#šātech
1 messages Ā· Page 1038 of 1
Aida64 is not a stability test or stress test
Tomorrow when I wake up from coma.
In winter I could definitely hit 60°C
What wattage
144W or whatever the limit was
Weren't you unable to prove that last time I asked for proof?
No, I got you your proof
Whatās the technical reason for this?
I remember hearing something about infinity fabric or something.
:(
welp, that's a major bummer
so clockspeeds will continue to drop as time goes on?
Only n7p can hit 5g practically, and it really can't sustain it under load
š¤·āāļø tell that to intel
has always been the case for new nodes, but improvements and all can always go the other way
intel
14+++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
let's hit 10GHz
no
what node jump that didn't fundamentally change something else did not come with a clock regression at first?
Key word being at first
First implementations of a node have always been higher clock than the first implementations of the previous node
true that
I mean yeah, thats what I said, at first and the improvements come š
Point is clock goes up over time
Donāt need multiprocessing if a single core can compute everything. 
- Intel, probably
That's implying single cores don't do multiprocessing
Ooe is magic
yea, point was N7 couldn't go higher but n7p can go higher as its no longer the first iteration
also don't you mean OoO?
so pumped for that 4 core running at 735 GHz in 75 years
rolls eyes You donāt need parallel multicore processing.
Happy @astral haven?
Out of Order Execution
Big difference between those statements
Practical limits are gonna be in the 300ghz range
According to what?
Thatās an arbitrary as fuck number.
Did some scientist do the theoretical math?
Am I missing something?
Em waves have to be larger than the band gap of the material
Right, I getchu.
Silicon can do around 40ghz in practice, 50ghz in theory
Makes sense.
Carbon can do 700ghz in theory 600 in practice
But CPUs use heavy clock multiplication so the 300ghz range is where we will likely max out
š¤·āāļø I doubt it will matter tbh
Someone will make a quantum computer room temperature and this won't matter any more
Not for ages.
Less than 10 years out by my prediction
Room temp quantumās arenāt really possible with tech for a LONG time.
Not commercially.
Thermodynamics is a bitch.
Cause who says the internal temperature of silicon has to be higher than the external :P
time to ditch electronics for photonics!
Basic thermodynamics does.
Nah aeon
Look at the peltier
Unless you contain shit in a perfect vacuum and somehow deal with the various radiation.
Literally Intel sells a module that does that for their CPUs...
No no.
That doesnāt do that.
Itās a nifty tech, but it doesnāt stop thermodynamics.
It exploits it, yes, but it canāt do nearly enough for quantums.
Says general thermodynamics.
You gotta remove ALL thermodynamic interference because it fucks with your read.
Not all
Just enough
Means you canāt have a medium to transfer molecular vibration or you have to cool it down enough.
Canāt be done in room temp.
Still needs impressive cooling.
Basic laws of reality, my guy.
Impossible? No
Tbh tho, itās probably doable under LN2.
So Iād probably just run a few dewers and call it a day.
If random folks on the internet can do cascade with home equipment, Intel can do it inside of a chip in 10 years
Talking about Qbits still, no?
Idk man, Iām working at something like 2% brain.
Sleep dood
Water
Yes but drink it
Moar
Then put down the phone and put in earplugs or something
Shhhhhhhhhh. The vertigo isnāt going away.
Iām so lucky spellcheck exists.
Omg itās working so much overtime.
Ok, Iām gonna go actually pass out and wake up in 11 hours.
Ded.
Unique to steam maybe
So, Steam Controller 2.0
$400 for a controller is rather steep :P
That's just the price to pay to finally have that screen on the controller
Did you copy paste some weird stuff on accident?
yea that was wilfully ignored and misinterpreted when I posted it in the discussion hours ago š
"Disclaimer: The following article is machine translated from the original German, and has not been edited or checked for errors. Thank you for understanding!"
yeah the important parts are accurate though, at most the grammar making it confusing
Thing is tho
This isn't the cause of the problem
This is an effect
The fan control might have a bug where it goes haywire when the power of the card goes over it's power limit?
But that's not the problem
The problem is that it's going over it's power limit
I've never seen an Nvidia card ever self report going over it's own power limit
THAT is the bug here
GPU performance gains have become absolutely insane over the last few years... If VideoCardz is right, we're looking at 2-3 times the performance gains between lovelace and RDNA 3
RDNA 2 in itself is already pretty solid, really not far behind in raster at 4K and technically ahead in perf at 1080p
RDNA 3 though, AMD is really shooting for that performance crown.
I expect it to win, but nvidia will have >>> volume
More for money
Compared to the ballpark of ~20 to ~50% increase that used to be the norm each generation, things have really become insane
high end went higher up in price yes, but with next gen the middle class will also have a big jump
What I mean is they made higher end a lot cheaper
wut
2080ti was a lot msrp
look, one generation of extreme price hike doesnt mean they now are cheaper xD
like compare it to the 1080ti
Eh
Idk how to say this right
They put all their cards on the playing field and reduced the price
I don't think it would have leaped so much
If it wasn't for them needing to compete
I don't either
still, when you account for things like inflation, even production cost, and compare it to where prices are now compared to say... 10 years ago (and the comparison only gets worse as you go back into the past), nothing really justifies the current prices.
I feel like they give more for less this gen because they knew they couldn't sell for much more
Is what I mean
Like the prices were insanely reasonable
ok I dont care anymore
Compared to previous gens
intel just announced they finalyl renamed their node names
finally no more nm bullshit
Quite excited for next gen Intel
They are about as creative as a chicken without a head
it's not about creative names, it's about not using false naming anymore
this is great
10nm has nothing to do with 10 nanometer
the livestream is still going but press was briefed ahead of time
I'm rather skeptical to be frank. Intel basically had to hurt their own multi-thread performance in order to get back the single-thread crown, and they still lost in actual gaming with the 11th gen. who knows how they are getting their numbers with the 12th gen leaks?
because 11th gen came out even better than people expected
Oh I'm excited but I ain't going Intel again for a longggggg time
11th gen was never planned for 14nm, that was the biggest issue
but 13th gen is the big boy that gives them a chance to be #1 again
12th can be on par
It's kinda funny how it's to the point where we are talking about future gens where intel can 'be on par' with AMD š
I mean, given how many years of stagnation they had, the fact that a lost lead can be balanced out after just a year is impressive
we'll see. they have such an enormous R&D budget that they just try to bruteforce it with funding research teams in different companies. the fact that AMD can maintain a lead for so long with a comparatively tiny budget, is so much more impressive to me.
you keep forgetting, amd is not doing this alone, tsmc has had insane funding from apple, qualcomm, even intel aswell to keep the process going
Well nothing in tech is ever truly 'alone', i think the point still stands that a very limited R&D is beating an insanely big R&D into the ground.
the fact that they have to play catch up, and can maybe do it in a year, is not really impressive, it's more humiliating.
oh yeah, just keep in mind part of it is that the process had issue, which is 95% what tsmc did without amd
and intel's spending is not that much more gigantic than tsmc's
@ashen spindle heu
https://www.statista.com/statistics/263562/intel-expenditure-on-research-and-development-since-2004/
see for yourself
There is a massive difference
yes now include TSMC
@ashen spindle Are you good with charts I need help lol
uhh depends
Science stuff
I need to combine two different charts but there's just too much data
I'm just contemplating on not doing it thought to ask here one more time
so you need a dense display for lots of datapoints?
yeah sec
those two combined
Doubt I can but meh two pictures is fine too
I need 3D chart
They're the same specimen but two different properties of pH of the same specimen
hmm yeah can't directly think of something that would fit it well in a 2d way aside from interactable things you can't put in a paper or so
@ashen spindle tsmc spent 3.72 billion on R&D in 2020 according to an arstechnica article.
Oh well worth a try thanks anyway
add that to AMD's R&D, not even accounting for the fact that they divide their R&D between two major tech segments, it's not even close to what Intel spends in total with R&D by themselves
@ashen spindle here's the article https://arstechnica.com/gadgets/2021/04/chip-maker-tsmc-to-invest-100-billion-in-research-expansion/
TSMC is skynet
now add the billions apple spends on a node development that amd benefits a year later etc.
yes intel still spends more, has more capacity etc.
but it's not that massive as you might think if you just look at amd vs intel
the statista charts shows 13.56 billion for Intel and 1.986 billion for AMD
(in 2020)
yea and amd also doesn't have their own fabs, does not have to spend on other ventures like mobileye(ai for self driving), network equipment, optane
oh i get that Intel spends on other tech segments, but you seem to be forgetting that AMD spend's on GPU R&D, in rigorous competition with Nvidia no less.
I use an Intel network card, and i do think Intel is far better than the realtek on my motherboard, so i certainly don't forget that intel makes other stuff
which is also majorly supported by tsmc's spending etc. you make it look like I'm saying that makes them equal, I merely say it's not apples to apples xD
yes, much smaller teams and all and it's extremely impressive and as a customer I'm very glad we have competition
but david has a lot of ladders from other people to beat goliath in this case š
At this point, i just don't want david to be the new goliath
thankfully looks like a good fight for at least 5years from what we know so far
gotta pray it stays like that for longer
it's interesting how AMD and intel have so thoroughly reversed roles, I now have intel in my mind as the brand with frequent sales on their chips that young gamers on a budget would go for.
like you, I do hope they keep AMD on some sort of edge
they will, one thing intel has going, they are not as limited as tsmc in terms of how to develop nodes, so (if it works) they can take bigger steps at once
And me with the amd chip sees Intel as the platform of old stability still...
Amd really needs to figure out their workflow for agesa
And you can get faster Intel chips than AMD
It's just horriyingly expensive
In my system i have a 3900x. frankly, the only time i've had compatibility issues when dealing with AMD processors is with outdated motherboard firmware. I'm letting my mother use a 5600x i had in my old motherboard for 6 months (which i put on hold building up due to absurd prices or low stocks on parts)
As it always has been tbh
Just the difference has never been so small
yeah amd's software in general is not up to competition yet
both gpu drivers and cpu bios etc.
but soooo much better than before
unless you're on Linux
AMD GPU drivers on linux are vastly superior to the Nvidia drivers for linux
USB drives are broken, overclocking is busted, pbo is busted, ECC is broken, raid is unuseable for ssds
granted, I have a GTX 970 in my system
Not on supported distros
too annoyed about linux from work lately so I won't comment on it 
Which a problem to be fair, but on supported distros (which are relatively limited) you get a very buttery experience ignoring inherent Linux issues
what are you calling 'supported' distros when it comes to Nvidia drivers on Linux?
I've never had a problem free Linux experience, but tbf the only times I've tried Linux were to try things that windows didn't support
The ones listed on the supported list
In the past it was limited to ltsb releases only, which really was the thing that annoyed the Linux crowd
They never had a single release for current kernel
And apparently they changed that, but I can't find the supported os list any more
if by supported, you mean you can technically 'install' proprietary drivers on a distro, there is more to it than that. i remember looking up a benchmark comparison between the 6900xt and the 3090 on linux, and the 6900xt performed considerably better than in similar comparisons with most of the same games and the same cards on windows.
No, you can install the drivers on anything
They only are Nvidia supported on a subset of distros
intel's next big show to talk about things will be in october as Pat just mentioned in the live Q&A, wonder if they launch alder lake there too
if that's so about nvidia supporting specific distro's, then I've learned something new
But i'm going to do more research into it
It definitely was true when I last tried switching to Linux at the beginning of 2020
this guy makes solid linux benchmarks https://www.youtube.com/watch?v=ykiU49gTNak
Helpful Links:
- @Level1Techs Main Channel Video: https://youtu.be/yMz-5uP8sBw
- https://forum.level1techs.com/t/fedora-33-ultimiate-vfio-guie-for-2020-2021-wip/163814
- https://forum.level1techs.com/t/vfio-in-2019-pop-os-how-to-general-guide-though-draft/142287
About Stocks & Availability: Please be patient and wait about 1-2 weeks before g...
Yea
L1t is solid
Just simply can't yet
You can't, they ain't making enough
today is my day off and i've had a couple of beers mid day, don't judge
Nothing wrong with that
Some people get really judgemental about such things haha
Processor availability is becoming vastly better lately, it's the gpu's that are still a big problem
regional thing, as you might remember, has been in stock at/below msrp here for quite a while
it's starting to stabilise at basically intel pricing levels for CPUs
here, at least
i saw the 5600x for 280 usd not too long ago
of course, mobos are a fuckton less expensive, still making ryzen the value option
price for performance, i still consider it the value option, especially when looking at the default 'retail' price, ignoring sales prices.
retail prices are about the same here for CPUs at a given tier, like R7 and i7, sale or not
I just also look at motherboards when comparing CPU value
since most likely, you're going to need a new socket, with intel at least
so an 11600K would cost an identical amount to the 5600X where you live?
better CPU for less than the worse
at least at the high end, 7s and 9s, where I look, pricing is pretty similar
500 CAD for 7s, 700 for 9s
maybe 10-20 dollars difference between two given CPUs
knock of 140 from the 9 and you get roughly the price of the 5900x in usd retail
Z590 to X570
of course, B550 is more than enough
PCI-E 4.0
or is that CPU dependent?
I coulda sworn that was a Z590 thing
I'm very happy with my x570 tomahawk from MSI, particularly the cooling on it.
200 for premium cooling? absolutely!
oh yea, they're terrible in that regard
i just went with the best cooling for the buck, that also had other nice features on it
i regularly use the bluetooth on that motherboard, actually it has become what i use by default for my headphones
lets hope this will be mostly on track
10th gen is faster outside of avx512 and that doesn't have gen 4
And some z490 have gen 4, the only difference is the significantly higher memory overclocking limits and the extra chipset bandwidth
Also what cooling?
I didn't think msi made any competent CPU coolers
Oh yea but buying a motherboard with components that would care about having a heatsink on am4 is very dumb
Unless you are in the sub $140 range
Then you have to be careful
Is there any normal way to get ācustom windows taskbarā on top, even if just to show basic stuff like time⦠?
Time and basic controls being shown in bottom right just look⦠not the best.
Honestly, idk why you wouldn't heatsink em anyway.
Thermal epoxy and a little aluminium finstack would cost you 5 cents per chip.
you would
but for aesthetic reasons
I mean for the hyper cheap boards.
Those sometimes just don't have cooling.
Why even?
shrugs Sometimes taking the hit can be worthwhile.
Having the single best budget items is a pretty good position to have.
Look at AMD of 6 years ago.
Uncontested by everyone and completely dominating their entire space.
š¬
no
amd was not in a good spot in 2015
amd was not in a good spot in 2016
or 2017
or 2018 even
That's true, but they had the entire low end.
How so?
Intel's offerings were often eh at the low end (AMD had some banger value chips), and nvidia is nvidia.
They've always been more interested in good GPUs starting in the several hundred range.
amd had alright value chips
but they werent purchased en masse
I feel like that's not a great way of looking at it.
It's the WW2 plane thing all over again.
?
Intel has the most chip sales because Intel had the mid to high end, but it's a PC, most people are in the mid to high end and thus have little use of hyper budget stuff.
Low meaning what in the context of 2016 to 2018? 4 cores hyperthread being the mid.
im talking i3s and below and a series (because ryzen wasnt there at the time)
amd didnt have anything for high end till ryzen
and even that took a while for the 1200 to exist
but the 1200 was an utterly garbage chip for low end since no igp
My thoughts were more on the FX 8350 during 2016, acually.
Those were some REALLY cheap chips.
I had one, my man.
but volume was very very very low
Was not great, but y'know what? It was pretty good for the price. $30.
bulldozer sold less than 100k units since the launch of skylake iirc
Wasn't really something you could compete against at that price.
Logical.
Skylake was a good generation.
Not as good as the older ones, but still good.
Beautiful Core i7 2600, we miss yee.
skylake was a fantastic generation lol
broadwell would have been better but it didnt exist and was very expensive
I know. My 6700k still runs my media PC. 
Got me confused there.
Hype for a downgrade.
Pog?
More meant the GPU, but eh.
What specs on the display?
raised eyebrow
Puke bar.
It really is rainbow puke.
I have 2 of them.
The range is a little too low on the FTW3s.
It looks somewhat washed out.
Puke bar.
There's just way better offerings for RGB.
MSI has a couple nice ones.
Want me to send you a pic?
I have 2.
the ftw3 looks like garbage
i bought one and im selling it right now
Hereās 2, hashing away.
Iād have to say the MSI Suprim line looks quite nice.
But it aināt much compared to the XC3s. Such a smooth little card.
Look at em go!! :D
it looks worse in person
It does.
I have em in person.
Looks fugly.
It's a lot brighter in person to the point where it looks obnoxious.
Nah.
i can assure you that it does not
They look really nice on video since the puke doesn't wash out anything else, but in person? Eeeh.
if you dont believe that i have a card then you can shoo
not any more
im selling it right now lmao

It looks great if you're still in the childish EVERYTHING RGB phase.
I've more moved on from that these days.
Rainbow puke just gets boring fast.

:O
Oh no!
Mods!
Mods help!
Aboose!
D:
ABOOSE SCREENCAP.
MODS HELP.

Bruh, grow up.
But, to keep this channel actually on topic, next gen AMD versus Nvidia.
Anyone got ideas on which way that'll go?
both ways
amd is gonna win again on raster and nvidia isnt gonna drop the ball on rt
Fml
How we've fallen. ._.

Bruh, this guy.
Lmfao
I wanna say this dude is a troll, but he's going at it man.
<@&490097707487330305> Aight, can we get a timeout here or something? As amusing as the last day has been, we really should keep this on topic.

Actual child, I swear.

Oh very much so.
Reputation as an immature liar are super good.
Watch, he'll be back in 5 minutes because he doesn't know how to drop things.
I'd happily bet you he'll steam away, only to log back into discord and check the messages.
i thought these prices are very very good tbh
i would have expected an extra $10 on all of them
O_O
wouldnt have been happy about it but that is what i expected
Not bad!
4
POG
Which laptop is this that just won the laptop olympics?
its missing touchscreen and amd (but its usb 4 so amd isnt really an option anyways)
thats it
its ultra pog
It's upgradable, by default that earns it pog satus.
and if you want to assemble it yourself you can get it for 750 instead of 1k
Very pog!
Can't just put one on?
I see nothing stopping you.
Maybe a little extra jank, but hey.
Daughterboards for that, no?
this is a very very thin laptop
Iirc the little uber tiny ones exist.
its within a mm of the mbp
Hmm.
it wont fit on a daughterboard
Ok that's kinda jank.
it would have to be integrated on the main motherboard
or attach to one of the usb ports
which is just no
not elegantly
Still technically doable. 

My hopeful assumption?
AMD pulls a 'Threadripper x64' on the GPU market
I just wanna see chiplet GPUs.
We get them, next gen
MORE SPEED! MORE FRAMERATE! MORE SPACE HEATING!
The only question is: how far will they go
gen after next :/
Speaking of chiplet designs, when's anyone else picking that up?
RDNA3 will be chiplet based
At least Navi 31
We know Intel in theory has it happening in a real sense, but when're we getting chiplet things that aren't AMD on market?
š¤·āāļø ibm, intel already has it in a product that they just discontinued.
thing is tho
Sapphire Rapids is great and all, but we can't be letting AMD have no competition.
Chiplet is scalable.
practically speaking it cant scale beyond what monoliths allow
Unless we can get something else cheap that scales well it's kinda the best tech for end users atm.
Binning is.
you literally can transplant a chiplet design into a monolithic design and make it 10% faster for 10x the cost
but 10x the cost is only if you are over the inflection point
the smaller it is the more better a monolith is
and the less cheaper the chiplet is
While that's all true, a solution needs be made for higher cores for lower price.
I remember the Skylake days.
$1600 USD for a 12 core is...
No.
š¤·āāļø chiplet is good for cpu
its less good for gpu
itll still be good sure
Less of a generational leap.
It does
You bin individual parts of a chip and mix and match the best ones
That's technically possible, but not at all practical
when you dont have an easily divisible architecture already, you cant freely divide
Yes, if Intel pulls the stops and tries to make an incredible chip they can do it.
But usually at obscene thermal problems, power requirements, and cost.
Wasn't the theoretical cost for the Everest Skylake series something like 5 grand each?
And they all ran like an oven?
Something stupid like 400 watts off a CPU or something.
Don't overestimate them
Sure they could make a larger chip, like 1 kW large
But they couldn't make a faster chip, or they would already do it
there are only 3 types of chips that will actually meaningfully benefit from chiplet and that is cheap manycore cpus, gpus, and ai chips
cerebras proved that the third gets monched by monoliths
what
thats uh
throttling disabled epyc
reminder
Back in the Skylake days with 8 cores, in context.
Epyc can have 64 cores.
skylake was 22 cores for the crazy insano chips
Skylake Blackops and Everest could pull 400 watts off a CPU if you could even cool the thing.
intel can't just "pull all stops" and make a CPU that has 50%, or 100% faster cores
They could make one that's 50% or 100% larger though, but at what cost?
amd doesnt have 50% or 100% faster cores
I know. Neither can AMD do it
heres the point
monolith will always do a bit better
equalize for everything else and you get 1 thing 10% better
Monolith isn't practical for cost.
If the monolithic CPUs have to cost 30% more to compete with an equivalent chiplet CPU it makes no sense to buy them.
Even if you were bulking them.
hardware is cheap in alot of servers...
You wanna tell people they're gonna spend 20% more per unit for the same power? 
Nobody is doing that unless they NEED monolithic.
our it department pays alot less for hardware and then related support contracts than we do for the software
and this is very very very common
look at the 8c 256MB chips
thats because you pay per core for some of this software
And if you're setting up a 30 million dollar supercomputer it's a no brainer to save 20% by means of not going with a difficult technology.
and you need maximum performance per core
If you're setting up an office? Whatever is cheaper on the licence cost.
ah but depending on your use it might be cheaper and faster to go with a pair of 8c cpus than a pair of 64c cpus
Kinda.
That's also dependent on the setup you need.
transaction latency not bandwidth
theres no doubt that splitting up your load is the better thing to do if splitting up your load is possible
Well the chiplet design will still translate over.
but when your software costs 6k per core
uh
you watch very very carefully how many cores you put in your server

and in terms of for example maximum throughput on 8 cores
the winner right now is icelake
cause sure rome is fast
and those bigboi cache chips are cool and all
but they arent super duper speedy if we are being honest
ice lake isnt signficantly lower latency than skylake but its still better
and rome and milan are higher latency
tldr free 10% boost to whatever stat you want for a very high price
most people myself included wont pay that very high price
but its still a 10% guaranteed boost
š¤·āāļø mature in what way?
We'll see.
thing is
whatever improvements you make to a chiplet design could just be made into a monolithic design...
cerebras again is proof of that
in what way
It happened before therefore it will continue to happen.
no
as in
its copper moving through substrate
instead of silicon
you can always just
not exit the silicon
Mmmm, this is true, but a maturing of the tech could yield another interesting innovation.
I don't see it stopping dead.
think from a theoretical point
chiplet by definition is not what you are interested in i think
In theory, yes they're the same in functionality and either can do it.
i think what you really are interested in is 3d stacking
because 3d stacking can do something new
Both.
Newer 3D stacking is one of those things I haven't looked super deep into, haven't had much time.
by definition anything chiplet can do a monolith could do better if you have the cash
3d stacking is different
Dude, chiplets allow for more consistent yields and higher quality cores for cheaper.
That's kinda the good thing behind it.
Nobody would be sane paying that much for 10% improvement.
RTX 3090 buyers everywhere

to be fair, if you need the extra VRAM, then get it
It's not more performance in that you can pack more performance into a monolith, but that comes with a hell of a lot of cost as you said.
Efficiency is extremely interesting with chiplets.
3D stacking only really adds to that.
chiplet is not more efficient
its more silicon efficient maybe
not more power efficient
Ah, but on the yields it very much is.
š¤·āāļø thats investment
Means you can lower the gap in core count across the board.
Gives more incentive to do things like multicore processing.
if you need the performance you can always go monolith to replace a chiplet design
itll be alot more expensive but its always possible
again aeon
cerebras proves that wrong
I'd value just that incentive alone as more valuable than anything a monolithic design can possibly give you.
Unless you're hung by latency or licence fees.
its incredible pog
Ikr.
Tbh, I want chips to be bigger like that normally.
Not that size.
But like... 2x bigger than normal, maybe 3x bigger.
wait .04sqm
It's a weirdly good feeling, holding any of the TR chips.
since you bought 3900x š
anything that fills out the hand feels nice
3950x?
Got a good deal on it when it was new.
still not tr
The 16 cores have suited me.
I did, however, buy and assemble a TR system recently.
It was awesome.
Wish I owned it. :(
its pretty nice
It's so cool.
but you didnt for yourself
and the point is that most people wont
Ah but no.
.01% will
ā
Wanted to wait a gen or two.
So I went with someone cheaper.
And no, I'm not main running the TR. I'm not that foolish.
you going 5000 tr or are you waiting for ddr5?
yea i dont believe you
Oh?
24 core lga cpu is msdt not hedt
TR board + ITX Ryzen
and you could fit that sucker in just one case
This.

ill bet you wont get the tr
specially since there isnt a point
ill bet uh
i dunno
hunh
I bet you I will.
At a minimum the dedi server is upgrading to TR for more RAM soon anyway.
Want 256GB on it.
And 32 cores.
how about i donate $50 to charity if you buy a tr ddr5 cpu
and dont return it in 2 months
shrugs When they eventually come out, sure.
By that time I'll be able to afford it easily.
i really dont think you will
are they expensive, yes
are they pointless, yes
do they perform worse for the money than a standard ATX build with the same money, yes
are they cool, yes
id bet you will just buy 24c msdt ryzen
that's all you need to make a decision
For my personal machine? Probably.
Whatever the best Ryzen at the time is.
But I defo wanna do dual PC shenanigans.
128b ddr5 can do 512GB dimms unregistered eventually
I'm nut ready for that DDR5 practical speed, tbh.
tbh its better to do dual pc nonsense via vms than physically
Yeah, I know, I know.
especially with 24c msdt ryzen
My main plan was to run a TR and just have everything on proxmox.
š¤·āāļø i have 2 pcs and no case
Literally all the things.
very cool
if 2pc 1 case is better than 2pc 2c then 2pc 0c is best
currently im only using one
I have technically 6 PCs in 4 cases. 
but thats because im waiting on a nic
shush you miner
Dedi server is pulling some Hyper-V time.
would you be willing to share with a fellow comrade
a mining rig doesnt count as a personal computer
I'm not counting them.
Then I have personal PC, VR/Media PC, and my roommate has a PC.
im talking physical computers
oh i forgot that this hdd is not on this main computer
shoot
3 pcs 1 case
eh that one doesnt count since i didnt build it š
i dont
my system is far too abnormally shaped and my roof isnt that tall
i really cant decide between this and a more reasonable setup
Do it, man.
...
heres the problem
its around $700 and i will never be able to afford a switch for it
It's not a bad price.
idk if I should get those and ram them into my PC
just to connect my other PC to that instantly
this is the cheapest switch
Probably something like 7 grand or so.
For the super amazing ones.
1.3 grand ain't bad.
this does up to 16 ports of 100gbe
Super doable.
but
You'd have an AMAZING LAN.
thats vs $uh 300 max?
Ok, I can see how that's a problem.
and i already have amazing lan
Save for it slowly?
Think about the fact that you could transfer movies in 2 seconds.
like if i paid microsoft the cash for rdma it would already be faster than my drives
100g wouldnt be fasterer
but it would allow me to use single mode fiber
because 100g smf tranceivers are pennies
thanks intel
silicon photonics for the win
I say go for the fiber.
oh its fiber vs fiber
RDMA would be hella fast, but 100GbE is coolio.
but its single mode with own connectors vs multimode direct attach cable
Iirc multimode would be slightly better.
Or I'm braindead.
$300 cable for the same performance as $20 cable
smf is more expensive for the tranceivers if intel didnt do magic
Idk man, I can barely remember fiber stuff.
Last time I did anything with fiber was ages ago. xD
single mode is expensive on the adapters and cheap asf on the cable
but
intel is siphoning off 100g adapters for pennies
I mean...
so cheap that i bought a pair for $60 already
my server has 7GB/s of drives
youve seen it right?
pile of iodrives with a nic at the end
I have not, actually.
its somewhere in here
its just over 1 gen 4 nvme with rdma disabled
could just about double that by forking microsoft over the $200
one off
Worthwhile.
Hmm.
i have yet to actually give them any money
Go sailing?
nah
Not a chance someone hasnāt plundered it.
its possible to do but annoying
Whatāre the required specs for it?
Well then.
windows 10 pro for workstations
Guess I have some plundering to do.
Ah.
infiniband
:(
and 40gbe+
š¤·āāļø you can just install the "free trial"
and it works
but thats not very cash money of me
I could do that.
Or...
I could plunder.
YAR HAR!!!
Itās Microsoft, for every core they charge WAY too much for Iām gonna plunder something.
its not microsoft that has the obscene per core fees
like lets be honest, nvdimm support is absolutely worth $200 for a workstation
$310 or whatever for a standalone key
throw on pnp rdma and its a steal
Their other stuff, tho.
It makes doing shit legitimate so hard. Yāknow how much money Iām gonna lose when I incorporate?
Rn probably like 10% of my annual.
There is zero possible reason for me to do that when I can sail.
Oh shit.
My big crypto miner just tripped a fuse. xD
oh my god sun i hate you with a fury that cannot be expressed
Lmfao wot?
Ah yes,
Those are called...
"Fuck you, dickhead, life is hard."
- Oracle, probably
this is entirely bs fuck off oracle
The fuck even are those cables for?
infiniband

instead of using the standard qsfp28 that literally everyone else on the planet uses they use this
What in the fuck, why?
What possible reason could there be for this garbage tier connector?
QSFP+ exists.
this is what anticonsumer looks like
this evokes joy
im not happy with mezzanines but i acknowledge the need for their existence
this is just fuck you
This does not invoke joy.
š¦
tbh its probably power
each side of a qsfp28 cable consumes about 3 watts
thats 12w per card
basically 1.5x the power consumption
where just direct straight through copper is 0 watts basically
I see zero reason why that can't be solved on the host device?
The card itself. Each side of a QSFP cable using that much power should be largely irrelevant.
I don't see a possible reason why the cable needs be different unless QSFP can't take enough power to run.
Mmmmmm, but an entirely different standard?
and and
each side of a serdes costs around $40
No possible way to not just fuck off the QSFP formfactor?
qsfp is closer to uh
pcie
than an actual network connector
it contains nothing that allows data to be sent long distance
you have to provide those modules yourself
yup its pretty fucked
Fragile pin city, ffs.
oh nah
this gets plugged in once and is done for good
since you cant use these cards outside of their system
Dude, you wouldn't know this about me, but contact pads are maybe my favorite tech thing in general.
I fucking love contact pads.
A better system for modular designs one could not have.
Y'know, assuming someone can just fucking make clips that work and hold in place. -_-
LGA can fuck off.
Really seriously.
A more moronic pin array one could not ask for.
lga is the best thing for processors to happen since pga
It's also supremely fragile.
not on the expensive side š
Tell that to the people coming up to me saying "I dRopPed tHE CpU iN tHe SockEt, WhaT Do i dO nOW?!"
my cpu made a crunch noise when i put it in
that doesnt happen on lga
uh
if you drop your pga cpu you are gonna have a bad time
LGA is fantastic when people aren't retarded.
Unfortunately people are retarded.
Not even once have I ever dropped a CPU when putting it in, it's not hard.
But apparently enough people have this problem.
ive coached people through installing their first cpu several times
none of them dropped it
however
ive seen many many bent pins on brand new am4 cpus
ive never seen someone bend a pin sure
but they will come bent from the factory
but also
my cpu
came with perfectly straight pins
but wouldnt slide into the socket

i examined it many many times
all perfectly straight
Fuck you, Doge.
:D
That's the life of the resident tech man.
still wouldnt fall in
love ya too
i had to press on it
fairly hard too
and it made a crunch noise
absolutely horrifying

