|
scrubs season six posted:Watching noted idiot Kyle Bennett destroy a $400 motherboard and rub his greasy jizz covered fingers all over the contacts of a 1950x has already made this the best AMD launch ever. That's the guy that RAID-5'ed 3 RAID controllers running RAID-5, right?
|
#
?
Aug 10, 2017 17:13
|
|
|
|
| # ? Apr 25, 2024 15:57 |
|
|
CommieGIR posted:That's the guy that RAID-5'ed 3 RAID controllers running RAID-5, right? I thought that was Linus, of Linus Tech Tips.
|
|
#
?
Aug 10, 2017 17:13
|
|
|
Twerk from Home posted:I thought that was Linus, of Linus Tech Tips. Yeah, that was him, nevermind.
|
|
#
?
Aug 10, 2017 17:18
|
|
|
https://www.youtube.com/watch?v=G9JR_v-4BaQ rip i9
|
|
#
?
Aug 10, 2017 17:19
|
|
|
gently caress youuuuuuu, NCIX! (Backordered) Should be ~$1250CAD with the exchange you dicks!!
|
|
#
?
Aug 10, 2017 17:32
|
|
|
rip and tear I haven't even ordered anything else yet. Had enough fun with the 1800X ordering multiples of things and keeping the combo that sucked the least.
|
|
#
?
Aug 10, 2017 17:34
|
|
|
The gaming performance was a bit better than I was thinking. I was worried that NUMA would gently caress with things, but it looks like by turning NUMA on, Win 10 is essentially just making games run all on one die and not even touching the others dies cores. The scaling isnt quite linear on the real world encoding tasks (aka not cinebench), but it does edge a healthy margin in front of the 7900X in basically all of those tests. Wouldn't really call it a blow out though considering the cost. AMD is probably pricing this thing just right honestly... enough to snake the build-it-yourself HEDT market for video makers and home rendering, but still with a good margin. Its got premium performance. Really thats the only people who should be even thinking about these platforms, everyone else should be getting 1800X or Coffee Lakes. Can get an 1800X+mobo+16GB of 3200 ram for less than that sampled ASUS X399 board by itself. Also jesus christ that power pull 
|
|
#
?
Aug 10, 2017 17:43
|
|
|
It'll be nice when Anandtech finishes with the benches for the DDR4-3200 vs the stock 2400 numbers.
|
|
#
?
Aug 10, 2017 18:04
|
|
|
I've been putting off a new build for a while, I'm guessing I'll end up with Coffee Lake.
|
|
#
?
Aug 10, 2017 18:09
|
|
|
Cygni posted:
|
|
#
?
Aug 10, 2017 18:11
|
|
|
I was planning on upgrading my ancient i7 920 to a 1700 in a month or two, but lost my job Monday, so that's been put off just a little bit 
|
|
#
?
Aug 10, 2017 18:11
|
|
|
Khorne posted:Honestly, there are 1700s for $265 out there right now. If you oc to 4 it out performs stock 1800X. Yeah, good call. Probably shoulda wrote 1700 there instead of 1800X.
|
|
#
?
Aug 10, 2017 18:15
|
|
|
Cygni posted:Also jesus christ that power pull That's not the half of it (literally). Here's what the total system power draw looks like:  (from Overclock3d) And that's even throwing GPU into the mix to hide Threadripper's consumption a bit. Paul MaudDib fucked around with this message at 18:20 on Aug 10, 2017 |
|
#
?
Aug 10, 2017 18:17
|
|
|
Khorne posted:I think that guy is wrong. Look at the anand benchmarks. With a die disabled 8c/16t would never beat 16c/32t in the exact compute heavy workload that disabling SMT would give an advantage in, and that's exactly what happens. According to AMD's information, game mode turns on legacy mode. When in game mode a 16c/32t will run at 8c/16t on one die. A 12c/24t also becomes 6c/12t. https://community.amd.com/community/gaming/blog/2017/08/10/amd-ryzen-threadripper-for-gaming  AMD posted:Game Mode is a new feature in AMD Ryzen™ Master that reconfigures the platform in two key ways: If Anandtech was only running on 8C/8T than they must have disabled SMT manually. SlayVus fucked around with this message at 18:43 on Aug 10, 2017 |
|
#
?
Aug 10, 2017 18:38
|
|
Paul MaudDib posted:That's not the half of it (literally). Here's what the total system power draw looks like: I just did some quick math and that would put a fully OCed dual Vega/TR system at 1400-1600W power draw. Like I said, that was quick math so take it with a very large grain of salt but holy hell that is a lot of power.
|
|
|
#
?
Aug 10, 2017 18:38
|
|
|
I'm not sure you'd even want to OC a TR system if you're running games. XFR boosts up to 4.2GHz on 4 cores on the fully enabled chip, right?
|
|
#
?
Aug 10, 2017 18:40
|
|
|
AVeryLargeRadish posted:I just did some quick math and that would put a fully OCed dual Vega/TR system at 1400-1600W power draw. Like I said, that was quick math so take it with a very large grain of salt but holy hell that is a lot of power. Who the hell even sells that kind of power supply to handle that? Aren't most manufacturers topping out at 1500W in their lineups?
|
|
#
?
Aug 10, 2017 18:56
|
|
|
Wirth1000 posted:Who the hell even sells that kind of power supply to handle that? Aren't most manufacturers topping out at 1500W in their lineups? Even 1 kW to 1.5 kW units are harder to come by than they used to be. Prices are up like 50-100% within the last year or two. Nobody is running quad-SLI on their overkill high-end gaming rigs anymore, the only people who need that kind of power are running mining ops and don't really care about splitting across two 750W units instead of one 1.5 kW unit. It may even be an advantage in terms of the number of connectors you get.
|
|
#
?
Aug 10, 2017 19:02
|
|
|
Wirth1000 posted:Who the hell even sells that kind of power supply to handle that? Aren't most manufacturers topping out at 1500W in their lineups? Super Flower has a 2000W power supply 
|
|
#
?
Aug 10, 2017 19:02
|
|
Wirth1000 posted:Who the hell even sells that kind of power supply to handle that? Aren't most manufacturers topping out at 1500W in their lineups? There are things like the EVGA Supernova P2 1600W and a few 2kW PSUs, but if one were to build a system like that it would probably be better to use two PSUs.
|
|
|
#
?
Aug 10, 2017 19:03
|
|
|
AMD CPU and Platfrom Discussion: A New Platfrom has Ryzen for Trading Framez for Lanez
|
|
#
?
Aug 10, 2017 19:10
|
|
|
AVeryLargeRadish posted:I just did some quick math and that would put a fully OCed dual Vega/TR system at 1400-1600W power draw. Like I said, that was quick math so take it with a very large grain of salt but holy hell that is a lot of power. I guess the good news is that if a whole bunch of that power consumption is tied up in the platform then multi-socket shouldn't be bad at all, just another 200-300W over that.
|
|
#
?
Aug 10, 2017 19:17
|
|
|
Watts (Lower is Better*) *[Citation needed] *tim allen grunt* EDIT: Also anything over like ~~1700W means you can't just plug it into the wall (in 'murika). SoftNum fucked around with this message at 19:37 on Aug 10, 2017 |
|
#
?
Aug 10, 2017 19:34
|
|
Paul MaudDib posted:I guess the good news is that if a whole bunch of that power consumption is tied up in the platform then multi-socket shouldn't be bad at all, just another 200-300W over that. Well, if you leave things at stock you would be seeing around 200W extra draw for another CPU, but if you OC you will be seeing more like 400W-500W extra per CPU. But I expect that almost no one interested in a multi-socket system would OC and that most interested in multi socket will go for some flavor of EPYC CPU or multi socket EPYC.
|
|
|
#
?
Aug 10, 2017 19:40
|
|
|
well why not posted:
Goddamnit, Linus.
|
|
#
?
Aug 10, 2017 20:19
|
|
|
Gwaihir posted:It'll be nice when Anandtech finishes with the benches for the DDR4-3200 vs the stock 2400 numbers.
|
|
#
?
Aug 10, 2017 20:22
|
|
|
https://www.youtube.com/watch?v=bmRQmr_G3ew I really like the bit where he's recording lossless video while playing a game and rendering another video in the background.
|
|
#
?
Aug 10, 2017 20:23
|
|
|
SwissArmyDruid posted:Goddamnit, Linus. Look at the disparity in viewcount 
|
|
#
?
Aug 10, 2017 20:24
|
|
|
I know, I saw his video where he addressed the goofy thumbnails.
|
|
#
?
Aug 10, 2017 20:25
|
|
|
that extra 13-15W power consumption for running the memory controllers at 3200 instead of 2400 is the weirdest thing. does regular ryzen do that too? has anybody checked? maybe the TR uefi steps up the SoC voltage for higher RAM frequencies?
|
|
#
?
Aug 10, 2017 20:27
|
|
|
G.Skill to the rescue with Threadripper-tuned quad-channel memory kits! http://techreport.com/news/32382/g-skill-flare-x-threadripper-kits-go-up-to-128-gb-or-3600-mt-s Looks like an easy way to get 3600 or 3200 in 4 sticks. edit: All in all, I think it's a good sign that AMD is providing UMA/NUMA and legacy/non-legacy modes. It means they actually loving learned from Bulldozer, in providing a ramp to help compatibility while opening up the potential of their product. Can you imagine what would Bulldozer might have been like if Bulldozer had a mode which bound cores together in pairs to form a pseudo-quad-core? Single-threaded perf might have had enough perf to keep abreast... maybe. SwissArmyDruid fucked around with this message at 20:41 on Aug 10, 2017 |
|
#
?
Aug 10, 2017 20:29
|
|
|
Cygni posted:Also jesus christ that power pull So seriously, the implication I'm reading between these two charts is that the platform is eating literally 200W+ of power. What in the world are they doing with that much power?
|
|
#
?
Aug 10, 2017 20:36
|
|
|
Running two Ryzen CPUs.
|
|
#
?
Aug 10, 2017 20:38
|
|
|
Combat Pretzel posted:Running two Ryzen CPUs. No, according to those sets of numbers the processor itself is only eating 180W. So there's 200W going elsewhere in the system. I mean sure that includes efficiency losses, but that still means the platform itself is eating (380 x 0.8 - 180 = 125W). Doing what? I guess the alternative is the first set of CPU-only numbers are just wrong, after all a Ryzen 7 system is much closer to 110-120W at load. But I guess I never considered the possibility that Ryzen's platform was just terrible and chewing 30W+ of power. It hardly uses anything at idle (Ryzen's idle power is super great) so I never gave the possibility much thought, I guess maybe the platform could be ramping up its consumption heavily under load? But even if Ryzen's platform eats 30W under load, that still means Threadripper's platform is using more than twice as much power per die (which should still be 2 RAM channels per die, etc). It just seems very weird. Paul MaudDib fucked around with this message at 20:50 on Aug 10, 2017 |
|
#
?
Aug 10, 2017 20:40
|
|
|
Threadripper introduces a new system power draw offset 
|
|
#
?
Aug 10, 2017 20:45
|
|
|
200W of RGB LEDs.  seriously though even with extremely inefficient VRMs and a power hungry chipset, I don't see how they'd dissipate that kind of energy on a motherboard with dinky passive heatsinks.
|
|
#
?
Aug 10, 2017 20:46
|
|
|
 That seems a bit implausibly high. Like, "we forgot to subtract the GTX 1080" high, or "SB-E overclocked to 5 GHz at 1.5V" high.e: Actually something is fucky with that graph. It says the stock 7700K is pulling 156W with a GPU, and a stock 7700K is 91W? So that leaves 65W for platform + GPU. Even assuming the platform is only taking 15W, that leaves 50W for the GPU, and what even is that, a GT 730? Kazinsal fucked around with this message at 20:53 on Aug 10, 2017 |
|
#
?
Aug 10, 2017 20:49
|
|
|
i'm guessing the 64 pcie lanes seem to be "active" even if not connected because there's a not insignificant power cost for those in the first place
|
|
#
?
Aug 10, 2017 20:50
|
|
|
from the same review as the one in your pic: look at those max values. 3v vcore is fine i guess. I assume their HWinfo is returning wrong values or the motherboard has a 100% load line calibration.  adoredtv's video shows 156W maximum running prime95 on all cores of a 1950X.
|
|
#
?
Aug 10, 2017 20:51
|
|
|
|
| # ? Apr 25, 2024 15:57 |
|
|
eames posted:from the same review as the one in your pic: How would improper on-chip readouts affect a reading taken at the wall? Although I guess it's possible that Anandtech could be just trusting the package power measurements rather than actually measuring it themselves. If the processor is actually drawing more power than the reported package power that could account for some of the "platform power" as well as their abnormally low power measurements for Ryzen. AMD has been playing games with their power measurements lately and I don't really trust their numbers anymore. Eg they've switched to reporting only package power (rather than board power) with Polaris, and with Ryzen it's been the whole "TDP is just a number for our partners to match thermal solutions and isn't an actual power measurement" song-and-dance (with Ryzen 7 measuring pretty close to X99 in practice despite a 25-50% lower TDP on paper). (thanks for the help all, I just have no idea what's going on between those two measurements) Paul MaudDib fucked around with this message at 21:06 on Aug 10, 2017 |
|
#
?
Aug 10, 2017 20:56
|
|