|
We need to confirm clockspeeds being used, but yeah that does seem to be the case.
|
#
?
Jun 29, 2019 18:50
|
|
|
|
| # ? Apr 24, 2024 02:08 |
|
|
Dr. Fishopolis posted:It's worth delidding. I couldn't get mine over 4.4 without it, but a little dab of gallium and it's been going for 6 years solid at 4.6 with nary a complaint. You're never going to notice a 4.5% clockspeed improvement, especially on a 6 year old chip
|
|
#
?
Jun 29, 2019 18:54
|
|
|
Inept posted:You're never going to notice a 4.5% clockspeed improvement, especially on a 6 year old chip Nah, after spending all the time effort and money, It will feel like 20% in my mind.
|
|
#
?
Jun 29, 2019 18:58
|
|
|
Alpha Mayo posted:Still have a 60hz monitor and am overclocked to 4.5ghz. I don't get any lockups though the most intensive game I play is BLOPS4. But I would like to get a 144hz monitor and my CPU needs to keep up for that to happen, so it's time to upgrade. I had a 3770K at 4.3 for three months with a 144hz monitor. It was fine. I’m not saying it’ll be fine for BLOPS5 or whatever in the future, but anything right now.
|
|
#
?
Jun 29, 2019 19:01
|
|
|
it's so nice to see AMD back in the game aside from that it's a shame we'll probably never be fully aware of the behind the scene bullying that intel has probably never stopped doing, cause that'd be a fun story
|
|
#
?
Jun 29, 2019 19:11
|
|
|
ilkhan posted:We need to confirm clockspeeds being used, but yeah that does seem to be the case.
|
|
#
?
Jun 29, 2019 19:22
|
|
|
shrike82 posted:You don't really need that many cores for most GPU ML use-cases - 2 cores per GPU as a rule of thumb is typically more than sufficient. I have a home ML server (1950X with 4x 1080 Tis) and desktop (1700 w/ RTX Titan) and have never hit a case where the CPU was the bottleneck. This is completely dependent on your algorithm so it's interesting that you'd tell me the thing that often happens in fact doesn't happen.
|
|
#
?
Jun 29, 2019 19:25
|
|
|
How much are the cpubenchmark site’s submissions subject to gaming/faking?
|
|
#
?
Jun 29, 2019 19:31
|
|
|
I think you will find, much like games, it is impossible to cheat *pushes finger against earpiece* oh, well nevermind
|
|
#
?
Jun 29, 2019 19:34
|
|
|
https://twitter.com/0x22h/status/1144952007930683393?s=19
|
|
#
?
Jun 29, 2019 20:22
|
|
|
Bulgakov posted:it's so nice to see AMD back in the game Haven't you seen all the news about AMD helping enemy China. I can't imagine Intel had no influence on these coming out of nowhere.
|
|
#
?
Jun 29, 2019 21:35
|
|
|
Anyone used a radeon vii for ml development? I see rumors it's rtx 2080 ti perf for 2080 price and it has that sweet 16gb of vram No Cuda but rocm and plaidml are fine.
|
|
#
?
Jun 29, 2019 21:44
|
|
|
What's this tweet about cross-die latency? What the hell? There's no such thing with the IO die?
|
|
#
?
Jun 30, 2019 00:45
|
|
|
I’m personally waiting for the new radeons.
|
|
#
?
Jun 30, 2019 00:56
|
|
|
Wow I know they got the inter die latency down but dropping it to just a 2ns penalty above inter CCX latency really is nuts. Seems like they improved everything across the board a whole bunch for IF (latency, power, and bandwidth). Rome might even kick rear end a little more than people are expecting! Combat Pretzel posted:What's this tweet about cross-die latency? What the hell? There's no such thing with the IO die? No there would still be some degree of latency, you can't avoid that its physics at play, its just the impact was generally thought to be small enough to not be a big deal. What this guy is saying is that inter-die (by inter-die I mean from one Zen2 die to the IO die and then to the other Zen2 die, so this matters for the 3900X, 3950X, and Rome but not any of the other Zen2 products that only have 1 Zen2 die and the IO die, we already knew IO die to Zen2 die latency was negligible) latency is so small its going to have a negligible impact on performance at worst and you can pretty much ignore it as a factor on performance period. At least for the desktop chips. It could still potentially effect Rome's performance some in an extreme worst case scenario but its so low that even that isn't looking likely. Looks like TSMC's yields are improving dramatically over short periods of time. TSMC is saying their 7nm already yields better than their 10nm process. Should have near parity in yields to their 16nm process around year end and is already getting close to achieving that. So AMD's Zen2 chiplet yields are probably already better than what was leaked a month or so ago and will probably be as good or better than what Zen/+'s was by year end since Zen2's die is smaller. PC LOAD LETTER fucked around with this message at 05:14 on Jun 30, 2019 |
|
#
?
Jun 30, 2019 03:22
|
|
|
More synthetic benchmark leaks https://twitter.com/TUM_APISAK/status/1145162509399146497?s=19
|
|
#
?
Jun 30, 2019 03:53
|
|
|
I'm trying to piece together my first major upgrade since 2008, and the lack of mATX boards for x470 and x570 is really causing issues with my case choices. Is a B450 mATX board going to be worth it for a 3700X or so? I really don't want to wait another quarter until B550 boards are out NTRabbit fucked around with this message at 04:32 on Jun 30, 2019 |
|
#
?
Jun 30, 2019 04:14
|
|
|
NTRabbit posted:Is a B450 mATX board going to be worth it for a 3700X or so? I really don't want to wait another quarter until B550 boards are out If you're running stock clocks than even the 12C and 16C Zen2's will probably work since they're limited to 105W TDP's. If you want to do some overclocks, even modest ones, than the 8C Zen2 will the limit on any of the mATX B450 mobos. Its not unheard of for people with some of the cheaper B450 mobos and 2700's have been saying they get periodic shut downs due to the VRM overheating with even modest overclocks. The 6C Zen's should run OC fairly well on a decent B450 mobo though. The MSI B450 Gaming Plus seems to be relatively OK compared to the other mATX B450 mobos going by what newegg is listing. I'd still have a fan pointing at the VRM on that thing if you decide to overclock a 8C Zen2 on it though. Really though even the best mATX B450 mobos seem fairly mediocre compared to the decent ATX B450 and X470 mobos that go for $130-150-ish. You sure you can't go for one of those in your budget?
|
|
#
?
Jun 30, 2019 05:06
|
|
|
PC LOAD LETTER posted:If you're running stock clocks than even the 12C and 16C Zen2's will probably work since they're limited to 105W TDP's. It's less about budget, and more about space. Right now I have 2x 24" screens plus the usual up on my desk, and an old tower case under my desk. I've already bought a new 34" ultrawide, and was hoping to move my case up onto the desk for better cable management and easier vacuuming, but I've also been looking at cube cases with a horizontal layout because my desk is right in the corner, the case can only go on that side of the desk, so the 'open' side of a normal tower would face a wall with the rear fans pointing into the corner, and I like LEDs and effective cooling in summer. The difference in size between an ATX cube and an mATX cube is quite a lot, ie the Thermaltake XT is (HxDxW )515x623x394 while the VT is 348x430x350 NTRabbit fucked around with this message at 06:02 on Jun 30, 2019 |
|
#
?
Jun 30, 2019 05:51
|
|
|
If anyone ran OpenFOAM benchmarks on this new chip, that would test + stress the memory system in an important & common type of calculation. I can't afford to upgrade old sandy, but I could help walk an adopter thru a simulation. There isn't much to it, OpenFOAM has cases which are ready-to-calculate in their tutorial for laminar and turbulent flow regime cases. You can especially test core performance scaling, which levels out at around 8-10 real cores today.
|
|
#
?
Jun 30, 2019 06:00
|
|
|
NTRabbit posted:I'm trying to piece together my first major upgrade since 2008, and the lack of mATX boards for x470 and x570 is really causing issues with my case choices. I go by this, and don’t expect it to change much. Get an MSI Mortar if you have to, but get a Tomahawk if you can spare the space.
|
|
#
?
Jun 30, 2019 06:07
|
|
|
NTRabbit posted:It's less about budget, and more about space. Right now I have 2x 24" screens plus the usual up on my desk, and an old tower case under my desk. I've already bought a new 34" ultrawide, and was hoping to move my case up onto the desk for better cable management and easier vacuuming, but I've also been looking at cube cases with a horizontal layout because my desk is right in the corner, the case can only go on that side of the desk, so the 'open' side of a normal tower would face a wall with the rear fans pointing into the corner, and I like LEDs and effective cooling in summer. The difference in size between an ATX cube and an mATX cube is quite a lot, ie the Thermaltake XT is (HxDxW )515x623x394 while the VT is 348x430x350 Could you strap the case to the rear underside of the desk or something? Did some searching and "under desk mounts" are apparently a thing. I may actually get one myself at some point..
|
|
#
?
Jun 30, 2019 06:19
|
|
|
How important is RAM speed with Zen2? I've seen a bunch of posts about 3600mhz being the sweet spot which given my current 1700/64gb ram at 2400mhz makes me wonder how much performance I'm giving up if I were to drop in a new CPU.
|
|
#
?
Jun 30, 2019 06:20
|
|
|
NTRabbit posted:It's less about budget, and more about space. Maybe give the X470 ITX form factor mobos another look then? Actual VRM's are somewhat better (moderate 8C Zen2 overclocks will still be about as good as you could expect and you'll still want to point a fan at the VRM though) and build quality generally seems lots better for some of the ~$150-ish ones. Not sure if they'll work in that case and you'll lose the extra slots of course. shrike82 posted:How important is RAM speed with Zen2? With Zen1/+ CL14 DDR4 3200 could get you around 10% more performance though over DDR4 2600 in some apps. That was fairly expensive RAM that could do that though and so many just ran DDR4 3000 at CL15-16 which was much cheaper and got you like 90% of the performance of CL14 DDR4 3200. We'll have to wait for memory benches. 1usmus said he'd have info out on 7/7 FWIW. PC LOAD LETTER fucked around with this message at 06:25 on Jun 30, 2019 |
|
#
?
Jun 30, 2019 06:21
|
|
|
I think RAM speed will be more important than ever because from the benchmarks I have seen so far there is a big variance in memory latency depending on RAM used, even more so than previous Zen generations. I would definitely wait for launch to see exactly what to get however.
|
|
#
?
Jun 30, 2019 06:39
|
|
|
Pablo Bluth posted:See if you can get OpenFOAM CFD to work and run the benchmark case discussed here. CFD is great for giving memory access a workout. The benchmark template tarball on the post you linked to is inaccessible unless you're logged into the forum 
|
|
#
?
Jun 30, 2019 06:40
|
|
|
shrike82 posted:How important is RAM speed with Zen2? I've seen a bunch of posts about 3600mhz being the sweet spot which given my current 1700/64gb ram at 2400mhz makes me wonder how much performance I'm giving up if I were to drop in a new CPU. Zen 2? Unknown. Zen+? 5% speedup from switching from 2133MHz to 3000MHz. What I can say about Zen 2 is that a lot of people are gonna up their core count. If you are, and you plan to keep those cores busy, more bandwidth is always better.
|
|
#
?
Jun 30, 2019 06:46
|
|
|
Yeah its entirely possible that the 8C or less Zen2 sees little to no benefit past DDR4 3200-3000 while the 12-16C Zen2's could see major benefit from DDR4 3600-3733 like AMD is recommending due to increased bandwidth requirements from the extra cores. Or maybe DDR4 3600-3733 benefits everything across the board by a bunch. We dunno yet.
|
|
#
?
Jun 30, 2019 06:49
|
|
|
mdxi posted:The benchmark template tarball on the post you linked to is inaccessible unless you're logged into the forum You can just install OpenFOAM and run the motorbike tutorial: https://openfoam.org/download/ The benchmark case and script used in that thread is available here: https://openfoamwiki.net/index.php/Benchmarks If you want to do things without the above script: The tutorial is in tutorials/incompressible/simpleFoam/motorbike https://github.com/OpenFOAM/OpenFOAM-dev/tree/master/tutorials/incompressible/simpleFoam/motorBike/ ./Allrun Will create the mesh and run the case with 6 threads by default and reconstruct the results. It will run for 500 iterations. ./Allclean will reset the folder You can change the number of threads in system/decomposeParDict (and decomposition algorithms, I would recommend "scotch" as in the benchmark): https://github.com/OpenFOAM/OpenFOAM-dev/blob/master/tutorials/incompressible/simpleFoam/motorBike/system/decomposeParDict You can change the run length in system/controlDict: (by default 500 iterations, in the benchmark 100): https://github.com/OpenFOAM/OpenFOAM-dev/blob/master/tutorials/incompressible/simpleFoam/motorBike/system/controlDict DoLittle fucked around with this message at 07:42 on Jun 30, 2019 |
|
#
?
Jun 30, 2019 07:15
|
|
|
NTRabbit posted:I'm trying to piece together my first major upgrade since 2008, and the lack of mATX boards for x470 and x570 is really causing issues with my case choices. All of the B450 mATX boards are pretty mediocre. The aforementioned Gaming Plus has decentish VRM, but it's still hugely lacking in the extras department - just one M.2 slot, zero(!) gen 2 USB 3.1 slots, the lowest-tier onboard audio, only two RAM slots. ASRock is actually releasing a mATX x570 board, though, the only one announced so far. We'll have to wait for the technically breakdown, but the VRM components are at least doubled so I'd expect it to be up there with the Gaming Plus, and it has two PCIe 4.0 x4 M.2 slots (most X470 boards only have one 3.0 x4 plus one 3.0 x2 or 2.0 x4), a M.2 key E for wifi, proper top-tier audio, four RAM slots, and 8 rear USB ports (including two gen 2).
|
|
#
?
Jun 30, 2019 07:26
|
|
|
mdxi posted:The benchmark template tarball on the post you linked to is inaccessible unless you're logged into the forum Edit: beaten by Doolittle's edit
|
|
#
?
Jun 30, 2019 07:46
|
|
|
Stickman posted:ASRock is actually releasing a mATX x570 board, though, the only one announced so far. We'll have to wait for the technically breakdown, but the VRM components are at least doubled so I'd expect it to be up there with the Gaming Plus, and it has two PCIe 4.0 x4 M.2 slots (most X470 boards only have one 3.0 x4 plus one 3.0 x2 or 2.0 x4), a M.2 key E for wifi, proper top-tier audio, four RAM slots, and 8 rear USB ports (including two gen 2). I hadn't seen anything about that, bless their crazy little hearts! Now to hope it's good, and available in enough volume to reach the antipodes NTRabbit fucked around with this message at 07:56 on Jun 30, 2019 |
|
#
?
Jun 30, 2019 07:54
|
|
|
Stickman posted:All of the B450 mATX boards are pretty mediocre. The aforementioned Gaming Plus has decentish VRM, but it's still hugely lacking in the extras department - just one M.2 slot, zero(!) gen 2 USB 3.1 slots, the lowest-tier onboard audio, only two RAM slots. Are there any decent small matx cases to put it in?
|
|
#
?
Jun 30, 2019 09:53
|
|
|
The Cerberus owns.
|
|
#
?
Jun 30, 2019 09:57
|
|
|
GRINDCORE MEGGIDO posted:The Cerberus owns. It does but it's way out of budget sadly.
|
|
#
?
Jun 30, 2019 10:07
|
|
|
Lungboy posted:Are there any decent small matx cases to put it in? They're not SFF, but Fractal Design has mATX versions of the Meshify C and Define C, both in glass and solid-panel versions. There's no front USB-C option like the R6, but the mATX X570 Pro doesn't have a type-C front panel connector anyway! NZXT also an mATX case, the H400. They're all smaller than their ATX counterparts, but if you really want small you should probably just go ITX. If you want something a bit shorter, the Corsair 280X is Steve Burke approved and shrinks the height a bit by going with a dual-chamber design. E: The Cooler Master Q300L is decent in the lower-mid price range. Just be aware that the window is acrylic, so it doesn't look quite as nice as glass and can scratch easily. Stickman fucked around with this message at 10:46 on Jun 30, 2019 |
|
#
?
Jun 30, 2019 10:43
|
|
|
Stickman posted:They're all smaller than their ATX counterparts, but if you really want small you should probably just go ITX. This is my intention but that matx board looks very nice. If there are decent itx boards then I'll go that route. The ASRock board's main selling point seems to be thunderbolt which I have no use for, and the Impact has that M2 riser board which seems a bit over engineered. Thanks for all those case suggestions, it seems that matx cases are quite large and something like a Thetis with a full ATX board might be smaller. Lungboy fucked around with this message at 10:55 on Jun 30, 2019 |
|
#
?
Jun 30, 2019 10:50
|
|
|
I like itx, I never use more then 2 channel memory, one GPU, and built in WiFi anyway. Plus the boards hold value quite well.
|
|
#
?
Jun 30, 2019 10:53
|
|
|
Weird, I thought matx would be the most popular. I definitely can't fit an atx case l, so maybe I'll look into itx too...
|
|
#
?
Jun 30, 2019 13:02
|
|
|
|
| # ? Apr 24, 2024 02:08 |
|
|
For DIY'ers mATX has kind've fallen out of popularity because people either decide to go big (ATX) or doubled down on going small (ITX) so there are less options there than there used to be.
|
|
#
?
Jun 30, 2019 13:06
|
|