|
i dont actually play any of the games that i own lol doing dumb poo poo on my computers for no reason IS the game for me now
|
# ? Dec 23, 2019 17:59 |
|
|
# ? Jun 5, 2024 18:38 |
|
KillHour posted:They'll never do that. 95% of people buying Threadrippers are people buying them for productivity tasks. Those people aren't going for OC world records and running a few degrees cooler isn't going to make enough of a difference to be worth the effort. My 3950x never goes over 50 in any of the loads I regularly put on it anyways. Exactly. I didn't buy my 3900s to do stunts for the internet, to win e-peen contests, or to smugly idle within microdegrees of ambient temperature. I bought them run at 100% load, 24/7. The room they're in is currently 77F/25C, and the CPU temps look like this: code:
|
# ? Dec 23, 2019 18:05 |
|
Most of my life revolves around smugly idling tbqh
|
# ? Dec 23, 2019 18:12 |
|
I wouldn’t want a bare die MCM tbh, too much of a pain to get good mounting pressure across all the dies and such. For monolithic, bare die forever. T-bred Athlon days are back baby.
|
# ? Dec 23, 2019 18:16 |
|
KillHour posted:Those people aren't going for OC world records and running a few degrees cooler isn't going to make enough of a difference to be worth the effort. Looking at how blinged out the TR boards are, I wouldn't be surprised if people would be all over a GAMER XTREME TR. Cygni posted:I wouldn’t want a bare die MCM tbh, too much of a pain to get good mounting pressure across all the dies and such. But this.
|
# ? Dec 23, 2019 18:16 |
|
I always have the urge to spend money on a new PC but then I remember the most intensive game I play is Stardew Valley it maybe Baldurs Gate so it’s really not worth it and my laptops do just fine
|
# ? Dec 23, 2019 19:16 |
|
Statutory Ape posted:Most of my life revolves around smugly idling tbqh This should be threat title.
|
# ? Dec 23, 2019 19:44 |
|
Interesting Christmas sales stuff goin on locally. A 3800x is now only about 10 USD more expensive than the 3700x.
|
# ? Dec 24, 2019 16:18 |
|
Dr. Thunderbolt, Or: How I learned to stop the legalese and learn to love Titan Ridge. https://www.youtube.com/watch?v=59svND1ZtXI
|
# ? Dec 25, 2019 04:57 |
|
I have the Titan Ridge hooked up to a Gigabyte TRX40 board. It’s still finicky with an eGPU - I need to warm boot Ubuntu sometimes to get it to recognize it so it’s not a perfect solution
|
# ? Dec 25, 2019 05:58 |
|
Cygni posted:For monolithic, bare die forever. T-bred Athlon days are back baby. Anyone else remember the Thermaltake Orb heatsink and how its twist on mounting would destroy Athlons?
|
# ? Dec 25, 2019 23:43 |
|
Shims, bruh.
|
# ? Dec 26, 2019 01:33 |
|
Wiseblood posted:Anyone else remember the Thermaltake Orb heatsink and how its twist on mounting would destroy Athlons? Hell ya. First one (t-bird) I got scared and didn't tighten enough and let the magic smoke out. Second worked fine for three years with rounded corners. Thank god I worked returns at Fry's when I built that PC.
|
# ? Dec 26, 2019 02:48 |
|
monsterzero posted:Hell ya. First one (t-bird) I got scared and didn't tighten enough and let the magic smoke out. Second worked fine for three years with rounded corners. Thank god I worked returns at Fry's when I built that PC. https://www.youtube.com/watch?v=UoXRHexGIok Intel: AMD:
|
# ? Dec 26, 2019 03:38 |
|
my mom is visiting for the holidays and brought a dope as Tyan AMD Opteron tote bag that has to be ca. 2005 when I built my Athlon 64 3500+ gaming PC. I am absolutely stealing it. gently caress off socket 939
|
# ? Dec 26, 2019 14:07 |
|
Check out the southbridge on that goons mom
|
# ? Dec 26, 2019 15:17 |
|
What an absolute classic. Has anyone done this again since with several CPUs?
|
# ? Dec 26, 2019 15:17 |
|
Yeeess. When the thermometer read 370C I legit laughed out loud. HalloKitty posted:What an absolute classic. Has anyone done this again since with several CPUs? Anything newer than the AMD chips in that video would just shut down- I keep telling friends that PC building is easy these days. In 2000 you had to strap a block of aluminum to a bare die and if you didn't get it right and flat the CPU could burn up before you made it into BIOS. And the heatsink installation usually required you to put your body weight into a flat blade screwdriver that would core your mobo if you slipped.
|
# ? Dec 26, 2019 19:59 |
|
It got a lot easier for me in 2001 with the alpha 8045. A cooler + mounting method. E - looks like swiftech used screw mount before that. GRINDCORE MEGGIDO fucked around with this message at 20:24 on Dec 26, 2019 |
# ? Dec 26, 2019 20:15 |
|
monsterzero posted:Yeeess. When the thermometer read 370C I legit laughed out loud. I built my first PC in 1995 and the AT motherboard connector wasn't even keyed back then, lots of fun if you hosed that up.
|
# ? Dec 26, 2019 20:18 |
|
I actually spent more time setting up my current desktop than any previous desktop I've ever assembled. Cable routing is a much more pleasant rabbit hole though. Also an awkward several minutes where I had to google images of a 3600 since my "golden triangle" is just a little faded outline. When asked family members have remarked it looks like the components are just floating midair so that's cool.
|
# ? Dec 26, 2019 20:30 |
|
Pittsburgh Fentanyl Butt posted:I built my first PC in 1995 and the AT motherboard connector wasn't even keyed back then, lots of fun if you hosed that up. On a related note.. a few years down the road. Dell would often use power supplies that appeared at first glance to be ATX standard; the form factor and connectors where the same. Except the motherboard connector was a completely different pinout from standard. Using the Dell power supply on a regular mobo would cause fun things to happen.. Conversely replacing a failed Dell supply with a regular ATX supply would do the same to your Dell motherboard. https://www.smps.us/power-connectors.html shows the differences. A lot of the COM and voltage rails are switched around.
|
# ? Dec 26, 2019 20:38 |
|
stevewm posted:On a related note.. a few years down the road. Dell would often use power supplies that appeared at first glance to be ATX standard; the form factor and connectors where the same. Except the motherboard connector was a completely different pinout from standard. Using the Dell power supply on a regular mobo would cause fun things to happen.. Conversely replacing a failed Dell supply with a regular ATX supply would do the same to your Dell motherboard.
|
# ? Dec 26, 2019 20:40 |
|
hahaha! "Cloud To Butt" strikes again. https://chrome.google.com/webstore/detail/butt-to-butt-plus/apmlngnhgbnjpajelfkmabhkfapgnoai?hl=en
|
# ? Dec 26, 2019 20:48 |
|
BRB, changing my username to "the cloud"
|
# ? Dec 26, 2019 20:49 |
|
dad is this malware
|
# ? Dec 26, 2019 20:49 |
|
|
# ? Dec 26, 2019 20:51 |
|
I'm so used to seeing it, I forgot I even have that extension enabled.
|
# ? Dec 26, 2019 20:51 |
|
I like big clouds and I can not lie
|
# ? Dec 26, 2019 20:51 |
|
Statutory Ape posted:I like big clouds and I can not lie We get it, you vape.
|
# ? Dec 26, 2019 21:00 |
|
Is there anything fun I can do with a left over Wraith cooler? Is it worth scrap? Only good for a weird paperweight?
|
# ? Dec 27, 2019 00:26 |
|
The other day I wanted a big chunk of aluminum or copper and regretted getting rid of some coolers.
|
# ? Dec 27, 2019 00:44 |
|
You could sell it. Getting money is fun.
|
# ? Dec 27, 2019 00:54 |
|
Fabulousity posted:Is there anything fun I can do with a left over Wraith cooler? Is it worth scrap? Only good for a weird paperweight? It's louder than your mom, so keep it around to drown her out.
|
# ? Dec 27, 2019 01:14 |
|
Howard Phillips posted:It's louder than your mom, so keep it around to drown her out. Something something, your sister, something something, GeForce FX 5800 Ultra comparison, something, jet engines and suck. I got nothing. Paperweight it is.
|
# ? Dec 27, 2019 04:44 |
|
Your sister is a lot like VIA motherboards, they both just barely function after a new 4-in-1 every couple months
|
# ? Dec 27, 2019 05:33 |
|
I was bored so I've been messing around with voltage and PBO settings and stuff on my 3700x and I have to say, I'm so confused. How does a -120mV undervolt give me all-core clocks 100+ mhz higher, temps 10-15c lower, but benchmark almost 20% slower in Cinebench? Nothing I have thrown at it has produced any kind of crashing or hash failures or WHEA errors or anything like that, but obviously something is having trouble to lose that much performance. What is going on here? Just tripping some kind of error correction which doesn't seem to get reported anywhere and without any readily apparent effects?
|
# ? Dec 27, 2019 05:47 |
|
You'll lose your mind if you go down the rabbithole of trusting anything you set in shaky AMD partner BIOSes, and relying on false clock/temp readouts. Overdrive also bad.
|
# ? Dec 27, 2019 05:51 |
|
Lowclock posted:I was bored so I've been messing around with voltage and PBO settings and stuff on my 3700x and I have to say, I'm so confused. How does a -120mV undervolt give me all-core clocks 100+ mhz higher, temps 10-15c lower, but benchmark almost 20% slower in Cinebench? Nothing I have thrown at it has produced any kind of crashing or hash failures or WHEA errors or anything like that, but obviously something is having trouble to lose that much performance. What is going on here? Just tripping some kind of error correction which doesn't seem to get reported anywhere and without any readily apparent effects? There's a concept called "clock stretching" where if the power integrity circuitry thinks things are starting to get unstable it'll automatically pull down clocks a little bit to prevent data errors. It does this through an internal clock multiplier thing that you more or less can't externally check for (afaik), the normal main oscillator keeps going and this stretching gets added to it internally. It's kind of like some of those early timer bugs on zen1, effectively you're not getting 0.5% more score, the wall clock is measuring 1.005 seconds worth of work. As you pull your voltage down, internally the cores are starting to panic that the voltage is getting closer to threshold, and starting to stretch the clocks. So you're triggering this stability defense mechanism. It has nominally been in AMD's chips since Steamroller. I've never heard of it being super prominent before Zen2 though. Ideally it seems like it would be nice to give some cpu register that returns the highest clock stretch factor over the last second, or something else to let people monitor and tune for. But overall this is a good thing, it means your chip doesn't crash as abruptly when you are getting close to the OC limit. It's better to try and prevent a few errors by slowing the clocks a bit situationally, than to just let it blow up and hard-crash. This power management circuitry is practically a necessary one on 7nm and below AFAIK, power conditioning and management is now a serious consideration since the transistors operate so close to threshold voltage. SemiEng has written a fair bit about it. https://semiengineering.com/new-power-concerns-at-107nm/ https://semiengineering.com/power-delivery-affecting-performance-at-7nm/ My interpretation: more or less, power management is moving to the core itself, so it doesn't locally brown itself out. Basically a combination of the thermal density making it extremely spiky at a local level as things switch on and off. Transistors switch slower as you get them hotter, and on 7nm your core voltage is so close to the point it doesn't switch the transistor in time, and the core voltage itself varies as different units in the CPU fire and pull the voltage down. So you are trying to aim at a moving threshold with a moving supply voltage, and trying to maximize performance while managing temperatures and power use (which both feed back). And the design requirements on 7nm and below are more or less so tight that you can't just do some "well this should work" setting and have it actually work. And if you speed up one part of the core and pull a bunch more power in that area, you could droop another area past its point of stability. The design requirements are more or less unsolvable for "reasonable" clocks that you would actually want to clock processors at, without some kind of active management. So basically nowadays the core will more or less just have to watch if it's starting to brown out, and do something to bring power under control. Conceptually that's something like stalling some units (eg push them back on the execution unit scoreboard somehow) or stretch clocks. That way you can relax the design requirements to some reasonable level and not have to worry about the prime95 case where someone is torturing the power-heavy execution units. Ideally you would do that 1% or 0.1% of the time, not to the point where things are regularly stalling out. Think of it like Pascal and Vega in the GPU space - they have their own internal power management stuff that sometimes gives non-intuitive behaviors. I have heard of this specific behavior on both Pascal and Vega - the actual scores "get worse" when you push the clocks past a certain point, or if you push the boost controller out of "reasonable" values (eg super low shunt measurements). It's not a coincidence that AMD tried it out with Zen and Vega and NVIDIA tried it out with Pascal, they knew for a long time that it would be problematic on 7nm. These boost systems are waaaayyyy smarter than they ever used to be, because boost is now built into the chip itself. But in turn that also means that in a lot of cases, it's best to try and tune it by twiddling inputs to the existing boost algorithm rather than trying and using one fixed setting. Zen2 is the same as Pascal/Vega - the solution is to undervolt to let the chip have more headroom to boost, then twiddle the clocks to whatever's stable, and OC memory as far as you can. If this whole scheme of voltage manament sounds insane, yes it is. <=7nm computer chips are truly the most complicated thing humanity has ever produced. Making atomic-scale lithography (hundreds of atoms) with blasts of soft xray generated by hitting exotic chemicals with pulse lasers... and then you have to make it boot. At this level of lithography, "digital" logic is intensely analog. Paul MaudDib fucked around with this message at 08:21 on Dec 27, 2019 |
# ? Dec 27, 2019 06:18 |
|
|
# ? Jun 5, 2024 18:38 |
|
As one of my profs used to say, there isn’t analog and digital, there’s just analog and really really fast analog. The design of high-performance caches and other SRAMs are fascinating. I wonder if anyone was foreseeing the tight integration of power management and basically playing the game around a control loop to keep voltage above BOR levels as a significant area of investment. movax fucked around with this message at 21:04 on Dec 27, 2019 |
# ? Dec 27, 2019 21:01 |