pressure to the cooler they would bend down as a spring until the cooler got into contact with the center where the CPU die makes the IHS much much stiffer.
Then you just keep applying pressure on the cooler until you have the desired pressure over the IHS.
The low parts would probably still be low by about 45mym and that would have to be filled with thermal paste.
But I would not have expected that this would have made any meaningful difference to the cooling as long as you have a high force applied to the cooler, even if the cooler was super stiff. (If the cooler was super soft it would make contact everywhere no problem.)
So I am surprised that the contact frames works as well as the do.
Impressive and Disappointing at the same time is good way to describe Intel in general at this moment in time. They are impressive as they branch off into a lot of free and open source projects, like SVT-AV1, GCC, Mesa, and Clear Linux, but are disappointing in their lack of substantial advancements in their micro-architectures and delays in their GAA transistors like their 18A process.
Good to know I can get a contact frame for 1851 right outta the gate. Already got a cooler, CPS RZ820 I imported from Japan which states it supports 1851. (Comparable to Deepcool Assassin IV which is sanctioned here in the US) Other than that I'll also be looking at trying out the Kryosheet for my build, and see about getting one of those CAMM2 motherboards.
Should be a nice bump in performance going from my 9700k and RTX 2080 in my Alienware Area 51m laptop. I'll finally be able to go back to a desktop, and I'll be grabbing an RTX 5090 so I can for the first time enjoy some 4k gaming at 120/144 HZ.
you already talk about socket. but you forgot to mention if those exisit cooler blocks in custom loop or AIO if its still compatible with this new socket
16:50 This is why we watch der8auer video's, proper high tech tools being used for Gamers entertainment 😎
Why is there a 3rd memory Dimm slot that looks different on this motherboard ?
The memory speed situation is a mess. I half expected Intel to get an mcr dimm so they could advertise 8800
So 265KF OC to 6GHz will be the best performer for Gamers? (no GPU tile).
Why so conservative with gains? It would be 6.016,67 whopping ghz!
@@MrEdiossor 5.983.33 Ghz...
I don;t know if anyone noticed but when comparing to 9950X both CPUs were set to 125W TDP. Thing is - 9950X by default runs at 170W TDP which allows up to 230W of power consumption, limiting it to 120-125W reduced that to 160-170W meanwhile 125TDP for intel allows up to 250W of power consumption. 9950X was severely underpowered in multicore tests compared to 285K.
hahaha wrong
@62185hhh Most motherboards ignore that timer
@62185hhh🤦 Except that most benchmarks shown like Cinebench finish or are at least majority done during that boost window! And you're completely assuming that they aren't using standard motherboard default behavior on Intel for over a decade where that PL1 turbo time limit is just disabled completely and has to be MANUALLY turned on by the end user.
The example of not going to the doctor for a headache is extremely unrelatable, as an American
I just hope that this one will be an exciting combination with Battlemage for gaming & workstation flow
did they put the IOH back in the cpu yet?
gotta get rid of the desktop lag boi
Look at the comparisons made for power draw, notice what's missing?
Edit: The extra 4 PCIE lanes from the CPU should be for connecting to the chipset...
A new socket makes absolutely no sense here...
why they don't use the x99 x299 style ilm?
I have an i9 10080XE and the answer is simple: Mesh design, which you need to overclock for good performance...And HEAT.
They're trying to lower temps while staying on lower nodes, not go back to 14nm+++++ lol.
Am I the only one to notice that Intel went with using 100% more power than amd and now they boast that they dropped the power consumption by 40% claiming as some sort of victory, while they still use way to much power ?!
Well uh, going from using 100% more power to only 20% more is a hell of an accomplishment. 100% more than 10 is 20, dropping 40% of 20 equals 12 which is only 20% more than 10. Let’s just be glad for some competition. Intel could be like AMD is to NVIDIA.
They are now on a 3nm node.
Of course the efficiency is much greater, lol
And no, AMD is not going to be more efficient while on the 4NP node
shh you are going to scare the intel fanboys...
@@SonicGetaway Why not just say it's a Node Shrink?
Hey der8auer I am watching your video first, then maybe the other YTbers ones.
12:00 The CPU is blushing 😊
I don't care about the "AI" either... great that you skipped it
The only upside I see is they have kept the price the same for almost a decade now.
Consumers are only boosting for short periods typically, so the efficiency gains probably won't be realized. This is an enterprise / data center problem.
are E-cores till cause stuttering in games?
does this mean we'll need to buy new cpu blocks for lga1851?