If you look to the majority of PC products? Sure - it didn't take but essentially all other devices are on the cloud. Every phone and business device is cloud-first or use the cloud to back up.
Yup the AI deskmat, AI stapler & the AI paperclips are revolutionary @ Computex (or was that being sold by a guy in a trenchcoat in the dirty back alley behind Computex ?) LoL ;)
I always see Thermal Grizzly as a sponsor of GNs or your videos and i can only say, thank you. My current system has a vertical gpu installed which would slowly, over time become hotter and hotter. Hotspot temps reached 110 °C which led to 'fun' side effects. I repasted the gpu once since the first time i noticed these high temps and now, a year later. Turns out that due to the vertical orientation the gpu suffered from thermal paste pump out. After opening the gpu the thermal paste i applied a year prior was everywhere except the gpu die. After applying the graphine sheet temps droped by 30 °C and haven't gotten higher than 85 even under heavy load (Furmark). Thank you and also thanks Roman for your product.
I don't see the benefit for a home-only system. But I can see it being really useful for these handhelds. The ROG Ally X docked with an eGPU for example, like a PC version of the Switch. No, it won't get 100% of the bandwidth of an RTX 4090, not even close. But paired with an RTX 4060, wouldn't be a terrible setup if you play it on the go a lot.
You should try with these aliexpress Oculink dock. I have both docks (tb4 and oculink) and its so much better with my Oculink. Running a RX7800XT in my dock, it just blows my rtx4070 in my laptop (Slim 7i 2023)
@@Patrick-y4d1z why dont they use riser cables like when you want to mount your gpu in a different position? Maybe I am overlooking something like the max length perhaps, never used one myself.
@@Patrick-y4d1z Technically it can be an option for ultrabooks too. Like having a Framework 13, when you're at home, you plug in the GPU and blast away at some demons. But tomorrow you only take the small laptop with you to work and / or travel and enjoy the small weight, footprint, good battery life and thermals.
Its so sick and embarrassing now that AI is mainstream and the consequences is the average joe who doesnt even know what "inferencing" or LLM means now suddenly spouts stuff about AI and used it as their marketing tool, similar with how the word "gaming" was in pc parts a few years ago
The motherboard chipset is not the limiting factor! You’ve got the 8000 APUs easily running DDR5-8000 on X670 and B650 already. So the big question is whether the 9000 series will feature a memory controller that is more capable than the 7000 series.
Probably not, as up to current information, Ryzen 9000 will use the same die as the Ryzen 7000 series. However no-one knows if AMD makes newer revisions with error corrections and slight tweaks, but those IO-Dies would also reach Ryzen 7000.
X670 2DPC motherboards are unable to provide reliable RAM stability, I mean boot to boot stability with Karhu at least 30000%. Go on OCN for more details. Only Asus Gene (1DPC board) can guarantee reliable stability 1:2 8000. Maybe Aorus Xtreme and Master are the only 2DPC boards which can provide 8000 1:2 stability.
Got a 7600 and 7700 on a Jginyue and Colorful B650 using three different Hynix kits one with expo. All will do 7600mt/s cl36 and pass Memtest86. Haven’t messed with infinity fabric memory controller dividers yet. All are 6000 or 6400 kits. Oddly time spy cpu score is best at 5600mt/s likely due to memory controller being at higher clocks. Also 6000mhz does not boot on any combination of parts. I advice folks go with Hynix 5600kits.
I'm not as interested in ultra speed as whether I can populate my board with 4 Dual sided simms (128GB) and have it run at say 6000. Need the extra memory for UE5/Cad programs.
This. Current AM5 mainboards can already boot over DDR5-11000 mem and run ~8800 C38 bench-stable if an APU is used. Still - while IMC seems to be the limiting factor for most configurations - it's always dependent on the board (and to some degree on the RAM, tho less with hynix A-die) as well. Look at how the Z790 apex encore just laps any other board on the market when it comes to mem, even if CPU and the RAM itself is the same.
My suburban Catholic parish is running a survey at the moment. It is like five questions long (what is working, what can we do better type stuff). Someone approached the person organising it and politely blasted them for not "using AI for the survey"...
The Light Base 600 horizontal layout by Be Quiet looks like the ticket for me. A horizontal mount would mean the biggest, baddest and heaviest of all video cards will not require support and places no stress on the motherboard. Likewise with an air cooled system the unit can be enormous and heavy without strain to the motherboard.
I'm looking forward to a PC upgrade, but I can't do it until I see all your testing! Eagerly looking forward to the next few months of content. Thanks guys.
I was 100% against the whole rear connector motherboards but seeing as how more and more cases are being made to be compatible with both, it's becoming less concerning. I like the idea, I just want the compatibility to remain. Like how I know that if I buy an ATX case, my ATX board will fit. I was initially concerned about placement of the headers but case manufacturers seem to just be opening up the whole area around the edges which makes sense. I just hope that everyone gets on board so everything is compatible. Granted, small form factor cases won't likely have this option but at that point you already have to get an ITX or Micro ATX board anyway. But if all the new ATX cases have support for both configs, I think that would be ok.
An oculink EGPU isn't nearly as limiting as USB4 and Thunderbolt 4. A 4090 looses at most 25%, and it can be a much smaller loss in higher resolution/lower frame rate situations. Anything around the performance level of a 4070 TI or below should have little (single digit %) to no loss with an oculink connection vs direct PCIEx16 connection.
The higher clock speeds will arrive when that new desktop/laptop memory form factor for ddr5 is released. It’s less about the memory controller and more about improving board/connector signal integrity. So the delays relate to the availability of the new CAMM2 (acronym escapes me) DDR5 module.
To me BTF only makes sense with light colour cases and/or white motherboards to hide any colour of cables better. What I don't like the most about BTF form factor is the rear facing connectors will get covered with other cables, limiting clear access to specific cables, and them possibly getting mashed down and pins getting bent.
3:50 USB 4 (and Thunderbolt 4) eGPU's are heavely bandwidth limited, there are eGPU's that use OCULINK 4i and 8i connector that effectively doubles bandwidth compared to Thunderbolt solutions. The difference to full fat PCIE 4.0 x16 slot isnt that big since even current gen cards do not use full bandwidh of x16 slot, Oculink 8i (it really is just PCIE through Oculink connector) would give PCIE 4.0 x8 bandwidth for eGPU's. Oculink 4x gives PCIE 4.0 x4 bandwidth for the EGPU to use and difference to GPU performance is only something like 5-15%, where Thunderbolt eGPU is heavely limited, sometimes getting only 30% of the framerate in some games. Thunderbolt has latency issues yes but Oculink does not, Oculink does not need any controllers to work just straight connection from PCIE lines through Oculink to eGPU adapter with build in pcie re-drivers (signal boosters, no latency issues).
Derba8er and others also made a video from the GSkill Computex stand. And here they had Ryzen 8000 G series running with DDR5 Modules with 10600 MT speed. So I think Zen 5 will support adequate memory speeds
Those speeds were achieved because the ryzen 8000 are monolithic designs. it is essentially the same reason why intel's 12/13/14 cpu were generally able to cope better with faster ddr5 than the 7000 series. the 9000 is a chiplet design and while I am sure it will handle faster ddr5 better than the 7000 series, there is a limit on how much it will be able as the I/O is , according to AMD, essentially the same.
Those are monolithic APUs and they always supported higher memory speeds since they dont need infinity fabric to connect 2 seperate chiplets. Same way how you could easily run ryzen 5700G at DDR4 4400 while the chiplet ryzens like 5800x couldnt reliably go over 3600/3800 in 1:1 mode I dont expect ryzen 9000 to go over DDR5 6800
I've always seen VRMs like high wattage speakers, you probably will never use the max capacity, but the smaller percentage is utilized the more efficient and cleaner the signal is.
I think Project G-Assist has potential to be a really cool feature. Imagine an MMORPG assistant that can role play in character. Imagine it act like a space ship AI assistant in a space sim game.
Your rundown on all the AI improvements and enhancements left out a new line of AI coming to Australia. Just when will the AI enhanced Drop Bears be arriving down here?
About here 9:23 - I really never heard of the inverted cases. I build the 1st computer for personal use (built many for business) and here I have my beautiful PC setup... and it faces a shelf and not me for me to see all of it's tempered glass glory. LOL
I really hope that BTF doesn’t become predominant. It’ll be really annoying if the overclocking boards start using this, it’ll basically ruin test benches.
When you mentioned AI memory you first showed a CAMM2 module, and I figured that OKAY, CAMM2 might be better for local LLMs than other RAM. But then the view widened to all kinds of other memory and storage and I realized it was just executive vaporware. I think there might be some genuine use-cases for ML even if it's just used for the design (which executives seem to use as an excuse to slap on AI on their products these days). I think there were papers on using AI for CPU designs, and I can also imagine that it would be helpful with designing the traces of high performance data busses, like DRAM. Generally today's generative ML can be surprisingly effective when you need something technical that is somewhat creative. But if I were to design such a thing with 'AI' I'd certainly not announce that to the world, but keep it a trade secret (if I were an executive, as an engineer I'd like to write a paper or at least blog post on it).
It's only ever worth investing in eGPU compatibility if you don't have the budget for a good midrange GPU right now, but want one in the future. Buying a miniPC with USB4 support right now can be worthwhile, as it will allow you to use eGPUs in the future if/when you need a significant GPU upgrade, but if you need/want a powerful GPU right now, you should just get it now. An eGPU won't save a significant amount of money in the long run, because you'll need to buy an enclosure or adapter in addition to the graphics card, even if the card itself will be cheaper in the future.
8000 series with revised controller can already run 10k+ ram on existing boards so it’s cpu generation & not chipset gen that’s restricting speeds. But, the 8000 series responds so well to high ram speeds because it’s benefitting the gpu cores even more so can’t imagine ‘the sweet spot’ for straight cpus jumping massively. Total guess but i keep reading/seeing g series running 8400 so I’m just assuming 7200 (or thereabouts) will be the new 6000. & that’s well within spec of existing decent B650 & X670 boards.
About the eGPU topic, I could not find any valid test video from Jarrod on the topic. There were 2, one with a 2core CPU in a laptop and one comparing laptop eGPU setup to desktop dGPU setup which is not comparable since there are some constraints outside the eGPU connection between the teste configurations (RAM, cooling, power budget for CPU etc ...). Also the laptop with the superior CPU from the 2 laptops used having worse results just confirms that there are more factors at play here than just the eGPU connection ... but I bet you'll do a proper apples to apples test just to prove the point ...
Zen 2 and zen 3 used the same IO die, and their memory was speed was basically the same with zen 3 only going slightly higher in 1-1-1 ratio. I’m guessing zen 4 and 5 will be the same, roughly similar ram speeds with slightly faster infinity fabric since they use the same IO die.
If external GPUs have too much latency when distanced from a given device, why hasn't anyone made a lego eGPU chassis and portable minui-PC case combo design? Couldn't you just have a setup where you pull out a port plug to expose the PCI-e connector and then snap the two together so they become one solid frame. You would have a detachable GPU for travel and a clean setup for when the two are put together. Seems like easy money if a case manufacturer can get a custom motherboard layout with say the PCIe connector on the reverse side so it can be flush with the case wall.
I love the idea of CAMM2 memory, but the only motherboards I've seen only have one module of space. That doesn't save space over DDR if I can have 4 slots of DDR, instead of one slot of CAMM2. You must have the ability to upgrade your memory instead of replacing what you already have.
motherboards can easily support 4x 4high USB type A stacks, Mouse, keyboard, headset, camera, controller, audio deck etc is all type A. So why do we wanna have half that in type C ports? for plugging in our 8 phones?
I got rid of my a4 h2o sff case. Though temps were within tolerances the thing just blew out hot plastic smelling air and made my room stink. Swapped to a mff with some t30 fans and it’s like a mini fridge
Question: if I build my new PC with an AI PSU, AI memory and an AI case, but then buy another bigger AI case and put my newly build AI PC inside this big AI case, how much AI performance will I gain? Can I expect 10% of more better? And are all those AI components compatible with each other? Can the AI mouse communicate with the AI power cord?
So nearly every HUB video has the Thermal Grizzly Kyrosheet advertisement at the start ... yet I cannot find any testing done on it ? Not on HUB... shit not anywhere tbh (only "lower" channels) Plenty on the Carbonaut but I think that's an older "lesser" product
@@Hardwareunboxed Oh yeah there it is thanks. I think I zoned out of that review as the card is beyond affordable, was just pretty to look at =) Will you be doing further testing/reviewing on CPUs ? Do you think Kyrosheet is worthwhile? I'm about to swap from a 3600 to a 5700x3d and I'm out of paste *Just looked it up.. only $11 for paste vs $23 for sheet.. might as well go sheet right for zero maintenance which I'm quite fond of
I can't see why CAM2 is going to take off in PC's, as it erases modulation of RAM. It's a great option to pick up 4 dimms later on in life, as each spare dimm can run a PC.
For video editing on the go it will probably just make more sense to get Macbooks, lol. As far as I know, AMD has made their RDNA2 (and above) dedicated and Integrated GPUs to work together so maybe they could manage the same for external. It might need to use OCuLink instead of Thunderbolt/USB PCIe for better latency. But it will still cost fair amount for the an enclosure or be really expensive with all in one external GPU solution (especially in the long run). Framework does have the dedicated GPU on their 16 inch model being something you can detach, so that might also be a way to add more GPU power. But currently it is limited to RX 7700 and not something getting close to desktop dedicated GPUs.
i will say, that computex for me was underwhelming, all coverages i saw were mostly pc cases based on 4-5 older cases, like 80% of what i saw, another 15% would be cooling solutions funny thing is that i was most excited when GN went to thermalright air coolers, i think i had too many cases at that point tbh i started to forget about things like camm2 and any news, because there were hardly any monitors, all there was remaining was AI keynotes i think TCL dropping 1000hz 4k display and all their technologies had more tech news than whole computex at this point
The worst thing about this AI marketing spam is that it is working on leadership of organizations like mine. They've moved from pointlessly demanding we put things in the cloud that don't belong in the cloud, so now insisting that we start using AI. For what? How? Doesn't matter, they just insist that we go AI.
You know what would be awesome? An actual _Back to the Future_ Edition of a BTF mobo, kinda like the Sonic and Evangelion-branded boards, released on the first film's 40th anniversary. I would buy that in a heartbeat.
I don’t think it’s going to matter, I was able to achieve 7800MTS Y Cruncher VST stable on an MSI MAG B650 board, lots of others seeing success with really high speeds providing they have memory which can do so
"Except for a company like... " My brain: Noctua? "Noctua." Hard to beat a company knows exactly who they are and sticks to the roots. When we get a noctua AI fan... The robots have taken over and they want humans to worship silicon.
Too soon to make assumptions about the X870 and memory speeds - if the new memory speeds are supported on X870 on the Zen 5 already, but the boards are coming later for whatever production reason, they won't tell you either beforehand because people would waited to see/buy later and not get Zen 5 CPUs (and older chipset MBs) right away. And I'd expect AMD being fine with a gradual release and press with Intel coming later to party with Lunar Lake.
This video proves beyond any reasonable doubt that you are already AI Unboxed, powered by your AI overlord Balin. He appears human, but he’s not fooling anyone.
Welcome to AI Unboxed. Today I'll be reviewing AI screw from Asus, it helps the SSD stay still in the case and let it loose when a screwdriver unscrews it. A bit expensive to my taste and I don't see much value unless you really want to be screwed by a company that was once top notch in clients service
evolv x2 is the best case by a long way .. i hope it doesn't only support back connect though.. cos there are literally 0 of them for sale in my country and i just bought a new motherboard
It was the same thing with AM4 the B350/X370 were horrible for memory speeds I still haven't purchased an AM5 motherboard because I knew it would be the same thing they had 3 chipsets for AM4 and AM5 will likely be the same so I'm waiting until they come out with the 3rd gen AM5 chipset then I'll consider upgrading from AM4
the craziest thing is it isnt even AI, its pattern recognition. It has no intelligence. Its still just an ordinary computer. takes input, calculates output.
A few years ago I was planning a trip to see these guys in Australia. I was applying for my visa and I was asked if I had a criminal record. I told the immigration officer I didn't realise you still needed one.
I'm having flashbacks to 2012/2013 and everything being "Cloud"
What about the IoT - Internet of Things 🤣
If you look to the majority of PC products? Sure - it didn't take but essentially all other devices are on the cloud. Every phone and business device is cloud-first or use the cloud to back up.
Remember when all the things were 5G
Before AI, many of these features were called "smart"
Icloud turbo gaming X AI.
No AI RGB? AI thermal paste? AI SATA cables? There is still so much untapped potential to put AI in front of words.
You joke but there’s actually an “AI thermal paste” just announced by Cooler Master 😂😂 as you may imagine it has nothing to do with AI
I can remember when AI was a vetenary procedure for cows.
Yup the AI deskmat, AI stapler & the AI paperclips are revolutionary @ Computex (or was that being sold by a guy in a trenchcoat in the dirty back alley behind Computex ?) LoL ;)
@@einsteinx2It was a translation error (luckily). Supposed to say "Thermal paste for AI CPUs", like the Ryzen AI etc
AI asses
13:29 - Steve you missed a trick to say "opened my A-eyes".
😂😂
Underrated comment.
I always see Thermal Grizzly as a sponsor of GNs or your videos and i can only say, thank you.
My current system has a vertical gpu installed which would slowly, over time become hotter and hotter. Hotspot temps reached 110 °C which led to 'fun' side effects.
I repasted the gpu once since the first time i noticed these high temps and now, a year later. Turns out that due to the vertical orientation the gpu suffered from thermal paste pump out. After opening the gpu the thermal paste i applied a year prior was everywhere except the gpu die. After applying the graphine sheet temps droped by 30 °C and haven't gotten higher than 85 even under heavy load (Furmark). Thank you and also thanks Roman for your product.
4:32 ye egpu succ, maybe thunderbolt 5 will improve it a little but we'll see
I don't see the benefit for a home-only system.
But I can see it being really useful for these handhelds. The ROG Ally X docked with an eGPU for example, like a PC version of the Switch. No, it won't get 100% of the bandwidth of an RTX 4090, not even close.
But paired with an RTX 4060, wouldn't be a terrible setup if you play it on the go a lot.
You should try with these aliexpress Oculink dock. I have both docks (tb4 and oculink) and its so much better with my Oculink. Running a RX7800XT in my dock, it just blows my rtx4070 in my laptop (Slim 7i 2023)
@@Patrick-y4d1z why dont they use riser cables like when you want to mount your gpu in a different position? Maybe I am overlooking something like the max length perhaps, never used one myself.
@@Patrick-y4d1z Technically it can be an option for ultrabooks too. Like having a Framework 13, when you're at home, you plug in the GPU and blast away at some demons. But tomorrow you only take the small laptop with you to work and / or travel and enjoy the small weight, footprint, good battery life and thermals.
If I’m not mistaken, Rog ally has a dock available ( sold separately).
@Tim the logo is not big enough. I appreciate you taking feedback since the last video but we can go bigger.
My favourite AI thing of computex is AI thermal paste
COLORED A.I thermal paste!!
its just normal thermal paste just dyed and has a modified FORMULA
Its so sick and embarrassing now that AI is mainstream and the consequences is the average joe who doesnt even know what "inferencing" or LLM means now suddenly spouts stuff about AI and used it as their marketing tool, similar with how the word "gaming" was in pc parts a few years ago
awwww. poor baby.
@@mentatteg and your problem is ?
@@mentattegkeep yourself safe 😇
@@Eleganttf2 I'm worried about you is all.
Ayyyy Eyyeee
Guys, did Roman ask you to say AI to the camera? 😀
Blink twice if you need help.
theyve said it hundred times in their own videos...
heeey 1 mill+ lets gooo HUB!
4:25 for that amazing pool shot
The motherboard chipset is not the limiting factor! You’ve got the 8000 APUs easily running DDR5-8000 on X670 and B650 already. So the big question is whether the 9000 series will feature a memory controller that is more capable than the 7000 series.
Probably not, as up to current information, Ryzen 9000 will use the same die as the Ryzen 7000 series.
However no-one knows if AMD makes newer revisions with error corrections and slight tweaks, but those IO-Dies would also reach Ryzen 7000.
X670 2DPC motherboards are unable to provide reliable RAM stability, I mean boot to boot stability with Karhu at least 30000%. Go on OCN for more details.
Only Asus Gene (1DPC board) can guarantee reliable stability 1:2 8000.
Maybe Aorus Xtreme and Master are the only 2DPC boards which can provide 8000 1:2 stability.
Got a 7600 and 7700 on a Jginyue and Colorful B650 using three different Hynix kits one with expo. All will do 7600mt/s cl36 and pass Memtest86. Haven’t messed with infinity fabric memory controller dividers yet. All are 6000 or 6400 kits. Oddly time spy cpu score is best at 5600mt/s likely due to memory controller being at higher clocks. Also 6000mhz does not boot on any combination of parts. I advice folks go with Hynix 5600kits.
I'm not as interested in ultra speed as whether I can populate my board with 4 Dual sided simms (128GB) and have it run at say 6000. Need the extra memory for UE5/Cad programs.
This. Current AM5 mainboards can already boot over DDR5-11000 mem and run ~8800 C38 bench-stable if an APU is used. Still - while IMC seems to be the limiting factor for most configurations - it's always dependent on the board (and to some degree on the RAM, tho less with hynix A-die) as well. Look at how the Z790 apex encore just laps any other board on the market when it comes to mem, even if CPU and the RAM itself is the same.
My suburban Catholic parish is running a survey at the moment. It is like five questions long (what is working, what can we do better type stuff). Someone approached the person organising it and politely blasted them for not "using AI for the survey"...
what the heck
Congrats on the 1M sub mark gents XD Onwards and upwards!
Your coverage is always appreciated guys. This channel is one of my go to's when it comes to PC hardware.
The Light Base 600 horizontal layout by Be Quiet looks like the ticket for me. A horizontal mount would mean the biggest, baddest and heaviest of all video cards will not require support and places no stress on the motherboard. Likewise with an air cooled system the unit can be enormous and heavy without strain to the motherboard.
Lol! The AI segment was hilarious! SaRcAsm galore! 🤣
AI unboxed is really catchy :D
Got my AI thermal paste on order!
Just a hint.😮
I loved that bequiet case, especially in horizontal mode, and I don't normally like glass or RGB.
enjoy the humid southeast asia weather you guys!
Loved your tongue-in-cheek talk about ai, especially "... become an ai person..."
4:20 - I'd say OcuLink is considerably better than using USB4 since it's PCIe x8. A much better use for eGPUs.
Liked the video and the camera angles. Good stuff gents.
I'm looking forward to a PC upgrade, but I can't do it until I see all your testing! Eagerly looking forward to the next few months of content. Thanks guys.
I was 100% against the whole rear connector motherboards but seeing as how more and more cases are being made to be compatible with both, it's becoming less concerning. I like the idea, I just want the compatibility to remain. Like how I know that if I buy an ATX case, my ATX board will fit. I was initially concerned about placement of the headers but case manufacturers seem to just be opening up the whole area around the edges which makes sense. I just hope that everyone gets on board so everything is compatible. Granted, small form factor cases won't likely have this option but at that point you already have to get an ITX or Micro ATX board anyway. But if all the new ATX cases have support for both configs, I think that would be ok.
you guys should do more content at random pools
An oculink EGPU isn't nearly as limiting as USB4 and Thunderbolt 4. A 4090 looses at most 25%, and it can be a much smaller loss in higher resolution/lower frame rate situations. Anything around the performance level of a 4070 TI or below should have little (single digit %) to no loss with an oculink connection vs direct PCIEx16 connection.
No A.I. power cables to power the A.I. PCs?
that cube on black tshirt really stands out , its giving a pop out look haha.
I really think its way to big its so out of place and wrong side 😉
The higher clock speeds will arrive when that new desktop/laptop memory form factor for ddr5 is released. It’s less about the memory controller and more about improving board/connector signal integrity. So the delays relate to the availability of the new CAMM2 (acronym escapes me) DDR5 module.
To me BTF only makes sense with light colour cases and/or white motherboards to hide any colour of cables better.
What I don't like the most about BTF form factor is the rear facing connectors will get covered with other cables, limiting clear access to specific cables, and them possibly getting mashed down and pins getting bent.
3:50 USB 4 (and Thunderbolt 4) eGPU's are heavely bandwidth limited, there are eGPU's that use OCULINK 4i and 8i connector that effectively doubles bandwidth compared to Thunderbolt solutions.
The difference to full fat PCIE 4.0 x16 slot isnt that big since even current gen cards do not use full bandwidh of x16 slot, Oculink 8i (it really is just PCIE through Oculink connector) would give PCIE 4.0 x8 bandwidth for eGPU's. Oculink 4x gives PCIE 4.0 x4 bandwidth for the EGPU to use and difference to GPU performance is only something like 5-15%, where Thunderbolt eGPU is heavely limited, sometimes getting only 30% of the framerate in some games.
Thunderbolt has latency issues yes but Oculink does not, Oculink does not need any controllers to work just straight connection from PCIE lines through Oculink to eGPU adapter with build in pcie re-drivers (signal boosters, no latency issues).
When you inevitably do AI Unboxed, you MUST have AI presenters. Call them Min and Max Headroom. I'll let you decide who Min and Max are.
Max is already taken. They can be Min and Mid Headroom.
21:20 yeah but some mobos with those VRMS could power a dual socket epycs no issues whatsoever.
The problem with eGPUs *IS* bandwidth. An external PCIe connector like OCuLink with more bandwidth is needed.
Pretty sure latency is the primary issue.
Derba8er and others also made a video from the GSkill Computex stand. And here they had Ryzen 8000 G series running with DDR5 Modules with 10600 MT speed. So I think Zen 5 will support adequate memory speeds
Those speeds were achieved because the ryzen 8000 are monolithic designs. it is essentially the same reason why intel's 12/13/14 cpu were generally able to cope better with faster ddr5 than the 7000 series. the 9000 is a chiplet design and while I am sure it will handle faster ddr5 better than the 7000 series, there is a limit on how much it will be able as the I/O is , according to AMD, essentially the same.
Those are monolithic APUs and they always supported higher memory speeds since they dont need infinity fabric to connect 2 seperate chiplets. Same way how you could easily run ryzen 5700G at DDR4 4400 while the chiplet ryzens like 5800x couldnt reliably go over 3600/3800 in 1:1 mode
I dont expect ryzen 9000 to go over DDR5 6800
@@ThunderingRoar sadly most people do not have that awareness.
I've always seen VRMs like high wattage speakers, you probably will never use the max capacity, but the smaller percentage is utilized the more efficient and cleaner the signal is.
1 minute ago? must have zen5 speed
I want to know if I can actually get 4 sticks with a total of 128 gb ram to work properly with zen5
I think Project G-Assist has potential to be a really cool feature. Imagine an MMORPG assistant that can role play in character. Imagine it act like a space ship AI assistant in a space sim game.
Your rundown on all the AI improvements and enhancements left out a new line of AI coming to Australia. Just when will the AI enhanced Drop Bears be arriving down here?
I guess the power phases will help if you have lots of USB-C devices connected and charging?
About here 9:23 - I really never heard of the inverted cases. I build the 1st computer for personal use (built many for business) and here I have my beautiful PC setup... and it faces a shelf and not me for me to see all of it's tempered glass glory. LOL
I really hope that BTF doesn’t become predominant. It’ll be really annoying if the overclocking boards start using this, it’ll basically ruin test benches.
i love your intro
When you mentioned AI memory you first showed a CAMM2 module, and I figured that OKAY, CAMM2 might be better for local LLMs than other RAM. But then the view widened to all kinds of other memory and storage and I realized it was just executive vaporware.
I think there might be some genuine use-cases for ML even if it's just used for the design (which executives seem to use as an excuse to slap on AI on their products these days). I think there were papers on using AI for CPU designs, and I can also imagine that it would be helpful with designing the traces of high performance data busses, like DRAM. Generally today's generative ML can be surprisingly effective when you need something technical that is somewhat creative.
But if I were to design such a thing with 'AI' I'd certainly not announce that to the world, but keep it a trade secret (if I were an executive, as an engineer I'd like to write a paper or at least blog post on it).
It's only ever worth investing in eGPU compatibility if you don't have the budget for a good midrange GPU right now, but want one in the future.
Buying a miniPC with USB4 support right now can be worthwhile, as it will allow you to use eGPUs in the future if/when you need a significant GPU upgrade, but if you need/want a powerful GPU right now, you should just get it now. An eGPU won't save a significant amount of money in the long run, because you'll need to buy an enclosure or adapter in addition to the graphics card, even if the card itself will be cheaper in the future.
8000 series with revised controller can already run 10k+ ram on existing boards so it’s cpu generation & not chipset gen that’s restricting speeds.
But, the 8000 series responds so well to high ram speeds because it’s benefitting the gpu cores even more so can’t imagine ‘the sweet spot’ for straight cpus jumping massively.
Total guess but i keep reading/seeing g series running 8400 so I’m just assuming 7200 (or thereabouts) will be the new 6000.
& that’s well within spec of existing decent B650 & X670 boards.
About the eGPU topic, I could not find any valid test video from Jarrod on the topic. There were 2, one with a 2core CPU in a laptop and one comparing laptop eGPU setup to desktop dGPU setup which is not comparable since there are some constraints outside the eGPU connection between the teste configurations (RAM, cooling, power budget for CPU etc ...). Also the laptop with the superior CPU from the 2 laptops used having worse results just confirms that there are more factors at play here than just the eGPU connection ... but I bet you'll do a proper apples to apples test just to prove the point ...
Zen 2 and zen 3 used the same IO die, and their memory was speed was basically the same with zen 3 only going slightly higher in 1-1-1 ratio.
I’m guessing zen 4 and 5 will be the same, roughly similar ram speeds with slightly faster infinity fabric since they use the same IO die.
Higher standard clocks for the IF are confirmed. So yes, I think the Zen 2 => Zen 3 example is spot on.
If external GPUs have too much latency when distanced from a given device, why hasn't anyone made a lego eGPU chassis and portable minui-PC case combo design? Couldn't you just have a setup where you pull out a port plug to expose the PCI-e connector and then snap the two together so they become one solid frame. You would have a detachable GPU for travel and a clean setup for when the two are put together. Seems like easy money if a case manufacturer can get a custom motherboard layout with say the PCIe connector on the reverse side so it can be flush with the case wall.
I love the idea of CAMM2 memory, but the only motherboards I've seen only have one module of space. That doesn't save space over DDR if I can have 4 slots of DDR, instead of one slot of CAMM2. You must have the ability to upgrade your memory instead of replacing what you already have.
more VRM - more chances of needing repair (in case one goes bad) or less so?
The flickering LEDs on Tim's mic are driving my inner Monk nuts... 😄
motherboards can easily support 4x 4high USB type A stacks, Mouse, keyboard, headset, camera, controller, audio deck etc is all type A.
So why do we wanna have half that in type C ports? for plugging in our 8 phones?
Nice romantic trip boys? ;-)
Yeah we loved it.
@@Hardwareunboxed Sorry I couldn't help myself. Thanks for the coverage, it's been awesome.
@@cairnex4473Wholesome bromance
i often wonder whos the giver and whos the taker 🤣🤣🤣🤣
@@dougkomonkeyboy8020Steve is the power tool guy that builds his own shed. Nuff said 😂
8000mt/s is already a thing on AM5, so, I'd say that will just have to do for now
My next pc will be all AI, even the fans. 🤣
I got rid of my a4 h2o sff case. Though temps were within tolerances the thing just blew out hot plastic smelling air and made my room stink.
Swapped to a mff with some t30 fans and it’s like a mini fridge
Question: if I build my new PC with an AI PSU, AI memory and an AI case, but then buy another bigger AI case and put my newly build AI PC inside this big AI case, how much AI performance will I gain? Can I expect 10% of more better?
And are all those AI components compatible with each other? Can the AI mouse communicate with the AI power cord?
No 1DPC motherboard shown at vendors for X870.
I am afraid for 8000+ stability
Putting AI in the model name just shows how desperate that company is
Bad thing is it will sell. And they are all going to do it.
So nearly every HUB video has the Thermal Grizzly Kyrosheet advertisement at the start ... yet I cannot find any testing done on it ?
Not on HUB... shit not anywhere tbh (only "lower" channels)
Plenty on the Carbonaut but I think that's an older "lesser" product
We tested it in our Asus Matrix review.
@@Hardwareunboxed Oh yeah there it is thanks.
I think I zoned out of that review as the card is beyond affordable, was just pretty to look at =)
Will you be doing further testing/reviewing on CPUs ?
Do you think Kyrosheet is worthwhile? I'm about to swap from a 3600 to a 5700x3d and I'm out of paste
*Just looked it up.. only $11 for paste vs $23 for sheet.. might as well go sheet right for zero maintenance which I'm quite fond of
I can't see why CAM2 is going to take off in PC's, as it erases modulation of RAM. It's a great option to pick up 4 dimms later on in life, as each spare dimm can run a PC.
For video editing on the go it will probably just make more sense to get Macbooks, lol.
As far as I know, AMD has made their RDNA2 (and above) dedicated and Integrated GPUs to work together so maybe they could manage the same for external. It might need to use OCuLink instead of Thunderbolt/USB PCIe for better latency. But it will still cost fair amount for the an enclosure or be really expensive with all in one external GPU solution (especially in the long run).
Framework does have the dedicated GPU on their 16 inch model being something you can detach, so that might also be a way to add more GPU power. But currently it is limited to RX 7700 and not something getting close to desktop dedicated GPUs.
i will say, that computex for me was underwhelming, all coverages i saw were mostly pc cases based on 4-5 older cases, like 80% of what i saw, another 15% would be cooling solutions
funny thing is that i was most excited when GN went to thermalright air coolers, i think i had too many cases at that point
tbh i started to forget about things like camm2 and any news, because there were hardly any monitors, all there was remaining was AI keynotes
i think TCL dropping 1000hz 4k display and all their technologies had more tech news than whole computex at this point
The worst thing about this AI marketing spam is that it is working on leadership of organizations like mine. They've moved from pointlessly demanding we put things in the cloud that don't belong in the cloud, so now insisting that we start using AI. For what? How? Doesn't matter, they just insist that we go AI.
I thought BTF was for Back To Front?
It is
What i want is Smart icloud gaming X turbo AI products.
You know what would be awesome? An actual _Back to the Future_ Edition of a BTF mobo, kinda like the Sonic and Evangelion-branded boards, released on the first film's 40th anniversary. I would buy that in a heartbeat.
Am5, zen 5, ddr5 , its the triple 5
What about Oculink for eGPUs? Greater bandwidth and lower latency that Thunderbolt 4
I don’t think it’s going to matter, I was able to achieve 7800MTS Y Cruncher VST stable on an MSI MAG B650 board, lots of others seeing success with really high speeds providing they have memory which can do so
"Except for a company like... "
My brain: Noctua?
"Noctua."
Hard to beat a company knows exactly who they are and sticks to the roots. When we get a noctua AI fan... The robots have taken over and they want humans to worship silicon.
Them not marketing their stuff as AI-whatever is a reason to buy Noctua in and off itself.
Too soon to make assumptions about the X870 and memory speeds - if the new memory speeds are supported on X870 on the Zen 5 already, but the boards are coming later for whatever production reason, they won't tell you either beforehand because people would waited to see/buy later and not get Zen 5 CPUs (and older chipset MBs) right away. And I'd expect AMD being fine with a gradual release and press with Intel coming later to party with Lunar Lake.
Lemme get myself a cup of coffee.
I hope you finished it before you got to the AI segment.
Why did you remove the "Ryzen 7 7800X3D vs. Core i9-14900K 253W Benchmarks" video ?
This video proves beyond any reasonable doubt that you are already AI Unboxed, powered by your AI overlord Balin. He appears human, but he’s not fooling anyone.
Oh, I thought BTF stood for back to front
You might say, it opened your A-Eyes
i have a lian li lancool 3 in white its awesome, but i really wanna try a be quiet one
Thunderbolt 5 is what I'm after. Might need to wait for Intel and cross my fingers hoping the platform isn't crap.
I loved the new AI computer mouse, that improved my CSGO skills with 10 000%, and I have never played the game!
Ai is the new Gaming, as everything was gaming at one stage!
Welcome to AI Unboxed. Today I'll be reviewing AI screw from Asus, it helps the SSD stay still in the case and let it loose when a screwdriver unscrews it. A bit expensive to my taste and I don't see much value unless you really want to be screwed by a company that was once top notch in clients service
evolv x2 is the best case by a long way .. i hope it doesn't only support back connect though.. cos there are literally 0 of them for sale in my country and i just bought a new motherboard
It was the same thing with AM4 the B350/X370 were horrible for memory speeds I still haven't purchased an AM5 motherboard because I knew it would be the same thing they had 3 chipsets for AM4 and AM5 will likely be the same so I'm waiting until they come out with the 3rd gen AM5 chipset then I'll consider upgrading from AM4
@unholydonuts and the AM5 high-end chips at that time will most likely still be very viable options for gaming like we see with AM4 now
the craziest thing is it isnt even AI, its pattern recognition. It has no intelligence. Its still just an ordinary computer. takes input, calculates output.
And it’s only going to get worse as it trains on its own bullshit.
AI DP Cable?
what motherboard should I get for a Ryzen 7 7700x?
A few years ago I was planning a trip to see these guys in Australia. I was applying for my visa and I was asked if I had a criminal record. I told the immigration officer I didn't realise you still needed one.
🙂
Just a question, is 12450h better than 5800u in terms of large number crunching??
Not seeing one x870 motherboard with 1DPC worries me... If you want to run 8000Mhz one AMD, you pretty need that.
The 386 was also originally marketed as being the processor that could handle the AI revolution...
Not gonna lie, an AI-based controller for SSD storage might actually be beneficial for more efficient storage of data on the drives.
I would much rather have overkill VRMs than a repeat of the MSI X570-A Pro.
what about AI gaming chairs for thous clutch 360 no scope kills?
Ebony, & Ivory living in perfect harmony. I just love ❤️ this Tech Channel.
After the AI discussion, I am now wondering if FTDI are doing AI UART ports...
thunderbolt 5 actually does solve those issues though, for like 95% of cases...
Amd can't even fix 4x dims at expo so why think about changing the speeds higher