#🔌│tech
1 messages · Page 179 of 1
It's not
Gamers have less amd than supercomputers do
Half of the currently installed gpu tflops outside of China are Nvidia and half are amd
AMD is underdocumented and their support system does not reply.
I've sent AMD several emails inquring about their API and it's just a bot that responds with links to the incomplete documentation.
Once aurora comes online it will be 1/3 each Nvidia amd intel
Almost all of our simulations still are being written in OpenCL still because of this.
The "not replying to emails" part is a huge issue with AMD in general. I've had this issue when inquring to buy their EPYC Rome processors aswell
We wanted specifications on instruction latencies and branch costs for performance profiling for our DAQ server, and they reply with links to their marketing page with 0 processor specifications.
If you want any useful info you have to sign up for Twitter, get a few thousand followers, and at Lisa su lmao
At least threadripper might be making a come back
Or, I could just buy Nvidia & Intel systems 🙃
Which was the decision we came to. Running 8450H with A100s
But then you are giving money to the company that makes good products rather than the one that makes the cheaper gaming chips 😦
AMD makes good products, they just don't talk to you unless you have over $100M to give them.
yeah
AMD's EPYC series are more or less simply better in every way than Intel's current options for datacenter processors.
the only technical info you can get from them is how much it cost to get in top500
to be fair nvidia is like thast
Nvidia provides support if you email them.
not true
We've had Nvidia devs basically walk us through fixing bugs on our cluster's DMA interconnects.
Even I've managed it and our company spent less than a grand with Nvidia over the past 10 years lol
Literally just the GPUs in my and my coworker's computers and they still helped me out
spent months trying to get them to help with a mosaic setup
and it ended up being easier to make a fake render port
Actually nvm come to think of it we also had an entire gen of desktops with dgpus for admin, but that was still only like 50 systems
than displaying it over a firepro monitor array
It helps if your email is from a corporate domain and not your personal
so quadros were rendering it
all my emails are for bis
and this was rendering on a6000s
did they just ghost you or what?
AND autodesk tech is who walked me though doing the dual brand setup
so they knew nvidia was useless for it too
Some of the gpus have maximum monitor count limitations
Nvidia is extremely consistent you always get four per core
this was 5 a6000s with a sync 2
And so what happened with Nvidia what did their support tell you
Did you keep opening separate tickets or was this just in one ticket
At some point you just got to say you're not helping escalate this to your whoever
one ticket
think its still open
lol
pair of old as fuck w600s
and a amd sync card
fixed it
That's pretty shocking
Usually they blitz through tickets in my experience
they were fast to respond
with shit that didnt help
they kept treating it like a software issue
when the entire wall would be showing different frames
imagine the literal worst possible frame tearing
there would legit be 4 or 5 difference frames being displayed in the array
Yeah I know video matrixes lol
amds 10+ year old cards worked
🤷 idk that sounds like a bug and a bad l1 tech
id send in a video of it
and they told me to update revit
like fam
its impossible for the software to do that
and then id send a picture of a video playing
and they said the system isnt desigined to display video
yea, what happened when you escalated?
back to square one
same shit
we gave up because of deadline
dropped 3 of the a6000s for pennys worth of amd shit and it worked
thats pretty crazy
maybe the tech passed you to a colleague to avoid escalating properly?
my coworker had that happen to him for this pile of thinkpads
he didnt know how to insist to escalate hard enough
now this issue is a bajillion times worse than anything nvidia could do
these laptops wont wake up from shut down unless you cmos clear every time
this was the only tr pro i worked on
yea well what option do you have at this point 🤷
3995wx with 512gb of ram and the a600s
thing was a beast
and its funny to think some old shitty gpus fixed it
really unfortunate, but i suspect something could have been done if you escalated hard enough
i think its more nvidia dosent actually support 3d rendering over big display setups
they do there are just max res limits
it had no issue rendering it, just displaying
we just got 8k screen simulators
and mirrored it
so it wasnt a render bug
something in the actual display system
look fam
if disney can do the mandalorian video wall on rtx 8000s, the gpus actually can do the displaying
the question is if theres a bug that they had to work around or something
they could also be doing sync external
isnt that the entire purpose of the sync card?
yea but that passes it along to the sync card right?
no
what?
the computers just output video
it literally has a sync input bnc...
with the frame number encoded in the image
the frame holder holds the frame
until the sync pulse for that frame comes
its how you get a huge number of computers to do one image
are you saying the nvidia sync 2 card holds video data?????
if you use more than 4 monitors im pretty sure you need one?
ah
so you were trying to use 3rd party software to get around needing a sync card
right?
??
we are talking about how someone like ILM dose a video wall
they likely dont use nvidia sync cards
use something like a BM quad SDI
they have to lol
the only way to get the cards rendering in sync is with a sync card
no
you may be using a 3rd party timer, but the only way to get timing info into an nvidia card is via the bnc on the sync card
so how does the gpu know when to start rendering?
video walls that arent synced are insanely awful to look at
thats why you use something like a decklink
a decklink cant genlock a frame rendered on a gpu
yeah it pulls them out of the frame buffer
well i mean i suppose it could by prerendering it, capturing it, then playing it back on some genlocked output
and has a frame sync api for what ever program is rendering the frame
ah so again its not the gpu that is syncing that is the issue
i dont see why that would be nvidias fault tho?
ilm uses genlock
the frames are generated when requested for by the sync server
they might be using nvidia as the syncing server, but probably not
either way, they are making sure all of the various timecodes are within nanoseconds of each other, and the only way to do that on nvidia is with the sync card...
they definetly arent rendering to a capture card, then having that capture card perform the timecoding
heck they might be using a digital matrix switcher, or heck even one of those programmable video transform boxes and having the quadros only render the scene, with the video matrix handling the 2d transform
but i can guarantee you that if you follow the bncs that are coming out of each of those computers, you will eventually find a single master clock box that is hooked up to both the pcs and the cameras
yea its pretty pedestrian resolution
the hard part is genlock
its well within the capabilities of a single gpu tbh besides the raw rendering horsepower required
i know they use TR pro 5000
sounds to me like you dont need genlock tho, you just needed matrix
oh right the canvas max per single gpu is 16kx16k, but a single gpu doesnt have enough video port bandwidth to actually have 16k worth of monitors
but yea 16k is the maxerino no matter what on nvidia
i think quad sdi 12 and the matrix are the only acutal 16x16
also
16x16 is only a windows limit
but nvidia treats linux like its only terminal
well i mean nvidia doesnt have mosaic for linux 🤷
hahan yeah
does linux even need it?
actually nvm i already knew it does bc of how atrocious my experience was with just dual monitor
sweet
what is that from?
the quadro resolution pdf
im really curious how they expect us to have that many pixels physically attached to their gpus...
whats?
dont they have a tiny quadro with like 8 mini dp?
it doesnt support sli or 8k
its maxwell at best
32k 
yeah but 4x8 = 32k
oh right it doesnt have to be a square
so yeah dci from 8 mini dp would be 32k
yea
so you wouldnt run into the limitation with any single gpu
or any 2 gpus
you need 3 minimum
and fully horizontal tiling
i though nvidia had a 8 mini dp card?
it doesnt support sli
and afaik it also doesnt support sync
well maybe lemme check
but
nvs820 afaik
yea
which is within the limit
810
yea 810
oh
810 is dual gpu
that has to be the WEAKEST dual gpu card made
dual 512 cuda cores lol
and you can nvidia mosaic 2 of them sweet
so yea using 2 of them you run out if you do more than 8x horizontal 4k
actually you could run out before that horizontally with a ton of og g9's
pretty sure thats more than 360 degrees of g9 lmao
signal integerity engineers fear him
lmfao
what a monstrosity
i suspect you arent looking at more than 1080p for each of those monitors individually
dude never got more thjan 16 displays working lol
yea nvidia advertises that as the max
1920x1200
I connected up 16 screens in a 8 by 2 matrix to the cards in slots 2 and 6 and it worked OK. The PC was still struggling when using Windows display settings and nvidia control panel. After applying the video settings it took at least a minute to settle
hahahah
you need a quadro sync card for more than that
yeah
now i want to a TR pro
with a fire pro sync
and like 7 firepros
just for meme
wait
i have the perfect machine
im pretty sure i cant really add any more
my 1366 machine
my little 1L lenovo is already struggling
get like 5 W600
and fit it into what pcie slots lol
im talking about my 1366
it has quad quadros, a og tesla and a pair of ssds in it
i can technically go up to 7 monitors, but then id be using the intel igpu, and it was struggling until i disabled it
it has 7 pcie x16 though
so i could get 7 w600s
and 2 fire pro syncs
and run 56 displays
much to the dismay of the windows display manager
tempting
I was malding quite a bit when windows swapped to a compositing window manager in win10
It really sucks you can't disable the compositor anymore
what is the cheapest small 1080p lcd i cna get?
x56 thats still really expensive
are they dp?
no
find a school offloading a computer lab worth of monitors, they are often under $5 each in bulk
oh wait no under $10
$5 might be hard
those are rarely 1080p
yeah they are these days
they are now
ive been watching a few local school's surplus, and its pretty rare to see the 4:3 stuff nowdays
actually come to think of it
it would be pretty nice to have a 4k:3k monitor
triple ultrawide 4k:3k monitors 🤤
3:2 afaik
rats
yea thats a standard photography aspect ratio
i wouldnt be too beat up by 3:2 but i dont think its really enough more to make a huge difference
i still miss 5x1 portait spanned
well you can see my setup here and i like it plentyh
the little tiny boi is where im talking to you and controlling my music
right now i have 5x1
but only 3x1 in the center spanned
with 2 wings
then a single display up top
so like if i rotated my right monitor and moved the small boy up top and made it larger
left
rotate left
ooooh no you mean like video wall spanned
yea ive got multimon for multitasking
each monitor has at least 1 item on it, got a few monitors with more than 1 thing open
i should be able to do 5x4k60 on amd
depends on the gpu but yea
any with 1.4a and mst
🤷 nvidia is always 4
intel is always 3
amd is 3 or 6 and they only advertise which it is on their expensive stuff...
r9 280
non x
and a huge list of their lower end gcn stuff
or was it r7 280
nah it was an r9
on the 280x
they would do 6
yes
if the gpu has 6 clockgens it will work
if it advertises 6 monitor max, it will work
if it doesnt, its a gamble
dont need clockgens for dp
mst is one clock gen
no
like the gpu uses one clock gen
the mst hub has clock gens for the outputs
its how mst works
nvidia still hard limits 4 displays in driver
🤷 if windows can see the monitor as separate, it needs its own clockgen
yes
but mst hubs have clock gens
so with a 280 you can get a 4 output mst
plus the hdmi and dvi
1.2
"The MST Branch device must support four lanes both on the upstream uPacket RX port and the downstream
uPacket TX port, and must always set the downstream link configuration to the maximum link bandwidth in
order to avoid the interruption of stream transport due to the link configuration change. "
page 463 of the 1.2 spec
?
every single block diagram i can find shows that the source device generates the datastream and then sends them muxed to the phy
the only exception is if you are mirroring, where you can send a single stream that is cloned by the "branch device"
i dont see any examples where you can take a single stream, and then extend it between monitors
because if that is true, everyone is bsing us
you should in theory be able to video wall 18 monitors on a single w5700 using mst
https://www.ebay.com/itm/255740041316
what a beaut
ive never heard of an mst hub where the gpu thinks its actually 1 monitor...
Multi-stream transport (MST) format supporting the transmission of up to 63 independent AV streams from a
single DP connector, taking full advantage of the micro-packet transport architecture
right from dp 1.2 spec
yes but that still needs 63 independent frame generators...
yes and it doesnt exist because you need 63 clockgens...
the most anyone has connected to a single phy is amd's 6
and apples 6
actually hmm
i wonder how much chaining you can do with apples mwhatever biggest chip
cause im pretty sure you have 12 clockgens on the big boi
shame they dont allow mst 🤣
actually yea why is thunderbolt daisy chain allowed when mst isnt?????
ye
but apple has a chip now with 12 clockgens
so we may be able to get 12 monitors on one port
without mirroring
well
"a chip"
what
tesla gpus
if you go old enough, you can even get video out of em lol
what tesla has video?
its only the a100 and h100 that are unable to render
i have a orginal g80 and it has no video
🤷 the one techpowerup got a photo of had display outputs
yea again not all of em had em
but there were 2 or 3 with dvi and 1 or 2 with a minidp
this is for the c870 pdf
and has a c870 having a 3dmark03 score lol
and it has no outputs
lol
VDI was hype in like 2000
maybe they just did the testing with a FX 5600
thin clients have existed forever basically, the teslas supported that...
90% sure the 870 only dose cuda
The Nvidia Tesla C2075 was a fantastic workstation graphics card back in the day and it was never intended for gaming. However, Nvidia decided to attach a DVI port to this one so I couldn't help but try to run a few games, albeit with disastrous results.
Thanks for watching :)
lol
nvidia fun
that isnt the same gpu
now this is very possible, but until kepler there were editions with video connectors
and kepler and later definetly support vdi bs
yes there were 3 or 4
it didnt have video output
guess youve gotta remote in then
the a100 and h100 dont support rendering at all
but prior to that they were the same silicon as the quadros
v100 (same as quadro gv100) is the last flagship die with rendering support from nvidia
potentially forever... :/
these predate vgpu
again not vgpu
kepler predates vgpu
nvidia grid style proprietary VDI Bs
VDI is not VGPU
then what software
dunno
remote3d?
didnt exist
🤷 yea i cant find the name
it was that 3d api that microsoft dropped support for 3 or so years ago because it was insecure
remoteFX
that allowed for multiple clients per gpu
but without that there was dedicated thin client software
didnt exist either
VDI existed in 1994
not with gpu support
but oracle has purged their ancient documentation
looks fine as an icon
its not ai
The eyes are too low res for me to tell
But normal ai eyes give me the Heebie jeebies
ai is great to ignore copyright (even if not legally at least from moral standpoint) and put cursed danny devitos into movies scenes
Firefly is proof that's not necessary for good looking ai pictures
100% self trained
By adobe on their own stock photo gallery
I trust adobe as much as I trust them to make it easy to cancel subscriptions for a fair price
Artists don't always either tbf
For me so far I've used it as a reference point in my next commission.
Which is why the ai does it sometimes
But it does it wrong, and the eyes look messed up too tho this is one of the better examples
I've commissioned so many projects
And I won't stop cos or ai that's for sure
This is more an art style really
I really think that geohot got it right
The ais are currently at the level of $5 on fiverr
For coding, art, philosophy
Anything
But they will get better
Their shirt merging with their jacket lol
It's good enough
But it's no masterpeice
oh yeah it will get better just for now it's good at fucking artists over more than it helps
but hey, gotta take what we get for it
Can anyone help me, my wooting stopped working all of a sudden and I need it to work for my job
Feel line that will take a long time..
All of the artists who don't use it are definetly ruined
But it's a tool
Anyone still cutting out film plates got their careers turned over by photoshop
it will be faster than it getting buried here because its not a support channel
Less time than arguing with us lol
Maybe some of yall had the same idk
oh it being a tool like a calcualtor sure, the issue is how it got the power to do what it does with training data and all is the issue I see for art
That's how art be
When a human is inspired it's ok but it's not if it's a machine 🤷
People will need to get over that sentiment, or make it a law
HAHHA
I mean sure but it's more like tracing art vs copying it using your own skills
i asked it to merge musk
It depends on how much you are tracing
There's plenty of beautiful art that takes some other art and then traces it and then reformulates it somehow
and yes laws arent always there for it hence I said (even if not legally at least from moral standpoint)
And the law can't do anything about that yet because there's no clear metric on how close is too close
But the USPTO or whoever will decide that and that's how it's going to be in the us...
https://twitter.com/Buildzoid1/status/1681454391252094976 ok what AI did amd use to suddenly make the imc insane??
was always good
more like when did asrock get good board designers
maybe they snagged evga's team
lol
maybe, but it was that past 6000 wasnt really a good think on any board so unless they found some good fix
Well carbonaut is tiny
A lot of AI are claimed to be trained on ethically sourced images, such as stable diffusion.
But if you type "morocco stock image" it will still show the dreamstime watermark 
from diffusion 1.5:
yea but theres a difference between "ethically sourced"
and "we are a stock image site trained on our own images"
Thanks to Keeps for sponsoring this video! Head to https://keeps.com/robdahm for a special offer.
T-Shirts: https://www.dahmracing.com/collections/shop
Patreon for Build Videos: http://bit.ly/2koiw8G
Instagram: http://bit.ly/RobDahmIG
Subscribe to Rob Dahm: http://bit.ly/1OAN9yi
AWD 4 Rotor Video Series: https://bit.ly/2QiFduE
–––––––––––––...
what a monstrosity
Yeah "ethically sourced" is almost always a lie
Mang, we gotta use AI to replaced the biggest cost of business, the executive leadership
that time I found a pentium 4 northwood extreme in my Lego bin
That was earlier today
And was the one positive memory I have of any Intel product
noice
Also yummy cap juice isn’t actually that yummy
Probably for the best
…
ok
Personally I go back to wasteland at most
Speaking of which
Best game

My dad might have played it in uni idk
fallout 2…
crap I haven’t played that in
several years
…another game to revisit
Rage inducer nine thousand
Clearing Brahmin woods, the tutorial level, took me three hours
The first time
Now it’s a twenty minute adventure in murder
pesky raiders
Either way, St. Louis is my least favourite mission of them all
the muties have good guns, thick hides and my kalash collection at that point in the game tends to not do that well
But the M2s with the .50 BMG depleted uranium rounds
A bitch to obtain
But so lovely
They just melt things 
Usually I just set up farsight with a sniper and just have her go at it
But sometimes the muties actually start shooting back even when out of range
true
But then again
That’s how video games should be played
Unless it requires a wheel, flight stick, or fight pad
Which are very specific
Never heard of it
Should I be curious?
seems…
interesting
That does seem interesting
Oh damn a lot happened yesterday
At least 2 different sources came out with exact same images for a new Switch 2
That looks slick
Probably a fake rumour, but that looks sick
looks superbly fake
what is that apple ui?
xbox 720 ui
it looks just like apple tv ui
The mockup of the console itself looks pretty terrible and not very Nintendo like.
Yeah that's ugly AF
Also mechanically seems like a terrible design
Also
That UI is too animated and fluid for anything Nintendo make
Nintendo like very fast and linear animation
Lol the rumour is getting a bit faker
Announcement "tomorrow"
It's a pity cause I like that design
Idk why the joycons are looking like that, there whole bottom portion being weirdly empty, maybe a speaker is over there?
But it looks slick
Too bad, likely fake
Actually the biggest giveaway is the what is shown on screen.
wtf is that chin
also why would they market a switch 2 with old games right on the box
and only old games
Actually the fact that there are games on the box is already a dead giveaway
Isn't that cause we have no confirmed next gen games from Switch and also it's backwards compatible?
as doing so would cause confusion about said games being included
cod

Nintendo has only ever included games on the box actually included with the system
prime 4 100%
The only thing we really know that will be maybe next gen is Metroid 4
fortnite moment
Also Switch 2 is supposedly backwards compatible
but yea
So them marketing it with Mario Wonder makes sense
Since that's their "next big thing"
putting on old games and only old games is something they wouldn't do
looks fragile
Though that stand/dock looks hella weird
those joycons looks like they are attached via magnets?
or something similar
definetley not something you want
If Switch 2 retains Switch sizing, will that dome dock even have enough space for HDMI and shit?
that ledge would also be uncomfortable
Biggest issue with the joycon design is even if you secure them properly the angles would allow for a lot of stress / leverage when pushed from the top edge
and making the joycons wider than they need to be without actually more comfortable is also something I doubt they would do
yes
yea you don't want them to just break off
so reducing the contact area with the main console is a horrible idea
Well if announcement is tomorrow then we won't have to see for long if some intern actually leaked the promo material or not
thats false
have u seen the tears of the kingdom box?
but yes the tablet itself is either off or shows some colors nothing else
ngl
id still welcome a new switch announced tomorrow
Fair enough, just saying they would never show a reel of games on the display itself like that.
Why not?
The only big game they really have coming up is Mario Wonder
And it's literally in the center of attention on promo material
ToTK is there too, Xenoblade and whatever, the big titles
If they aren't ready to announce next gen games yet, and just want to announce the console early, doing it with last-gen games to announce backwards compatibility isn't a bad shout
ya no deffo not a reel of games on the tablet
in general everyone would jump on a slightly more powerful switch
super mario rpg?
Like, I do hate Nintendo for their ninjas, exclusivity and everything, but they are the only company whose console I would actually preorder/buy
cause it's Nintendo
this succa
Isn't that getting a remaster or somethin
Either wayt I don't think any reasonable person would call that image anything but fake as well as the render in general.
yes
The titles on the screen to me scream anything but fake
could be based on real product details and just fake renders
The design of the display though looks a bit too good
but i wouldnt be surprised either way
There was also this image
Apparently from instructions
I can't quite understand how the joycon locks in place
Says it's magnetic
But that is a bit expensive, no? And fragile, potentially
why would u need to slide it down for a magnetic lock
Wendell talks Genoa and Bergamo from the new 3d V-Cache CPUs from AMD
AMD Tuning Guides:
https://www.amd.com/en/processors/server-tech-docs/search?f[0]=document_type%3A14371&f[1]=document_type%3A14376&f[2]=product_type%3Aepyc_9004
0:00 Intro
0:49 Overview
4:05 96 cores to 128 cores
7:30 A few more details
9:28 Benchmarks
18:44 Rec...
Phoronics can’t find an increase build performance because they can’t fix their test for it lol
MORE CORES
fix? as in, tune it to give better results for 3d vcache?
No as in if you have 512 threads and don’t spawn 512 jobs it’s not wonder they don’t see more build power from more cores lol
Runs at 60%power because you don’t give it more work lmfao
To be fair to phoronix is just one guy
Yes but it appears to be known so if you can’t fix it don’t make graphs as if the product can’t do more lol
Fair enough
So do other benchmarkers show big gains in compilation then?
yo guys, can anyone tell me..are these keycaps compatible with wooting 60HE?
https://www.amazon.de/MOLGRIA-Tinten-Lotus-Tastenkappen-benutzerdefinierte-Farbsublimations-Tastenkappen-Tastenkappen-Abzieher/dp/B0C3QG99C8/ref=sr_1_66_sspa?keywords=corsair%2Bkeycaps&qid=1689772106&sprefix=crosaair%2Bke%2Caps%2C87&sr=8-66-spons&sp_csd=d2lkZ2V0TmFtZT1zcF9tdGY&th=1
This seems to do, haven’t gotten around to check it
But if hothardware can do it lel

Allegedly caused by some gmake thing but idk anything about that but at least power consumption seems to be represented well to be lower
complie is likely a bandwidth issue
hence the 32cx2 doing better than the 128c
teh real question
will we get a single ccx 16c 3d cache desktop part
I trust someone like andrei more than my own guesses on if bandwidth is an issue or the compile is set up wrong (the ex anandtech guy who did core latency measures and all that)
kinda disappointed by the io
2 fewer usb 4 ports than the smaller intel machine...
intel has TB on the soc
sorta
both need the same number of offboard chips to get things working
the chips intel needs are smaller and cheaper, but you still need the same number of chips
anyone know what kind of keyboard this is?
Hey guys
im planning on saving about 2-4k usd over the next year so i can build my own gaming setup. I have two questions. Should i go for a white and blue themed setup or a black themed setup? Also, should i build my own pc or buy a prebuilt
save first pc later
yea but i want an idea, currently i should have about 1.5 in a savings account from my birthday last year, so i want to start there
Colour is all personal preference
You can definitely get a better PC for the money by building it yourself rather than getting a prebuilt
thanks!
and use the money saved on building yourself to buy more storage or better monitors and keyboards and whatnot
the wooting keyboards give a hardware advantage but is there any mouse that gives a hardware advantage like the wooting?
zaunkoenig lol
Not yet, but plenty of people using razer pcb to make their own version
they are selling their remaining units, from folks who cancelled their orders for the m2k
m2s and m3k later
Only south korean companies offer that
i just wish someone would make an absolute positioning mouse
infinite g within the confines of the mousepad
I just meant zowie and vaxee
those dip switches alone weigh more than the zaunkoenig probably
China keeps spewing out clones with the same specs
Probably only going to buy from logitech or vaxee moving forward
Maybe EGG
Swift point uses one button😏
zaunkoenig doesnt even need a single button
nor a led
but it still allows you to set whatever you want, and inform you of the result
Would be cool if a company provide a usbc dongle for your mouse that has buttons to change settings. And you just unplug when you are done
why?
An altetnative to having the buttons on the bottom of the mouse like vaxee/zowie
Reduce weight as well
zaunkoenig uses zero additional buttons or leds, and has full control over all settings with immediate feedback
you dont need to copy the shape of the mouse to copy the ux
🤷 most of the settings are set and forget, and a normie mouse could have pictograms printed on the sticker on the bottom
@prisma crag what's happening with that other mouse you were looking at?
I have a question, so I bought the lekker switches for my K2 Pro that is said to be "hot swappable" and the switches aren't working (i don't know much about keyboards)
Looks cool, shame the shape is definitely not for me
seems wired g-pro ish
With a bit of g303
Like putting petrol in a diesel car.
One of my dad's friend did the reverse
more like putting batteries in a diesel car

dang okay, i like the 75% and didn't want the 100% or 60%, i only put one switch in to see if it works then came here. thank you
i mean technically diesel can be an alcaline solution 😛
so then would it just be better to just rip the keyboard apart, throw in a magnetic sensor or just return the switches and buy the 60HE
The later
you cant really just add the magnetic sensor even if you wanted to
actually i wonder how much one of those magnetic reed switches would cost these days 🤔
cherry mx hotswap to magnetic reed wooting switch compatible conversion kit when
Experience crystal-clear audio with the iFi Audio GO Pod, a pair of wearable Bluetooth DAC/headphone amplifiers that make your premium in-ear monitors (IEMs) wireless. This revolutionary device is the world's first to support Hi-Res 96kHz/24bits with LDAC and Qualcomm Snapdragon Sound codecs, delivering an unparalleled
I remembered some people being interested in this
Well it's on preorder now
me for 1
how much
400
yikes
Yeah
ill wait for reviews, but i suspect not
Mostly for people with $1000 iems already
i pulled out my old galaxy buds+ and ive been really enjoying them, their wonly sound is so much less wonky than the chus that im pretty happy with them now
Yeah the cobra is such a throwback type of mouse from razer
Rgb, shiny plastic, charging dock
Heavier than my RVU

yeah no one likes the cobra lmao
i saw one yt review ask who it was even for
i had the viper before i switched to my glorious model o, my endgame mouse is prob the viper v2 pro
oh yeah i forgot zankoenig lol but is there anything else?
besides weight and having a perfect sensor, theres nothing objective about mice
everything else is personal preference
these folks allow you to take a shape and pair it with a pcb you already have
People wanting a g703 superlight for years
Ooo
Would have been nice to have options when buying the 60he
The Module will be a big jump into that sort of thing.
omg Calder leaking things left and right
Next think you know he'll be leaking the additional Alumaze colors....
Ffs imagine putting text as a graph
8490H is the sapphire rapids chip from Intel, no?
So their latest
among others but yes
Wasn't there also some xeon max? That isn't compared
xeon max is a competitor to genoa x, only difference is it has HBM onboard
xeon max is what Mi300C would slaughter but people dont want it
i mean you say that but intel has installed more flops than AMD has in the top500 in just 1 installation
yeah but amds customers did not want a mi300c, the ones that are using xeon max did not want to switch so it's not a product anyone would have bought hence it didnt get made
no quin you dont get it
aurora, which is pure intel, has more flops than everything amd has in the top500 combined
because LLNL is getting el capitain up before aurora sure
but also how does that relate to the product itself
they werent going to put mi300 into aurora and others werent interrested in it so the sku didnt come alive
I'm not saying people dont want xeon max
yes because xeon max is very specialized so for those it doesnt make sense to swap even if hardware would be a bit better
Intel Xeon's haven't been used in HPC for a while. They're mainly bought for high reliability datacenters.
but it's not a massive sku that will save intel any more that other parts
We used PowerPC processors for HPC until EPYC Rome released
intel has been out of the industry for about 8 years due to their low IO bandwidth
I mostly think its funny intel does fancy cpu version and amd is like "so we can just make that in better if we remove the gpu in our next gen APU"
common misconception
intel has all the gigantor customers, they stopped selling to the small dudes due to lack of stock,
if they had stock, they would have priced competively
We are the gigantic customers
Summit is the largest supercomputing cluster in the world on construction
intel was pretty much never a consideration due to poor performance, not price. The IO bandwidth of Xeons is dogshit
gotta remember that xeon can go up to 8p, epyc is 2p
you go epyc because each p is the same price, but if all 8p of the intel were the price of the 2p of the amd, it would be a no brainer
powerpc processors are triple the cost of intel and epycs and is only 2P
Yet we still built computers with them because 2P on powerPC is more IO throughput than 8P intel
also what was the oak ridge expansion called?
bro a amd employee basicly saying theya re using buildzoids oc is amzing
i cant find anyone talking about expanding summit, but i remember it being a thing
that sounds wrong, biggest power chips are like 15k in bulk, intel at the time was 10k, now closer to 20k
an 8p system would be 80k in cpus alone vs 30k on ppc
8P? It provides near 0 performance benefits as something that parallelizable can ran on GPU
gpu dont have good pref and reliabilty on numbers
Power9 chips with nvlink was $32,000
Worked fine for us
for trading?
No
exactaly
last 8p xeon i worked on was like 10+ years ago about
and it was trading
and they do use gpus too now
the guy that manages that client now dose only those
they are 8x56 core, with 16-32TB of ram and tesla too
they run like 6KW under load
for just cpu/memory lol
and how much would the whole set of nvswitches cost if it wasnt built in?
afaik thats not a benefit to intel
Yes because intel doesn't have any IO bandwidth on chip
which was the problem I pointed out
its also not true
they had 48 PCIe 3 lanes
sure each individual intel chip only has about 1/4 of what the others have
but you can attach 8 of them together without cutting into that IO
its a fkn dumb architecture ofc
but it was the direct comparison
Its lower than 2 power9 chips
350GB/s per socket
2P Power9 is almost double 8P intel
didnt a significant portion get sacrificed for the intercpu interface?
Its not counting the intercpu interface
the 350GB/s is purely to extension devices
not including memory/interconnects
oh right
ok so power9 has 48 gen 4 lanes per cpu
its not pcie
8p intel plat is up to like over 100k lol
yeah the 60c is 17K list
so 136 grand for 8
Its substantially cheaper if you buy over 10k processors
Titan's xeons were bought for 35% below msrp
yea, on top of that, they have a few proprietary interfaces, you get 300GB/s dedicated to nvlink per cpu
sure thats alot
but total io is still in the same magnitude if you add nvswitches to the xeons, and you end up with the same latency (including the penalty from 8p xeon), and you get to separate gpu-gpu comms from cpu-cpu comms for better optimization
if xeons + nvswitches were the same price, im confident they would have won the bid
but they werent
they were near an order of magnitude more expensive
Separate GPU/CPU comms is not more performant in clusters. The inter-chassis communications is over SPF, which requires the CPU to stream data to the GPUs.
A high bandwidth channel between the CPU and GPUs are at some point needed.
which exists
and was the same bandwidth for both systems
sure the xeons could only talk to a maximum of 4 of the gpus at full bandwidth at once
but they could and did in other systems
The Power9 system achieves ~300GB/s on DMA to the Nvidia V100s
Intel skylake Xeon platinums could get 16GB/s to each GPU.
over 192 pcie lanes
Intel CPUs did not have 192 lanes.
That is lower bandwidth than 2 Power9 processors still
Even if they were cheaper, they are simply worse for GPU clusters.
That is almost never used aside from ML workloads
🤷 and when you have large datasets loaded into gpu memory
sure nowdays thats ML
but flip and other volumetrics need hella data
The workload at ORNL was primarily physics simulations. The communications needed was passing boundary information, which requires more node-node bandwidth than gpu-gpu bandwidth
2 of the gpus on power9 had full 150GB/s each
but the other 2 were stuck talking through a shared 64GB/s link
but this was a workload architected around that limitation
if they could have put more data into the gpus local memory to make things faster, they would have
but this argument is purely academic because intel wasnt selling their cpus for anywhere near cheap enough to consider, and their (of the time) curreng gen server chips (ice lake had a ton more bandwidth) got delayed by 5 years 🤣
8x 64 gen 4 would have been plenty of bandwidth, even if you had to reserve half of it for networking
but that chip was still vapor in intel's silicon boule forges when you needed the stuff delivered
I would do this, but only for the exhaust
thats what i got rn cause moving countries soon, otherwise wouldve gotten another good 1
1 good and 1 cheap is better than 2 mediocre imo
there are some real nice cheap monitors these days
enjoy your fps issues with screens mismatched too hard 
hey its good for now
for now(potentially) (if you are lucky)(till next driver update)
i have both of my original boxes
they are being a table
yea i dont have the space either, yet there they remain
its actually a problem
woah
i wonder when i last turned mine on
in my room, my pc causes the ambient temp to increase 5c in just 4 hours
well i mean if it was bought at that time, the 1060 was tied for best encoder with all of the other pascal cards with an encoder
yeah, aka it makes me turn on AC a few degree earlier 
😭 i would like ac
saaame
dex is awesome
my phone is already on, might as well put it to use lol
its just the portable AC inside, I modded it to have 2 hoses and put a polycarbonat plate into the windows instead of those cloth things for the hoses but its so nice
for gamin i have a deck!
you cant boot into proper linux and use that as your work os?
i have a pair of "reclaimed" chromeboxes that now have ubuntu installed on my desk
at work
you are only throwing out the security when you are booted into linux?
yea but if you disable dev mode, it should boot back into chromeos?
i mean mine dont because i wiped the boot drive, but if you are booting linux from usb or sd card, that should work fine
i thought it was just a screw?
these are old devices, so it may be different on the new stuff
but they are new enough to use the custom uefi
its very possible
these machines are rather old, and i dont even know if i could have usb booted without wiping
but i imagine its probably possible to boot to usb without losing your data?
dang
dang it seems theres no way to usb boot without dev mode
unfortunate
unfortunate
have you investigated microsoft's trusted core machines?
sorry secured core
as long as you dont use the administrator account as your login, and enable all the features to get the "Your device has all Secured-core PC features enabled" message its supposed to be real secure
i get updates on my phone about that late :/
sad that thats actually relatively speedy for android
🤷 theres plenty of options, they are just all obscenely expensive
secured core yeets a bunch of insecure stuff, and virtualizes plenty
yea microsoft really just has their enterprise users do stuff
you want a locked down windows machine?
https://www.microsoft.com/en-us/microsoft-365/enterprise-mobility-security/compare-plans-and-pricing
heres the subscription you can pay to have microsoft do it for you lol
windows 11 S mode 🤷
you want to run an app outside of a container? get rekt
its effective
Get a machine with pluton, with how new and unknown that is it could be something for penetesting could it not?
and theres a crapton of malware for mac now, and the mac antivirus options are relatively poor
pretty sure theres both a wireguard and openvpn client on the windows store?
and if the provider doesnt provide those as options, you shouldnt use them...
but also the big thing is that pentesting on windows assumes the logged in user is an admin
because thats almost always true
yea
thanks apple
tho technically per the letter of the tos, you can run mac vms on a non apple device, you just have to have created the vm on apple hardware
see if stuff like the 2012 mac mini still existed, with non soldered pcie ssd, 2.5 hdd, and memory slots
sure, get an apple
you can now run macos on a ryzen machine in a vm
ryzen + nvidia even
with 3d accel (tho limited)
when the last updates come to intel machines
they usually cut off official means long before the unofficial ones
because powerpc was ended support in 09
its identical to the ppc to intel transition
except this time apple is moving alot slower
the only thing ever ported back to ppc was the beta stuff
that apple made of ppc then abandoned
nothing from the launch of SL or after has been ported back to ppc
the ppc to intel transition was 2006
the last update for ppc was in 09
with the SL beta
offical SL launch didnt support PPC
yea and the last time they sold ppc was 2006
they transitioned the mac pro literally 2 days ago
a solid 3 years after they started the transition
if they follow the exact same timeline (last ppc mac discontinued to last update) we are looking at 2026
but in 2006 the transition was done within a single year
this time they took 4 years to transition
they are moving alot slower this time...
To be fair: The Mac Pro is the tier that Apple always forgets about. Like during those three years they never updated the Mac pro platform at all, and it was even still running at the same price too
i still love my g5
i have a mostly working quad core
i just need to fix the liquid cooler one day
except thats not true
they kept releasing gpus for it
the new mac pros cant accept gpus
at all
eventually they will release expansion cards to add more thunderbolt, but those arent here yet for the new mac pro
Probably because AMD wanted to cease production on the older cards rather then Apple wanting to update it. The CPU and mainboard on those Mac Pros as far as I can tell never got an update
the machine runs but it gets hot
i think the pump is dead
and its running just thermal cycle lol
🤷 except amd still hasnt discontinued those gpus, they are enterprise class and you can still buy them
🤷 apple chose not to, however you can upgrade it yourself if you want
i thinjk they are DDC pumps
Yeah the new Apple Silicon Mac Pro seems real pointless to me. All those slots and no GPU support? The RAM options also seemed disappointing
its actually just a mac studio but takes up more desk space until they release thunderbolt expansion cards
Suppose that is the issue of using your own silicon. Apple just doesn't move enough Enterprise units to justify making an enterprise centric chip
You do run into the question of "Why bother?"
I mean couldn't they have just released a Fatter Mac Studio with extra thunderbolt to start with?

