Vayra86The PC wants to do it locally. Every other device that got shrunk over time, is aimed at doing things in the cloud. Even laptops; they're not equipped with sufficient storage to carry everything on local disk. To 'complete' a workflow on a laptop, you're generally looking at some external influence, whether peripherals, data, whatever.

TheinsanegamerNGPU sizes have increased over the last decade because we are no longer limited by production node. A GTX 780ti could only overclock so far. Newer nodes can clock higher, and handle higher power use, the major issue today is not clock speeds, it's thermal output, and thats why GPU coolers have gotten big, to handle the heat and also to run quieter then ever before. Even small cases today like the dan A4 can handle long triple slot GPUs. So...why not make em that way? The only market this doesnt really benefit is the LP community, but considering we got the 4060LP this year, even we're doing great!

Solaris17If you dont get it, then maybe we should ask you. How big is to big? Is dedicating a room of your house to a PC ala eniac totally fine as long as they are ICs instead of vac tubes? Anything for progress?

OnasiDoes it, though? How many tasks, apart from gaming (if we forget the Cloud Gaming attempts), are actually required to be local for CONSUMER PCs nowadays? Even many 3D artists I know prefer renting compute farms for their projects rather than building their own monster rigs. Storage, if you really want it local, is better done via NAS than cramming HDDs inside your case. You might say “video editing”, but Apple has shown that it’s more a function of proper acceleration and optimization rather than brute power with their M chips. Of course, nobody stops one from building a full-tower with insane specs, but that’s really not what the market should revolve around, methinks.

wolfSeems like Nvidia has a fairly different definiton of SFF than I do. I suppose this is better than nothing, but I was hoping for something more around LP cards, or "ITX" size/ length cards

Top card shown below fits NVidia's new SFF guidelines, with the 4060 LP under it. And an ITX sized card pictured for good measure.

Solaris17If you dont get it, then maybe we should ask you. How big is to big? Is dedicating a room of your house to a PC ala eniac totally fine as long as they are ICs instead of vac tubes? Anything for progress?

OnasiI feel like you are looking at it backwards. The size increase is caused by the increase in power pumped into the chip.

OnasiWhich is itself caused by the fact that new nodes are slower to come online and seem to have a less prominent potential performance effect. So, in order to compensate and win the numbers game, every modern GPU is pushed near its limit to eke out every drop of performance. Which causes more heat to be produced. Which causes coolers to explode in size. Thus a vicious cycle is born.
By the way, you do realize that the size of modern GPUs is technically a violation of the PCI-E AIB spec? It used to be that both NV and AMD at least tried to stick to it with their reference models. Now, that’s gone.

A Computer GuyI'm really liking the 4060LP. Half the battle is getting a decent PSU with good wattage to fit a sub 7L SFF case. HDPLEX 250w is a very convenient size but I wish they made a 500w one compatible with Flex PSU mounting for some of those smaller ITX cases that are designed around Flex ATX and need the PSU to be shorter.

I managed to squeeze a HDPLEX 250w into InWin Chopin Max as a replacement PSU.

TheinsanegamerNI'd like to point out, as a matter of course, that the 4090 not only significantly outperforms the 3090ti across the board, but also pulls significantly less power. Nvidia isnt just "shoving more power in and hope it works", there's a lot of optimization going on int he background. Just like with maxwell on 28nm.

TheinsanegamerNIt's not just power use. It's thermal density. Look at the 200+ watt power use of a FX 9590, then look at the temps people were hitting on 360mm AIO cooling, compared to a modern intel I7 hitting the same power workload. You'd think the i7s were just broken. Modern transistors dont leak nearly as much as older ones, which as a by product, means that getting heat out has gotten significantly harder.

OnasiThe 4090 is overall significantly more efficient than the 3090Ti, true. The latter is also based on a notoriously scuffed Samsung 8nm, so I am not sure it’s a stellar comparison.

OnasiThat’s true, however it’s not like modern CPUs are THAT much harder to cool. If we take your example - in identical workload the i7 will complete said workload far faster than the 9590 and will return to idle, making it more thermally efficient. And during said load, well, I just dug around some old posts and the 9590 was running at about 70-ish degrees on a full load with a NH-D15. So, a maximum heat load of 220 W (chips TDP). A 14700K with a same cooler under a same load runs at… 75-80 degrees. Is it hotter? Sure. But it’s a far cry from going in just three generations from a 1080 that can be easily cooled by a dual-fan dual-slot cooler to a 4080 which… cannot. And again, ultimately the modern i7 would be so much faster that it still would be at a thermal advantage.

OnasiAnd before you bring up modern Zen, while thermal density absolutely plays a role in those CPUs running quite hot (or, rather, being hard to cool effectively), the main issue currently is the IHS and the… debatable decision of designing it around cooler compatibility.

OnasiOverall, I don’t think we are in disagreement on a factual basis. This seems to me more like a difference in philosophy. You seem to be arguing that progress demands the current design direction. I am more inclined to believe that saner design decisions should be made, even at a cost of absolute performance if need be.

TheinsanegamerNFair enough, but TSMC 5nm isnt THAT big of a jump from samsung 8nm. It's not like the 32-22 jump, or the 90-65 jump of years gone by.

TheinsanegamerNAnd if you put a water cooler on those chips, the i7 will STILL run 75-80c under full load, whereas the FX could push down into the 50c range. IDK why you bring performance into this, thermal efficiency in workloads has nothing to do with my point.

TheinsanegamerNThe 1080 left quite a bit of OC performance on the table, which the 4080 did not. Fully OC a 1080 and most of the smaller coolers, while still working, would get rather loud. Which people greatly dislike. The only solution to this is either a bigger heatsink or a smaller chip, which misses the point completely.

TheinsanegamerNIt also plays into my point about nodes being a limiting factor. Pascal's 1080 was only 314mm2, compared to 379mm2 for the 4080. the 1080ti topped out at 471mm2 while the 4090 is a monsterous 609mm2. We never saw how demanding a 550+mm2 pascal chip would have been on cooling, something tells me it wouldnt have been dual slot.

TheinsanegamerNSo, like I keep saying, the 4070ti? Isnt that what a 4070ti is? Why the insistence that the 4090 should have been limited to the 4070ti design wise? Why not just buy a 4070ti and ignore the 4090?

And hell, look how PISSED people got that the 4070ti was going to be the 4080. Can you IMAGINE if nvidia made a 4090 that was 4070ti level. People would have blown gaskets and had heart attacks over it

TheinsanegamerNOr, for the sake of argument, they made the 4090 and under-clocked the snot out of it to make it a reliably 250w chip yield wise. You'd have forums full of buyers complaining that nvidia left so much performance on the table, looking for larger 3rd party heatsinks for their GPUs so they could have the performance the chip should have had originally, all while the 4090 was tying with a 7900xt in benchmarks. Is that preferable?

TheinsanegamerNwhat I dont understand is why people are so upset over the card being 3 or 4 slots.

Like....you guys have ATX cases. They have 7 slots. You dont use sound cards anymore, you dont use separate network cards, Controller cards? LMFAO no. 99% of "gaming PCs" are full of empty space.

Do people really want 72dba blowers back? Just to safeguard the empty space in their PC? The mini ITX guys I understand, because "mini ITX" used to mean sub 10L PCs, but have bloated to hold these big GPUs. But then, go buy that 2 slot jet engine 4080 and be happy? I seem to remember this exact same back and foth during the Thermi days, with people insisting the 480 was too much and 300w GPUs were way too power hungry and dual slot coolers were obnoxiously big and everything used to be single slot and blah blah blah.

TheinsanegamerNOh boy do I have GOOD NEWS for you.

They DO have a 500w supply. Listed right under that 250w supply you linked to. Same form factor, just thicker.

hdplex.com/hdplex-500w-gan-aio-atx-power-supply.html

You could probably 3d print a flex ATX to HDPLEX mount.

I too love my 4060LP, and I hope it sells well enough that we get better chips. a LP 4060ti 16GB int he form of the 5060 would be sweet. Or better yet, 1-2 generations later, the RTX 6/7000 series LP card will be powerful enough that I can retire my big desktop and get by solely on a LP setup.

kapone32Well you could also install storage in the x16 slots

OnasiI have no idea why you keep bringing up the 4090. I haven’t mentioned it once, I don’t care what the halo cards do. They can be as stupid as they want with them. But yes, the 4070Ti is still burdened by a ridiculous cooler for the vast majority of the models and that’s where I take issue. Even the “SFF” models in NVidias list are nothing like what actual SFF cards should be.

OnasiThat checks out, yeah. That’s what I would have preferred, caveat being replacing the 4090 in the analogy with 4080 and below. As mentioned above, the halo card can be as ridiculous as they want. Complaints of the buyers in this case means complaints by enthusiasts, which is a tiny minority. The vast majority would not have cared.

But I feel like the thread has been derailed enough. I’ve made my point known, you made yours. We are at an impasse since, as I keep saying, it’s a difference in philosophy, which is fundamentally unresolvable.

TheinsanegamerNYeah, you CAN, just like you can install sound cards or network cards. But almost nobody actually does anymore. And since the vast majority of builds are still full ATX, that would mean you could use 3 slots for storage, fitting in something like 12 M.2 drives, and still hold a 4 slot 4090.

Mainly because, here and elsewhere, when this subject comes up people bring up the 4080 and 4090. You're the first one to even type "4070" in response.

You're right in that the 4070ti could be pretty easy to make a true 2 slot card with, given the temps most run at now. But since noise rains supreme, and short of some older ITX cases most can take a 2.5 slot card without issue, thats what the whole industry defaults to, especially as sales decline slowly. Even going back to pascal, cards maintaining 65-70c and 30~ dba at full load was just unheard of. They were either pushing over 40 dba or over 80c at full load.

I would like to see more ITX sized cards, for computers like the deskmeet X600 that would be a really neat build if there were GPUs that fit. A 4070ti would fit, if theyd build it.


Fair enough.

wolfSeems like Nvidia has a fairly different definiton of SFF than I do. I suppose this is better than nothing, but I was hoping for something more around LP cards, or "ITX" size/ length cards

Top card shown below fits NVidia's new SFF guidelines, with the 4060 LP under it. And an ITX sized card pictured for good measure.

OnasiDoes it, though? How many tasks, apart from gaming (if we forget the Cloud Gaming attempts), are actually required to be local for CONSUMER PCs nowadays? Even many 3D artists I know prefer renting compute farms for their projects rather than building their own monster rigs. Storage, if you really want it local, is better done via NAS than cramming HDDs inside your case. You might say “video editing”, but Apple has shown that it’s more a function of proper acceleration and optimization rather than brute power with their M chips. Of course, nobody stops one from building a full-tower with insane specs, but that’s really not what the market should revolve around, methinks.

Keullo-eMy thoughts exactly. For me, a SFF machine means a low-profile one, not one where you can stuff a normal-sized card.

Gainward RTX 4080 Phoenix GS PCB

A Computer GuyI'm really liking the 4060LP. Half the battle is getting a decent PSU with good wattage to fit a sub 7L SFF case. HDPLEX 250w is a very convenient size but I wish they made a 500w one compatible with Flex PSU mounting for some of those smaller ITX cases that are designed around Flex ATX and need the PSU to be shorter.

I managed to squeeze a HDPLEX 250w into InWin Chopin Max as a replacement PSU.

SestoPTJust imagine, having a computer that is BARELY bigger than a regular ATX PSU with a x3d chip and a 4080 in it.... That is the ULTIMATE DREAM MACHINE !!!!

SestoPTYou do understand, 7 liter cases, are "massive" for today's real SFF cases...