|
i've always found asus overpriced and i'm not a fan of their design aesthetic the various board makers, mobo and GPU, are interchangeable to me these days - i've not noticed any major differences and i've (not deliberately) ended up switching makers almost very generation with a gigabyte mobo for my threadripper currently and a PNY A6000
|
#
?
Apr 9, 2022 01:09
|
|
|
|
| # ? Jun 1, 2024 07:37 |
|
|
Dr. Video Games 0031 posted:What are you basing this on? That seems like a big assumption. I was talkin about retail DIY sales, and basing it off of the statements from Steve on GN, Jay, and others during the 3090/New World explosion drama that EVGA had sold significantly more 30-series cards than their competitors. It is also backed up by what I've seen at Microcenter, but i put that "maybe?" in there for a reason cause I'm not really basing it on anything concrete I've seen. 
|
|
#
?
Apr 9, 2022 01:35
|
|
|
That was about 3090s though and I wouldn’t be surprised if evga did sell more 3090s. 3090s are probably just a drop in the bucket though compared to say 3060s. Remember the 1060 is still the most common card in the steam survey.
|
|
#
?
Apr 9, 2022 01:37
|
|
|
hobbesmaster posted:That was about 3090s though and I wouldn’t be surprised if evga did sell more 3090s. 3090s are probably just a drop in the bucket though compared to say 3060s. Remember the 1060 is still the most common card in the steam survey. The quote i remember was 30-series in general, and evga sold a lot of 1060s in the US too (they were regularly the cheapest 1060 6gb, along with the Zotac ITX variant). Like I said, im not basing it on much, but evga does sell a lot in the US. Like Paul mentioned, worldwide is a different story. At least by the last data i saw in digitimes, Palit was way ahead.
|
|
#
?
Apr 9, 2022 01:41
|
|
|
Seems like Threadripper is low in supply. https://www.theregister.com/2022/04/08/amd_threadripper_supply/ At a guess; such are Epyc orders, nothing is left over for Threadripper.
|
|
#
?
Apr 9, 2022 18:21
|
|
|
quote:That's because supply for Threadripper 3000 and Threadripper Pro 3000 processors has been severely low over the past several months, according to executives at six PC builders and one IT distributor in the US who spoke to The Register. I would think that AMD would want to be allocating 0 wafers to zen 2 chiplets at this point? The APUs are monolithic, right?
|
|
#
?
Apr 9, 2022 19:07
|
|
|
hobbesmaster posted:I would think that AMD would want to be allocating 0 wafers to zen 2 chiplets at this point? The APUs are monolithic, right? Ryzen 3000s were still on shelves to fill in the price holes that the 5000s were avoiding until just recently, so they must have been producing Zen 2 chips until at least the start of this year. But maybe this is the sign that zen2 is shut down, and threadrippers dried up first because the TR-5000s aren't available. I dunno though, when threadrippers were selling for $1k to $2k less than an Epyc chip with the same CCDs, and AMD has zero problem selling as many Epyc chips as they can make, I think there's a basic incentive problem. Threadripper started out as a cool bargain HEDT that was a bit janky, and while they've raised the price each gen this may be the sign that it is moving to serious-business only prices.
|
|
#
?
Apr 9, 2022 23:13
|
|
|
Really I’d think the 5950 covers whatever the threadripper market was decently. At least compared to Intel.
|
|
#
?
Apr 9, 2022 23:18
|
|
|
Klyith posted:Ryzen 3000s were still on shelves to fill in the price holes that the 5000s were avoiding until just recently, so they must have been producing Zen 2 chips until at least the start of this year. But maybe this is the sign that zen2 is shut down, and threadrippers dried up first because the TR-5000s aren't available. It's been hard to find a 3600X at a reasonable price for months (while the 5600X price was gradually falling), so I think supply has been running low for a while now.
|
|
#
?
Apr 9, 2022 23:26
|
|
|
hobbesmaster posted:Really I’d think the 5950 covers whatever the threadripper market was decently. At least compared to Intel. Memory bandwidth, meh. PCIe lanes, no.
|
|
#
?
Apr 10, 2022 00:03
|
|
|
just got a price drop alert i apparently still had for the 5900x at $399 if anyone's looking to do their last am4 upgrade other cpu's like the 5950x also got a drop gonna be fun to see the 5800X3D benchmark the 20th
|
|
#
?
Apr 11, 2022 14:50
|
|
|
That Peruvian site that got one early has it 20% ahead of the 12900KS in Shadow of the Tomb Raider, so gaming performance seems to be good at least. Gonna be a bit of a weird one though, considering it's basically a 5700x you can't overclock for tasks that don't take advantage of the extra cache.
|
|
#
?
Apr 11, 2022 15:08
|
|
|
i wish more specific optimizations could have been made for the cache size if amd hadn't already announced that it would be the last of its kind for the time being, but at least there's an actual cpu to program towards in case we end up back here again
|
|
#
?
Apr 11, 2022 15:30
|
|
|
Helter Skelter posted:Gonna be a bit of a weird one though, considering it's basically a 5700x you can't overclock for tasks that don't take advantage of the extra cache. The average PC enthusiast doesn't do any tasks besides games that really need massive CPU performance. kliras posted:i wish more specific optimizations could have been made for the cache size if amd hadn't already announced that it would be the last of its kind for the time being, but at least there's an actual cpu to program towards in case we end up back here again You can't really 'optimize' for a massive cpu cache. Either your task has datasets that take advantage of it, or it doesn't. And I'm sure there will be a Zen 4 with the 3d cache eventually. The Epyc server chips with v-cache are a massive success for HPC. Even if the 5800X3D is a total flop, they'd continue making v-cache stuff for the server lines. But it's probably going to work the same as Zen 3: 1. launch with standard consumer Ryzen, to check for any problems and get the fab process perfected 2. standard server Epyc chips 3. stack v-cache on the new chiplets after all other parts of the process are working 100%
|
|
#
?
Apr 11, 2022 15:54
|
|
|
Klyith posted:The average PC enthusiast doesn't do any tasks besides games that really need massive CPU performance. 3Dmark is on steam so it counts as a game right. As for a serious question, I’ve noticed that the higher end DDR4 ram is coming down in price so there’s less ridiculously priced DDR4-4400 out there. Do these higher speed bins guarantee that they can run at 3800 with tighter timings or is that something completely separate that would have to be binned for specifically?
|
|
#
?
Apr 11, 2022 16:05
|
|
|
hobbesmaster posted:As for a serious question, I’ve noticed that the higher end DDR4 ram is coming down in price so there’s less ridiculously priced DDR4-4400 out there. Do these higher speed bins guarantee that they can run at 3800 with tighter timings or is that something completely separate that would have to be binned for specifically? Been running 4x8GB of these Patriot ViperSteel 4400C19 at 3800 CL14 (14-16-13-21-34) for a couple years now. The price was $126 for 2x8GB when I picked them up, they're $109 now so they seem to have held their value. They're rated CL19 at 4400 and can run at 1.5V, both of which probably contribute to their ability to run 3800 CL14. Picked them up after seeing this buildzoid video: https://www.youtube.com/watch?v=j3z7Ggh2_xI Here're my current OC values, though I had to bump tRCDRD to 16T for the 5950x.
|
|
#
?
Apr 11, 2022 16:15
|
|
|
Klyith posted:The average PC enthusiast doesn't do any tasks besides games that really need massive CPU performance. Is it really gonna mean much if (for example) you're GPU-limited, though? There's a lot to look at and I'm intensely curious to see how it all shakes out.
|
|
#
?
Apr 11, 2022 16:24
|
|
|
We got a new waste of sand vid from GN hoorayy https://www.youtube.com/watch?v=PsdeJszdV7I Cygni fucked around with this message at 20:45 on Apr 11, 2022 |
|
#
?
Apr 11, 2022 17:27
|
|
|
hobbesmaster posted:As for a serious question, I’ve noticed that the higher end DDR4 ram is coming down in price so there’s less ridiculously priced DDR4-4400 out there. Do these higher speed bins guarantee that they can run at 3800 with tighter timings or is that something completely separate that would have to be binned for specifically? Yes, in general ram works pretty much proportionately: running 4400 at 3800 means you can reduce all the timings to ~86% of their rated numbers. Downsides: 1. Rounding is a problem 2. You're in manual memory tweaking land now, hope you enjoy spending a weekend twiddling numbers & running memtests Helter Skelter posted:Is it really gonna mean much if (for example) you're GPU-limited, though? There's a lot to look at and I'm intensely curious to see how it all shakes out. So even in GPU-limited scenarios where max FPS graph is nearly flat across the board, you can have differences in the 1% low FPS from better CPUs. So that's a reason that people may put forward to have the best CPU However, most games will run well above 60 FPS on the 1% low when you test for non-GPU constrained performance. A recent CPU, even midrange, you have to look around for a while to find a game where the CPU-influenced 1% lows can possibly be described as 'bad'. So, IMO, tldr, no. The CPU isn't really a big deal for gaming. If you are playing SP games, the inexpensive $200-300 CPUs will keep everything running nicely. Get a VFR display to reduce the impact of slow frames. If you are a competitive MP gamer that cares about FPS you're playing games that have minimal CPU needs. CSGO's 1% lows are probably still higher than your display's refresh.
|
|
#
?
Apr 11, 2022 17:28
|
|
|
Got some interesting datapoints recently, if anyone is curious about how desktop Ryzens behave when put on a very low power budget. My compute farm is now half-and-half 3950s and 5950s. The weather is also starting to warm up, and since last October(ish) those machines have been running in my garage, which has neither AC nor ventilation. To deal with this, I was pondering either (1) putting a cron job on them that would cycle their load on/off so that they only crunched at night, or (2) dramatically power-limiting them and seeing how the freq/thermals settled down. I decided to try the latter first, and set the PPT of all machines to 65W. I don't mean "Eco Mode (65W)", which is actually 88W. I mean PBO set to On, with the PPT limit set to 65W. Assuming that the IOD eats 11W, that leaves 3.375W per compute core. What I saw was that during all parts of the day, CPU temps were beween 58C and 63C. At night, freqs on active cores* were between 2400MHz and 3200MHz. When it warmed up, active cores would clock down to as little as 1700MHz. This pattern and the temps have been stable for over a week. It will be interesting to see what happens when things get really hot outside, but we're a ways off from that. Yesterday, however, in preparation for switching my CPUs from one project to another, I had them set to just complete their existing work and not fetch anything new. Last night they were all down to their last handful of workunits, and at that point things started to get interesting. Roughly around the point where half the cores** were idled, temps started to go up, along with clocks on the remaining active cores. Eventually, a machine would be down to 31 idle threads, and one thread controlling the work running on the GPU. That one core would be running at 4400 or 4500MHz, and the Tctl temp would be reporting at 75 to 80C. I'm not sure if this effect is due to a single core not sufficiently ramping up the overall package temp enough to cause the fans to shift into higher gear, or if it's because a Ryzen core running that hard simply generates that much heat even if the fans are going full tilt. I'm more inclined to believe the latter, but my machines are so rarely in this state that I've never given it much thought or investigated. They'll be back to work soon anyway. * My machines are normally run fully loaded with processes which are integer-heavy, so I'm not used to having idle cores reported at all. So over the past week I learned that X950 CPUs all report a clockspeed of 2200MHz for idle cores. This seems rather wasteful if that's an actual 2.2GHz of idle-wait -- but I believe that all modern CPUs are smarter than that, and that's just the clock frequency being sent into a mostly powered-off core. ** As mentioned in the previous note, I'm normally running workloads which are not FPU-bound, but for the past couple months I've been running Einstein@Home, which certainly seems to be. It runs stressy and hot (hallmarks of AVX usage), so I configured it to only run 15 workunits per CPU -- one per physical core/FPU, minus one for the GPU control process. So in this case, "half of cores idled" means 7 or 8 tasks in flight. mdxi fucked around with this message at 18:44 on Apr 11, 2022 |
|
#
?
Apr 11, 2022 18:28
|
|
|
mdxi posted:Last night they were all down to their last handful of workunits, and at that point things started to get interesting. Expected behavior. When only some cores are loaded, the CPU will put more watts into them. This leads to higher local temperatures on those cores. Running 30 watts on a single core it gets very hard to move that heat out of such a tiny area. Your cooler really doesn't matter as much anymore, the limitation is conduction in the CPU package itself. It's much easier to cool 80W across 8 cores. So the maximum temperatures you will see from tctl -- which has many temperature sensors across the CPU and reports the highest number -- is normally with ~2 threads of load. mdxi posted:* My machines are normally run fully loaded with processes which are integer-heavy, so I'm not used to having idle cores reported at all. So over the past week I learned that X950 CPUs all report a clockspeed of 2200MHz for idle cores. This seems rather wasteful if that's an actual 2.2GHz of idle-wait -- but I believe that all modern CPUs are smarter than that, and that's just the clock frequency being sent into a mostly powered-off core. When a core is sleeping and you ask what frequency it's running at, it reports the last active speed before it turned off. Whatever software you are using to measure CPU speed needs to ask both what clockspeed it's running at and how much time it spends sleeping to know the true effective clockspeed.
|
|
#
?
Apr 11, 2022 19:07
|
|
|
Cygni posted:We got a new waste of sand vid from GN hoorayy It’s recovering zen 2 APUs with errors in the video part, I’m not sure “waste of sand” applies when the alternative is to throw it away as unsellable. Klyith posted:
This was “fun” when I had a newborn that may only be asleep for 5 minutes before waking back up. All I found out was I had what must be the worst 3600x they allowed through QC. hobbesmaster fucked around with this message at 20:25 on Apr 11, 2022 |
|
#
?
Apr 11, 2022 20:21
|
|
|
hobbesmaster posted:It’s recovering zen 2 APUs with errors in the video part, I’m not sure “waste of sand” applies when the alternative is to throw it away as unsellable. Pricing matters. If they launched it at $60 against the Alder Lake Celerons and Pentiums, it would be fine. At its MSRP though, the extra energy, materials, and money they spend packaging and boxing the borked dies (which is more expensive than making the die itself) is pretty much a waste. They are completely uncompetitive and are destined to end up as ewaste rapidly. Worse yet is that in actual street prices, you can get a 12400 (with working IGP) for $98.** (edit, shady Newegg seller, likely not a real price anymore if ever) Cygni fucked around with this message at 22:43 on Apr 11, 2022 |
|
#
?
Apr 11, 2022 20:47
|
|
|
Price is the easiest property if a product to change. A “waste of sand” or e-waste would be a product that is just nonfunctional. This thing works it’s just at a bad price. That said even pretending to charge triple digits for zen+ performance in 2022 seems comical hobbesmaster fucked around with this message at 21:00 on Apr 11, 2022 |
|
#
?
Apr 11, 2022 20:58
|
|
|
How long did AMD sell the 5350? I miss being able to walk into Microcenter and they'd have that $40 CPU+MB combo.
|
|
#
?
Apr 11, 2022 20:59
|
|
|
kliras posted:just got a price drop alert i apparently still had for the 5900x at $399 if anyone's looking to do their last am4 upgrade Thanks for this, getting a price match on the 5900x.
|
|
#
?
Apr 11, 2022 21:13
|
|
|
Cygni posted:Worse yet is that in actual street prices, you can get a 12400 (with working IGP) for $98. Where? I might buy in at that price.
|
|
#
?
Apr 11, 2022 21:27
|
|
|
Cygni posted:Pricing matters. If they launched it at $60 against the Alder Lake Celerons and Pentiums, it would be fine. At its MSRP though, the extra energy, materials, and money they spend packaging and boxing the borked dies (which is more expensive than making the die itself) is pretty much a waste. They are completely uncompetitive and are destined to end up as ewaste rapidly. That can't be right that's literally 50% off a brand new part that was already priced competitively.
|
|
#
?
Apr 11, 2022 22:18
|
|
|
VorpalFish posted:That can't be right that's literally 50% off a brand new part that was already priced competitively. https://www.newegg.com/intel-core-i5-12400f-core-i5-12th-gen/p/N82E16819118358 There was a few other sellers at that price a few hours ago when i wrote that, but now its just one with zero ratings so uhhhhhh maybe dont buy that
|
|
#
?
Apr 11, 2022 22:40
|
|
|
Cygni posted:https://www.newegg.com/intel-core-i5-12400f-core-i5-12th-gen/p/N82E16819118358 There are other intel CPUs from random chinese sellers. all seemed to appear at the same time with unrealistically low prices, from different accounts with zero feedback. That's extremely suspect. I would not call this "actual street prices."
|
|
#
?
Apr 11, 2022 23:30
|
|
|
Dr. Video Games 0031 posted:There are other intel CPUs from random chinese sellers. all seemed to appear at the same time with unrealistically low prices, from different accounts with zero feedback. That's extremely suspect. I would not call this "actual street prices." Agreed, i edited in a disclaimer. I mean really it doesn't matter, the 4500 isnt competitive at Intel's MSRPs anyway.
|
|
#
?
Apr 11, 2022 23:34
|
|
|
https://twitter.com/VideoCardz/status/1513768891071307778 general parity with intel except when there are huge performance uplifts:   fun times
|
|
#
?
Apr 12, 2022 14:37
|
|
|
Holy balls look at that Witcher result Why does cache matter in games so much? You just keep larger chunks of the game loaded in without needing to go back to RAM?
|
|
#
?
Apr 12, 2022 14:43
|
|
|
gradenko_2000 posted:Holy balls look at that Witcher result Yes, cache misses will lead to fps drops or, in the worst case, hitching and poor frame consistency.
|
|
#
?
Apr 12, 2022 14:48
|
|
|
i really wanna see what a horribly programmed game like elden ring performs like with the 5800X3D
|
|
#
?
Apr 12, 2022 14:51
|
|
|
I want to see a 1440-2160P RTX results too please thank you.
|
|
#
?
Apr 12, 2022 15:04
|
|
|
kliras posted:i really wanna see what a horribly programmed game like elden ring performs like with the 5800X3D Makes no difference, the game's technical issues are from something else, not lack of CPU grunt/cache It runs fine on consoles/steam deck
|
|
#
?
Apr 12, 2022 15:36
|
|
|
TechPowerUp posted an early review of the X3D: https://www.techpowerup.com/review/amd-ryzen-7-5800x3d/  Pretty much parity with the 12700k and 12900k and splits their Microcenter pricing with its MSRP. Not sure if I would recommend it over Alder Lake for an all-new build, but a very nice final upgrade path for someone who bought a Zen+ or Zen2 many moons ago.
|
|
#
?
Apr 12, 2022 20:15
|
|
|
I'm waiting for the HWUB review. Want to see if it's enough to catch up to Intel in Factorio.
|
|
#
?
Apr 12, 2022 21:07
|
|
|
|
| # ? Jun 1, 2024 07:37 |
|
|
Cygni posted:TechPowerUp posted an early review of the X3D: e: also worth noting that their performance/$ chart doesn't take into account the 5900x and 5950x discounts i mentioned earlier kliras fucked around with this message at 21:38 on Apr 12, 2022 |
|
#
?
Apr 12, 2022 21:35
|
|