|
Zotix posted:But an x470 board does need a current ryzen processor to flash it right? Most boards do but there a few that can do usb flashing without a cpu. Some of the ASUS crosshair and a bunch of MSI boards have this feature. https://www.reddit.com/r/Amd/comments/bvfo57/list_of_b350_b450_x370_and_x470_motherboards_with/
|
#
?
Jun 25, 2019 23:08
|
|
|
|
| # ? Apr 25, 2024 22:30 |
|
|
It's X470, the high-end chipset. I would expect there to be more of those motherboards on the market that have BIOS flashback functionality than not, if not all of them.
|
|
#
?
Jun 26, 2019 01:38
|
|
|
Newegg was also very good when Raven Ridge came out about stick “Ryzen 2000 compatible!” on product pages for the launch duration. They are essentially promising you a pre-flashed board, so it’s their fault if you don’t get one.
|
|
#
?
Jun 26, 2019 01:50
|
|
|
SwissArmyDruid posted:It's X470, the high-end chipset. I would expect there to be more of those motherboards on the market that have BIOS flashback functionality than not, if not all of them. That's what I was thinking, but if you look at the list B-Mac posted it's pretty much - MSi top and mid-range B450 boards + their top-end X470 (the M7) - ASUS top-end X470 (Crosshair VII Hero) ... and that's it. No ASRock, no Gigabyte, only one board from ASUS, and only one of MSi's X470 boards. MSi really did go all-out on their B450s this cycle, though! E: I think I misread your post and you were actually saying the same thing... whoops! It really looks like decent support is pretty much just MSi on their newer B450 motherboards. Maybe it'll be more standard going forward - I think I remember seeing that Gigabyte's X570 board were supposed to support USB flashing.
|
|
#
?
Jun 26, 2019 02:16
|
|
|
Cygni posted:Chapuzas has now tested the 3600 with an x570 board and the performance is identical. Unless you are into extreme future proofing or going for the 16 core which might use those big VRMs, x470 is lookin pretty tasty. They haven't tested faster RAM speeds (3600 or 3733) to see if they work with the older motherboards right?
|
|
#
?
Jun 26, 2019 02:25
|
|
|
Still haven't decided what motherboard I want. Taichi having wifi + t-topology + no second nic throws a wrench in my gears, because I was just planning on running 2x16. It's also an awkward time to buy high clocking memory because new SKUs are coming out and 32gb sticks are coming out. Really disappointed with MSI and Gigabyte only offering 6 SATA ports on even their mid-high end stuff. x570 supports 4 or 8 (or 12... but no one did 12 as far as I know), so 6 is just being lazy and cheap. Kind of hoping ASRock drops something in the <$300 (preferably <$200) range that I want. The creator seems interesting but I have to imagine its price is high. Paul MaudDib posted:The memory latency is high/AIDA64 memory results are hosed up, but that actually showed up in another 3600 review today too so it may be a beta BIOS problem or something. From what I can tell from geekbench, it looks like 75ns-80ns is what 2933-3200 C16 gets. 65ns-67ns is 3200 C14 and similar (3600/3733 at higher cas too). Khorne fucked around with this message at 02:44 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 02:36
|
|
|
Khorne posted:From what I can tell from geekbench, it looks like 75ns-80ns is what 2933-3200 C16 gets. 65ns-67ns is 3200 C14 and similar (3600/3733 at higher cas too). That makes sense for the latency, but there's still no good reason that memory throughput should be as low as the Athlon 200ge running 2666 DDR4. Unless it's actually running at 2666 and they didn't notice, which seems pretty unlikely.
|
|
#
?
Jun 26, 2019 02:46
|
|
|
Someone put the 3950X under LN2 and ran Geekbench at 5.2GHz all core https://twitter.com/TUM_APISAK/status/1143704396103950336
|
|
#
?
Jun 26, 2019 03:40
|
|
|
Arghhh, I want the testing of 3900X!
|
|
#
?
Jun 26, 2019 04:39
|
|
|
There's like a zillion 3600 geekbench tests now and this one from today is the highest yet, pretty impressive for 4.3GHz https://browser.geekbench.com/v4/cpu/13667643
|
|
#
?
Jun 26, 2019 04:48
|
|
|
So that should put single score score for 4.5ghz around 6000 on the dot? Perfect.
|
|
#
?
Jun 26, 2019 05:07
|
|
|
Those of you looking at a 3800x or 3900x what are you going to cool it with? I'm doing my first build in probably ~7 years, and from what I read the Noctua basically beats everything aside from custom loops. However, it has the issue of weight on the mobo, and the space it takes up. Are any of you planning on using an off the shelf AIO? and if so which one?
|
|
#
?
Jun 26, 2019 05:35
|
|
|
I don't know about beats everything, the generally accepted wisdom that the D15 is a bit better than 240mm AIOs but still falls behind 280mm and 360mm ones in pure throughput. The Noctua has the advantage of having one or two really good fans and better price to performance. Llamadeus fucked around with this message at 06:05 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 06:00
|
|
|
MaxxBot posted:Someone put the 3950X under LN2 and ran Geekbench at 5.2GHz all core MaxxBot posted:There's like a zillion 3600 geekbench tests now and this one from today is the highest yet, pretty impressive for 4.3GHz My preliminary take after seeing all those linked benchmarks is that IPC is good but performance/frequency scaling and max frequency isn't so great. A 5 GHz Coffee Lake single core is 100 points faster than the 5.2 GHz LN2 score so the 3600X would have to run at ~5.3 GHz to match that unless higher core counts somehow diminish single core performance. Here's the above stock 3600 compared to my fully patched 5GHz 8700k daily system. Note that overclocking Zen 2 probably won't be very useful for regular daily use because PBO/XFR has likely improved further. I assume that's because a lot of the performance comes from the large cache and that performance doesn't scale well with frequency. Of course geekbench is just one synthetic and it's entirely possible that the large cache is a game changer for frametimes in video games, BIOS versions aren't final, RAM/IF dividers are a big question mark, etc. vvv You're right about single core performance but it could well be that higher RAM frequencies (= IF dividers) are required with the high core count CPUs because otherwise you'll starve the cores for bandwidth. vvv eames fucked around with this message at 06:39 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 06:18
|
|
|
Keep in mind that the 3950X score is likely held back some because of the infinity fabric running at 1/2 speed.
|
|
#
?
Jun 26, 2019 06:30
|
|
|
Llamadeus posted:I don't know about beats everything, the generally accepted wisdom that the D15 is a bit better than 240mm AIOs but still falls behind 280mm and 360mm ones in pure throughput. I know this is generally accepted as true, but I really do wonder sometimes. Most benchmarks I've seen that compare the two are a few years old and use Intel 4-cores, and I don't think I've seen one done with a soldered or delidded-with-liquid-metal CPU, so I strongly suspect there's a lot of bottlenecking on the TIM. Even in the tests I've seen where they use a beefier CPU they usually use relatively low heat loads (by today's standards) for whatever reason, like 120W or less, and as Buildzoid pointed out recently, a high heat load is very important in order to actually see the differences between different cooling solutions. Even GamersNexus has been guilty of doing this in their AIO tests - while they are using a 6-core (the i7-5930K), they're using P95's large FFT's and so I suspect they aren't pulling much over 120W. Now, to be fair, I'm pretty sure they use that particular CPU and configuration in order for them to stay comparable with a large number of older cooling benchmarks on the same test bench, but still. Failing to control for various confounding factors is also quite common by everyone who isn't GamersNexus - involving case fans in the comparison is quite common, for example (if you put the AIO at the front of the case where it can breathe fresh air, you should theoretically want to arrange so that the NH-D15 gets the same airflow where it's sitting way over at the CPU socket, but that's not at all practical, so just use an open air test bench like GamersNexus does). The best test I know of involving a Noctua and an AIO is the GamersNexus one on a first gen Threadripper, where they pushed 250-300W through a soldered CPU with a huge die area and a huge heatspreader (eliminating quite a bit of thermal transfer bottlenecking), but in that one they were using the TR4 variant of the NH-U14S, which isn't quite as enormous as the NH-D15 and which only has one fan. In that test the 240mm AIO won by about 7°C in the 250W test and 9° in the 310W one (illustrating Buildzoid's point about heat load being important for seeing the differences), but again, I really wonder how transferable those results are to other setups. Anecdotally, on a sample size of two delidded 8700K's pulling around 180W, my own under a NH-D15 gets about the same or perhaps a few degrees lower temperatures than my friend's under a Kraken X62 (one of the better 280mm AIO's). I've also seen people with delidded 8700K's and custom loops who brag about 60°C in some mild 120-140W benchmark where I get barely over 60°C myself (for some reason these people - who invariably daily a 5.2 or 5.3GHz overclock - never want to try an actually demanding benchmark, most likely because if they did it'd turn out that they're not actually stable). At least for me, in a ~200W or less scenario the few degrees of difference between a 280/360mm AIO and a NH-D15 isn't enough to motivate spending at least twice as much money on cooling. TheFluff fucked around with this message at 13:06 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 12:58
|
|
|
What I remember about AIOs is that they start off stronger than an air cooler but after all of the limited water supply warms up a bit they can’t keep up, and after a few hours of testing become less effective
|
|
#
?
Jun 26, 2019 13:05
|
|
|
The other thing to keep in mind with that gamers nexus review is the noise normalized results. The 240mm AIO lead is cut to 4 degrees at 40 dB vs the 140mm air cooler. You’re right though, there aren’t a ton of sites that out a large heat load into these coolers, most of the data has been with quad core CPUs from years back. Hopefully we will get some good tests once the 12 and 16 cores are released.
|
|
#
?
Jun 26, 2019 13:09
|
|
|
fishmech posted:I literally just said it, Microsoft states that if you're selling or using 8 or 10 on a computer, it is supposed to support a minimum horizontal resolution of 1024 and a minimum vertical of 768 to fit the UI elements, particularly since the former "metro"/now modern or whatever apps are meant to work with a 1024x768 minimum res. Gonna need a citation here, because Microsoft's official system requirements lists 800x600.
|
|
#
?
Jun 26, 2019 13:10
|
|
|
Lube banjo posted:What I remember about AIOs is that they start off stronger than an air cooler but after all of the limited water supply warms up a bit they can’t keep up, and after a few hours of testing become less effective Well yeah, heat soak in the liquid is absolutely a thing, but even the relatively unscientific benchmarkers know about that by now and let the water temperature stabilize before they take measurements. Heatpipe based coolers also have limitations of course. They tend to have a hard limit on how much heat they can handle, because at some point you'll be boiling the liquid in the heatpipes faster than it can condense and trickle back down, and once that happens the cooler basically stops working all of a sudden. For a huge 6-pipe thing like the NH-D15/D14 though that limit is somewhere north of 300W and you're not going to hit it in practice.
|
|
#
?
Jun 26, 2019 13:10
|
|
|
Zotix posted:Those of you looking at a 3800x or 3900x what are you going to cool it with? I'm doing my first build in probably ~7 years, and from what I read the Noctua basically beats everything aside from custom loops. However, it has the issue of weight on the mobo, and the space it takes up. Are any of you planning on using an off the shelf AIO? and if so which one? Depending on overclock results or a bunch of other stuff I'll maybe consider large AIOs. Weight isn't very relevant in my opinion. The downside to the D15 is the price, it's double the price of coolers with similar performance. Although I don't know the max performance of the mugen 5 rev b vs a d15, it's certainly capable of cooling an overclocked 9900k within 0c-3c of a D15. 3800x should be cooled by anything reasonable. 3900x and 3950x will come down to max voltages and actual oc headroom. It's hard to say. The $45+ air coolers and 240mm+ AIOs should do more than fine given the cooler they ship with, but it's unclear if you can get slightly more performance with a better setup or not. Zen+ hit a hard wall, and it seems like zen2 does as well. That means provided you keep your CPU under 80c or 85c or 90c or whatever is reasonable for zen2, additional cooling performance doesn't matter (unless VRM temps are getting too high or GPU temps are getting too high and throttling, but that's an overall case airflow thing). At least, that's my personal practical take on everyday OCs/cooling. Measly Twerp posted:Keep in mind that the 3950X score is likely held back some because of the infinity fabric running at 1/2 speed. Khorne fucked around with this message at 15:21 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 14:43
|
|
|
TheFluff posted:I know this is generally accepted as true, but I really do wonder sometimes. Most benchmarks I've seen that compare the two are a few years old and use Intel 4-cores, and I don't think I've seen one done with a soldered or delidded-with-liquid-metal CPU, so I strongly suspect there's a lot of bottlenecking on the TIM. Even in the tests I've seen where they use a beefier CPU they usually use relatively low heat loads (by today's standards) for whatever reason, like 120W or less, and as Buildzoid pointed out recently, a high heat load is very important in order to actually see the differences between different cooling solutions. Basically shows 240mm, the D15 and the 280s as being quite close, and I'd agree that nobody with six cores or fewer (or isn't too concerned about that last 100-200 MHz) should worry about needing more than D15-class cooling. Llamadeus fucked around with this message at 14:50 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 14:46
|
|
|
TheFluff posted:Well yeah, heat soak in the liquid is absolutely a thing, but even the relatively unscientific benchmarkers know about that by now and let the water temperature stabilize before they take measurements. It depends if the Ryzen-based 16C can get anywhere near current Threadripper power draw when you start overclocking them, Threadripper's already up to 250W TDP so 300W isn't exactly far.
|
|
#
?
Jun 26, 2019 14:46
|
|
|
Internal article about AMD for Intel employees 
|
|
#
?
Jun 26, 2019 16:02
|
|
|
hmm...
|
|
#
?
Jun 26, 2019 16:42
|
|
|
AKA until they get 10nm out in volume Intel is in bad shape. The same story they've had for the last umpteen years of 14nm.
|
|
#
?
Jun 26, 2019 16:44
|
|
|
the question is are they still planning to get 10nm out in volume, except in some mobile devices?  I think their 14nm CPUs will stay reasonably competitive in the 6-10C market, it's up to AMD's marketing to make all the pseudo-creators and wannabe-streamers want 16C or more. Real professionals will be quite happy with the new Threadripper. Competition is nice! eames fucked around with this message at 16:49 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 16:47
|
|
|
ilkhan posted:AKA until they get 10nm out in volume Intel is in bad shape. Actually given the leaked Intel roadmaps Intel's 10nm will still be blah to crappy even with volume production. That is why they're pushing hard for 7nm and for 7nm+ to follow quickly now. eames posted:the question is are they still planning to get 10nm out in volume, except in some mobile devices? So far as anyone knows its still just low low clocked (~2Ghz or less base clocks, which are what mostly matter for most of the devices they'll be going in) U and Y models for 10nm all the way to the end of 2021. eames posted:I think their 14nm CPUs will stay reasonably competitive in the 6-10C market Intel generally doesn't really like competing on price though and their recent cuts are kinda ho hum. You have to go back to the P4 era really to find the last time Intel was really willing to compete on price. PC LOAD LETTER fucked around with this message at 16:56 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 16:47
|
|
|
PC LOAD LETTER posted:Actually given the leaked Intel roadmaps Intel's 10nm will still be blah to crappy even with volume production.
|
|
#
?
Jun 26, 2019 16:55
|
|
|
ilkhan posted:Huh. Well thats lovely. Was looking forward to a quad core 3ghz surface laptop for my next laptop. IPC is rumored to go up so the 10nm parts might just perform like a 3 GHz Coffee Lake CPU. iGPU part is also rumored to get a pretty huge upgrade. I read that Microsoft is working with AMD and Qualcomm (link in german) for the next refresh of Surface devices.
|
|
#
?
Jun 26, 2019 17:02
|
|
|
I'm torn between the 3800x and 3900x for my next gaming rig. Aside from price, are there any situations where a 3800x would outperform a 3900x? I know the 3900x has ridiculous cache, which is nice. Only downside could be the 3900x is two 6-core chiplets while the 3800x is a single 8-core chiplet, right? Would that have any latency issues where a single chiplet may have the advantage over two, or is it another "we don't know yet, wait for the benchmarks" situation. On that note, the other last thing I'm debating is 16gb vs 32gb RAM. Usecase is strictly playing games, no rendering or video editing or workstation stuff. I know 16gb is more than enough for today, but I also wanna build a big beefy rig that'll be solid for 5+ years (maybe throw a new GPU in 2-3 years from now) and I'm wondering if resource requirements will continue to bloat and three years down the road having >16gb is desirable. And the reason I ask now is it'd be better to get 2x16 sticks now rather than end up running 4x8, at least on the mobo I want to get. Thinking 3600CL16 is the sweet spot.
|
|
#
?
Jun 26, 2019 17:12
|
|
|
DDR4 prices are at an all time low, so I'd say now is a pretty good time to pick up 32GB just because you can. You'll get 32GB for significantly less money than what you would have to have paid for 16GB of the same speed memory a year ago. For gaming you won't actually need it though. As for the CPU choice, I'd say just wait for benchmarks. Llamadeus posted:Here's a test on AM4 with small FFTs with a moderately overclocked 1700X, 230W total system power, comparing a D15 with 240mm and 280mm AIOs (one with Noctua fans). TheFluff fucked around with this message at 17:26 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 17:22
|
|
|
I'm tempted to do just that, but I'm sort of waiting to see what ram goes well with the new stuff. The wait for real reviews is getting unbearable.
|
|
#
?
Jun 26, 2019 17:25
|
|
|
Background processes related to gaming can eat up a ton of RAM. Let's say you have music running, voice communications, a copy of chrome open with a character build plan. Maybe a Netflix or Twitch running as well. That's my normal use case and I end up using 20GB out of 32GB on my 2500k. I don't think I ever ran out of RAM on this computer. My personal plan is 2x16GB so I have the option to move to 4x16GB if games end up needing more RAM. I'm sure I can also cut my RAM usage down by closing browser tabs. I will say it's really nice to never worry about having enough RAM and know you aren't hitting the page file (don't turn paging off, windows does not like it, at least 7 didn't)
|
|
#
?
Jun 26, 2019 17:27
|
|
|
Intel's Steve Collins posted:...the press likes to focus on Intel's top price being higher than AMD's top price... Darn those a-cussed knaves in the press, with their unfair price comparisons of the chips Intel trots out to win benchmarks, to the chips AMD trots out to win benchmarks! Edit: the opening part of that article was the most interesting to me. It started out being completely factual and honest, then had to veer slightly into sins of omission in an attempt to explain why AMD is suddenly doing so well without saying words like "It's because they decided to redefine the playing field by giving people poo poo-tons of cores for cheap." mdxi fucked around with this message at 17:39 on Jun 26, 2019 |
|
#
?
Jun 26, 2019 17:35
|
|
|
Azuren posted:"we don't know yet, wait for the benchmarks" situation.
|
|
#
?
Jun 26, 2019 17:40
|
|
|
pixaal posted:Background processes related to gaming can eat up a ton of RAM. Let's say you have music running, voice communications, a copy of chrome open with a character build plan. Maybe a Netflix or Twitch running as well. That's my normal use case and I end up using 20GB out of 32GB on my 2500k. I don't think I ever ran out of RAM on this computer. My personal plan is 2x16GB so I have the option to move to 4x16GB if games end up needing more RAM. I'm sure I can also cut my RAM usage down by closing browser tabs. Browsers behave similar to databases now, aggressively caching content and ballooning the amount of ram it appears to need. However, most of this memory is being spend on caching of content with very low cache rates and would be released effectively instantly the second there was memory pressure on the system and another application called for it. It distorts the perception of how much memory is actually needed to run a given workload.
|
|
#
?
Jun 26, 2019 17:57
|
|
|
eames posted:IPC is rumored to go up so the 10nm parts might just perform like a 3 GHz Coffee Lake CPU. iGPU part is also rumored to get a pretty huge upgrade. I'm begging you, AMD, please don't put Vega into anything anymore. I want that Surface Go 2 with Navi in it, at minimum.
|
|
#
?
Jun 26, 2019 18:12
|
|
|
As long as Microsoft is willing to pay for it I'm sure AMD's Semi Custom Division is happy to sell them chips with whatever graphics they want.
|
|
#
?
Jun 26, 2019 18:20
|
|
|
|
| # ? Apr 25, 2024 22:30 |
|
|
eames posted:IPC is rumored to go up so the 10nm parts might just perform like a 3 GHz Coffee Lake CPU. Intel has had the i5-8269U (a 28W 2.6Ghz base clocks 4C/8T Coffeelake) out since mid-ish 2018 for around $330 tray price. What Intel's 10nm will offer is a good reduction in TDP (maybe down to 15W TDP?) at least for that level of performance (CPU's like the i5-8269U and Core i5-8259U seem to be mostly in NUC's and not stuff like a MS Surface which really needs a 15W TDP or less) but actual effective peak performance isn't really going to change much from current parts. They did improve the iGPU a fair amount vs previous gen but its really going to come down to what RAM its paired with. If you get a model that has LPDDR4X 3733 paired with it iGPU performance will be pretty good. If the model has something more like DDR4 2400 or 2666 it'll be a different story.
|
|
#
?
Jun 26, 2019 19:44
|
|