|
B-Mac posted:That was overclocked running for 16 hours with the plexiglass cover on the test bench so it wasn’t open air at that point. At 9:55 in he gives the stock clock VRM temps on the open air test bench as a baseline and says its doing 88C under load for that mobo. https://www.youtube.com/watch?v=EqQcgwz1hYA
|
#
?
Aug 11, 2018 15:53
|
|
|
|
| # ? May 16, 2024 11:01 |
|
|
PC LOAD LETTER posted:At 9:55 in he gives the stock clock VRM temps on the open air test bench as a baseline and says its doing 88C under load for that mobo. Ah I thought you talking about the level1tech video, not hardware unboxed. I just finished the Buildzoid video you linked and found it funny that the Asus pro heat sinks make the vrms hotter than some of these B450 boards, with the vrm clocking in at 92C or higher. It would be nice if these boards came with functional heat sinks and I/O covers that didn’t restrict airflow. B-Mac fucked around with this message at 16:08 on Aug 11, 2018 |
|
#
?
Aug 11, 2018 16:03
|
|
|
B-Mac posted:level1tech video B-Mac posted:I just finished the Buildzoid video you linked and found it funny that the Asus pro heat sinks make the vrms hotter than some of these B450 boards, with the vrm clocking in at 92C or higher. I was also clear about it running too hot and how to address that in my original post too that you quoted. B-Mac posted:It would be nice if these boards came with functional heat sinks and I/O covers that didn’t restrict airflow. PC LOAD LETTER fucked around with this message at 16:16 on Aug 11, 2018 |
|
#
?
Aug 11, 2018 16:14
|
|
|
PC LOAD LETTER posted:Fair enough. I think its perfectly fine to hold such views. I just didn't get that from your original comment on this subject. You know more about VRMs than I do, which is why I wanted to talk to you about them instead of why I’m frivolously overspending on coolers. Edit: From what I see here, both boards have the same controller, but the Carbon has a doubler? I don’t know what that does exactly besides “not as good as actually having two times the phases”. Craptacular! fucked around with this message at 20:37 on Aug 11, 2018 |
|
#
?
Aug 11, 2018 20:31
|
|
|
Craptacular! posted:I bought a 120mm AIO to run at a machine at stock on quiet mode for five years. I know I’m overbuying on coolers. Buying a 110i/115i PRO for over $100 and putting it on a cheap motherboard looks silly from a performance standpoint, but performance isn’t all that matters so long as nothing overheats and dies. So that’s why I asked about the Gaming Pro Carbon AC, which is like $20 more than the Titanium but means stepping up to full ATX. (Newegg just debuted the Titanium this morning for $110, while the Carbon is $130.) For most intents and purposes, a doubled 4-phase is pretty much as good as an actual 8-phase. The point of having a multi-phase VRM is to let many different components share the current load - in an 8-phase VRM, each phase (essentially consisting of a pair of MOSFETs, an inductor and one or more capacitors) is only turned on 1/8th of the time. These things turn on and off very fast - 300 kHz switching frequency is pretty typical, while higher end VRM's can be optimized for 500 kHz. You want a higher switching frequency and more "real" phases to smooth out voltage ripples and reduce transient voltage spikes since that helps keeping the CPU stable when overclocking. A "fake" 8-phase with eight high/low MOSFET pairs and eight inductors but only four controller phases will turn on two "phases" at a time - each pair of phases will be turned on 1/4th of the time. For the most part it will behave like a "real" 8-phase thermally (since it splits the heat load over 8 phases' worth of components), but electrically it will be a 4-phase. A doubled 4-phase has the controller chip output PWM signals that turns on and off 4 doubler chips, and then each of those doubler chips will act as a pretty dumb 2-phase controller that turns on and off two phases alternately. Hence, you get 8 actual electrical phases, so it's almost as good as a "real" 8-phase. The only thing you usually miss out on is current balancing between the two phases in each pair because the doubler chips are usually too dumb to do that, but I think that's pretty much irrelevant in reality. Hope I managed to explain it somewhat. If you want the gory details I recommend Texas Instruments' Application Report SLVA882, Multiphase Buck Design From Start to Finish (Part 1). It's actually pretty approachable! e: WikiChips actually has an even more approachable explanation TheFluff fucked around with this message at 23:59 on Aug 11, 2018 |
|
#
?
Aug 11, 2018 23:23
|
|
|
I was really surprised my X470 ITX doesn't include any LLC settings in the UEFI. What's up with that?
|
|
#
?
Aug 11, 2018 23:41
|
|
|
TheFluff posted:For most intents and purposes, a doubled 4-phase is pretty much as good as an actual 8-phase. The point of having a multi-phase VRM is to let many different components share the current load - in an 8-phase VRM, each phase (essentially consisting of a pair of MOSFETs, an inductor and one or more capacitors) is only turned on 1/8th of the time. These things turn on and off very fast - 300 kHz switching frequency is pretty typical, while higher end VRM's can be optimized for 500 kHz. You want a higher switching frequency and more "real" phases to smooth out voltage ripples and reduce transient voltage spikes since that helps keeping the CPU stable when overclocking. This post was really informative, thanks. When you say doubled are you talking about have two high side and two low side mosfets per phase, or is there more to it?
|
|
#
?
Aug 12, 2018 00:36
|
|
|
B-Mac posted:This post was really informative, thanks. When you say doubled are you talking about have two high side and two low side mosfets per phase, or is there more to it? Sorry, I got a bit unclear with the terminology. When I said "doubled 4-phase" in the post above I meant "an 8-phase that uses a 4-phase controller plus doubler chips". That is, the controller outputs 4 on/off signals (PWM signals), but instead of going directly into a MOSFET/inductor (a "phase"), the signal first goes into a doubling chip that turns it into two interleaved (alternating) signals. Hence, you get 8 distinct on/off signals and each MOSFET is turned on 1/8th of the time. You have 8 actual phases in every sense of the word, it's just that the controller chip only sees 4 phases and can only do current balancing on pairs of phases. It would perhaps be more accurate to call it a "8-phase with doublers"...? When I was talking about a "fake 8-phase", I really meant "a 4-phase that tries to look like an 8-phase by having two of everything in each phase". So, just as with the doubled 4-phase, you have 4 PWM signals coming from the controller, and you have 8 power stages (high/low MOSFETs) and 8 inductors, but they turn on and off in pairs, so each one of them is turned on 1/4th of the time instead of 1/8th of the time. You have the components for 8 phases but from a control standpoint you only have 4. Hope that clarifies it. I really need to come up with a term other than "phase" when I want to talk about the actual power circuitry so I don't have to go "a power stage, inductor and capacitor" every time. e: for clarity, a power stage is the high and low side MOSFETs combined into a single chip, usually plus some temp/voltage monitoring bits TheFluff fucked around with this message at 01:06 on Aug 12, 2018 |
|
#
?
Aug 12, 2018 00:49
|
|
|
TheFluff posted:Sorry, I got a bit unclear with the terminology. When I said "doubled 4-phase" in the post above I meant "an 8-phase that uses a 4-phase controller plus doubler chips". That is, the controller outputs 4 on/off signals (PWM signals), but instead of going directly into a MOSFET/inductor (a "phase"), the signal first goes into a doubling chip that turns it into two interleaved (alternating) signals. Hence, you get 8 distinct on/off signals and each MOSFET is turned on 1/8th of the time. You have 8 actual phases in every sense of the word, it's just that the controller chip only sees 4 phases and can only do current balancing on pairs of phases. It would perhaps be more accurate to call it a "8-phase with doublers"...? No that’s great, thanks a lot for responding. Is there an easy way for a person to look at a motherboard and tell whether something is for example a true 8 phase or in actuality a doubled 4 phase. I’ve noticed a lot of manufacturer marketing BS seeping in AMD calling things something when it’s actually not. That wikichip link was very informative, especially to someone who admittedly has little knowledge in the subject and is trying learn. B-Mac fucked around with this message at 02:00 on Aug 12, 2018 |
|
#
?
Aug 12, 2018 01:55
|
|
|
Craptacular! posted:the Carbon has a doubler? I don’t know what that does exactly besides “not as good as actually having two times the phases”. But yeah a doubled 4 phase isn't exactly quite the same or as good as a proper 8 phase but in practical terms it doesn't matter and is close enough to not matter to even a picky type person. I'd be plenty happy with a doubled 4 phase VRM to feed a overclocked 8C/16T Zen, Zen+, or Zen2.
|
|
#
?
Aug 12, 2018 13:30
|
|
|
Well after all that researching and the realization that there are no X470 boards at mATX, I decided to just go ahead with my original plan of a Mortar and the eventual 8-core Matisse running at stock. I never felt the need to OC things at the start of their lifespan and by the end of AM4’s run in 2020 hopefully someone will have made a nice mATX board with more than four phases. The closest we have right now is ASRock’s doubled 3 phase. I guess I could just opt for an ITX board in there, though, if I get paranoid enough about it. MSI has a true 6 phase one.
|
|
#
?
Aug 12, 2018 19:46
|
|
|
I've now got 24GB of RAM running at 2666 Mhz on a Ryzen 2200G APU. Is it worth tightening the timings or try to OC the speed? (see attachment)
|
|
#
?
Aug 13, 2018 12:03
|
|
|
TR2 reviews are starting to trickle out at the usual places.
|
|
#
?
Aug 13, 2018 14:30
|
|
|
Mark Larson posted:I've now got 24GB of RAM running at 2666 Mhz on a Ryzen 2200G APU. Is it worth tightening the timings or try to OC the speed? (see attachment)
|
|
#
?
Aug 13, 2018 14:43
|
|
|
PC LOAD LETTER posted:Try to get it to around DDR4 2933 if you can and then see if you can reduce the timings. I don't think you'll be able to do much better than that even with the primo B die DRAM if you're running more than 2 sticks of RAM or a double sided DIMM and it'd get you ~90% of the performance gains anyways. You mean I should keep the same timings and just try to bump up the speed? Also how do I know if I have the B die stuff? They are Crucial Ballistix series if that matters (probably doesn't), I took a peek in HWInfo and two of the sticks are single-rank, the other two are dual-rank. The SDRAM Manufacturer in all cases is Micron Tech.
|
|
#
?
Aug 13, 2018 15:36
|
|
|
.
sincx fucked around with this message at 05:50 on Mar 23, 2021 |
|
#
?
Aug 13, 2018 15:46
|
|
|
Man, TR2 has some really loving great-looking packaging. I am normally allergic to physicaly *stuff*, but I almost want a TR2 box for no good reason at all.
|
|
#
?
Aug 13, 2018 15:55
|
|
|
I don't normally give PCWorld the time of day, but rather than just report the as-expected, "Intel wins single-threaded, AMD wins multi-threaded" Gordon Mah Ung did a little something out of the box, running multiple 3D renderers at the same time. The results are probably as expected, but the ramifications are that instead of walking away from your computer while it's rendering, you could, in fact, just continue working on something else, or render multiple things, is somewhat interesting.
|
|
#
?
Aug 13, 2018 17:07
|
|
|
Direct benchmark side-by-side: https://twitter.com/IanCutress/status/1028994758411202562
|
|
#
?
Aug 13, 2018 18:58
|
|
|
The 2990WX seems available next door in Germany already, but the newer cooling solutions like the Dark Rock Pro TR4 or the second version of the Liqtech still aren't. Hope that changes by the time the 2950X is out. :|
|
|
#
?
Aug 13, 2018 21:52
|
|
|
B-Mac posted:No that’s great, thanks a lot for responding. Is there an easy way for a person to look at a motherboard and tell whether something is for example a true 8 phase or in actuality a doubled 4 phase. It's almost impossible to tell at a glance because the heatsinks cover up all the important parts. Sometimes the doubler chips will be on the backside of the mobo because they're pretty tiny so that could be a clue, but that only seems likely if you have the thing in your hands since product shots rarely show the back. You have to actually own the thing so you can take the heatsinks off and ID the chips. OTOH true 8 or doubled 4 is so minor a distinction that even buildzoid, the guy with the highest bar for "acceptable VRMs" around, thinks doublers are fine and is ok with that being marketed as 8 phases. Unless you plan to go into heavy overclocking -- including overvolting -- it's really not worth worrying too much about this poo poo. Every AM4 motherboard will run every ryzen chip at stock settings. Make common sense decisions, like if you're buying a $350 cpu don't put it in a bargain basement motherboard.
|
|
#
?
Aug 14, 2018 03:49
|
|
|
Stock 2990WX can gulp ~400W under cinebench and ~450W under prime (whole system with an idle 1080ti) 
|
|
#
?
Aug 14, 2018 07:54
|
|
|
ufarn posted:Direct benchmark side-by-side: All numbers in the wild right now should be taken with a huge spoon of salt, because microsoft is loving something up. Yes, it's phoronix, but numbers are numbers and their windows numbers seem to be roughly on par with anandtech's: https://www.phoronix.com/scan.php?page=article&item=2990wx-linux-windows Microsoft *really* needs to get its poo poo in order here, jesus christ.
|
|
#
?
Aug 14, 2018 09:49
|
|
|
Has anyone done testing on Windows Server to compare?
|
|
#
?
Aug 14, 2018 11:19
|
|
|
Mark Larson posted:Has anyone done testing on Windows Server to compare? they did comparison 2990WX benches on windows and linux, and linux routinely blows windows out of the water: https://www.phoronix.com/scan.php?page=article&item=2990wx-linux-windows&num=1 e: f;b
|
|
#
?
Aug 14, 2018 13:30
|
|
|
Truga posted:Yes, it's phoronix Aren't they full on spergs though, and can be trusted with benchmarks?
|
|
#
?
Aug 14, 2018 13:44
|
|
|
GRINDCORE MEGGIDO posted:Aren't they full on spergs though, and can be trusted with benchmarks? they've come under fire in the past for the quality of their benchmarks
|
|
#
?
Aug 14, 2018 14:18
|
|
|
Any real difference between 1950X and 2950X performance? Is it the same difference as Ryzen, where you'd need to tune the 1950X manually and the 2950X can just enable XFR2 and be done?
|
|
#
?
Aug 14, 2018 23:40
|
|
|
Paul MaudDib posted:Any real difference between 1950X and 2950X performance? Is it the same difference as Ryzen, where you'd need to tune the 1950X manually and the 2950X can just enable XFR2 and be done? From what I could find the 2950x can clock about 200mhz higher at most or so than the 1950x, usually just a 100 MHz higher. Best case I saw was about 10% faster, usually in the 5% range. Seems xfr2 and PBO is what gets it to peak higher but voltage nears 1.45V. Basically yea what you said, just turn on xfr2 and PBO. The $600 1950X is a drat steal if it suits your needs. B-Mac fucked around with this message at 00:17 on Aug 15, 2018 |
|
#
?
Aug 15, 2018 00:11
|
|
|
Ok, here's some dumb theory: Since the IF has a considerable power draw, and it probably scales with frequency, there's gotta be a cross-over point where faster RAM eats enough into the TDP to mess with boost and XFR. Essentially RAM too fast for the CPU's own good.
|
|
#
?
Aug 15, 2018 02:06
|
|
|
What was the speculated percentile cutoff for ryzen die quality going into TR/Epyc? Top 95%? 97%?
|
|
#
?
Aug 15, 2018 03:35
|
|
|
AMD told the press it's still top 5% for TR second-gen.
|
|
#
?
Aug 15, 2018 03:48
|
|
|
Has AMD addressed the bench disparity between Linux and Windows yet?
|
|
#
?
Aug 15, 2018 04:06
|
|
|
Read something that they're working with Microsoft on improving the scheduler. Yet again (didn't that happen with Ryzen 1 itself initially?).
|
|
#
?
Aug 15, 2018 12:01
|
|
|
Combat Pretzel posted:Read something that they're working with Microsoft on improving the scheduler. Yet again (didn't that happen with Ryzen 1 itself initially?). it went from terribad to acceptable, and now from acceptable to better?
|
|
#
?
Aug 15, 2018 12:57
|
|
|
It happened with Bulldozer. It was the thing that the AMD fanboys were pinning their hopes on to 'fix' Bulldozer's performance issues.
|
|
#
?
Aug 15, 2018 13:03
|
|
|
I think it's to deal with TR2s weird asf topology.
|
|
#
?
Aug 15, 2018 13:23
|
|
|
EmpyreanFlux posted:I think it's to deal with TR2s weird asf topology. Is it weird or is it just not what Intel does?
|
|
#
?
Aug 15, 2018 14:31
|
|
|
Seeing that discrepancy between linux and windows benchmarks makes me think of the intel c/c++ compiler and how it's intentionally crippled for non-intel processors. Not that that's what's happening here, but Linux uses an open compiler that produces code agnostic to intel/amd, Microsoft might be using some custom intel optimizations or something
|
|
#
?
Aug 15, 2018 14:52
|
|
|
|
| # ? May 16, 2024 11:01 |
|
|
Munkeymon posted:Is it weird or is it just not what Intel does? Haha yes it is weird. The 2990WX raises issues for desktop PCs that were previously the domain of multi-socket systems and then some. You have four NUMA nodes (two DMA dies, two "compute" dies that can only talk to main memory through the DMA dies over Infinity Fabric) and managing where to keep threads while accounting for memory locality on this thing is not at all trivial. If the chip gets fully loaded down you had better hope your application is tolerant of variance in latency because it'll be all over the place. The 2990WX is crazy (like a fox, because it works well most of the time). Intel probably would never build something like this because Intel architectures seem to be about providing consistent performance and not chancing cases where even subpar code would perform badly for any reason except not scaling well with thread counts. AMD is apparently more than happy to provide a part that has an equal chance of scaling like mad or making programs and operating systems fall flat on their faces. If they tried to sell this thing into the server market it would be laughed out of the room but high-end desktops are a little more wild-West and AMD has nothing to lose by trying.
|
|
#
?
Aug 15, 2018 15:31
|
|