|
Seems like differential signalling would be a big improvement. I suppose it is just too many pins.
|
#
?
Apr 5, 2020 20:35
|
|
|
|
| # ? May 17, 2024 09:30 |
|
|
Are there PCBs designs that implement "twisted pair" directly on the traces? like alternating/criss-crossing between two or more layers? Or maybe that's pointless when you can just add solid shielding layers?
|
|
#
?
Apr 5, 2020 20:56
|
|
|
gradenko_2000 posted:https://www.youtube.com/watch?v=7TUjOb1H5pg The difference in GPU temperature is insane, the single-channel side might say it's at 100% usage but it's all idle cycles waiting for data.
|
|
#
?
Apr 5, 2020 21:08
|
|
|
peepsalot posted:Are there PCBs designs that implement "twisted pair" directly on the traces? like alternating/criss-crossing between two or more layers? Or maybe that's pointless when you can just add solid shielding layers? I haven't seen it but that doesn't mean it isn't a thing of course. Generally you want the impedance to be controlled and matched, using vias adds a lot of chaos to that. Plus the distance on the PCB is usually much much shorter than a cable.
|
|
#
?
Apr 5, 2020 23:12
|
|
|
Twisted pair is twisted to keep the wires at a small constant distance from each other, which makes any external interference produce roughly the same noise on both wires. With differential signalling the noise then cancels out when you measure the difference in the two signals on the other end. The helical physical shape of the wires isn't important in itself, it's just easy to manufacture. They also need to match the length, impedence etc. The equivalent on a PCB is a pair of matched traces which use differential signalling. All the nice fast serial links in a PC work like this: PCI-E, SATA, HDMI, USB, Ethernet etc. If you look at a board near these you can often see the pairs.  On RAM specifically I think only the clocks are differential. Doubling the data traces would be pretty extreme!
|
|
#
?
Apr 6, 2020 00:40
|
|
|
I remember when USB SS first came out and my friends bitching about having to get that high speed signal working on a 6ft cable.
|
|
#
?
Apr 6, 2020 02:06
|
|
|
Malcolm XML posted:Ddr is the last parallel single ended interface in wide use, they could switch to serial like IBM does.
|
|
#
?
Apr 6, 2020 13:48
|
|
|
ConanTheLibrarian posted:Uh whu?? Does IBM has some kind of proprietary memory format that's way faster than DDR? They use normal DDR memory chips, but have another chip called Centaur that's a memory controller and buffer.  The Centaur chip sits between the CPU and RAM (it's actually on the giant memory module PCBs), and each of those has a high-speed link to the CPU. Each CPU-Centaur link is way slower than CPU-DDR4 in a desktop, but you can have lots of them. Physically and electrically it makes a ton of sense to do it that way, the wires connecting CPU and DDR ram in normal PCs are a mess. Economically it would be painful, putting an expensive controller chip on every stick would suck. And for desktops I don't think it would scale down very well.
|
|
#
?
Apr 6, 2020 14:44
|
|
|
so it's just buffered ram? 
|
|
#
?
Apr 6, 2020 15:34
|
|
|
Late watercooling talk... I went with one for the water (more so via reservoir) as lowpass filter to temp spikes, and I don't want like 1.5kg lever on my mainboard (e.g. Dark Rock Pro 4). --edit: And I guess with a reservoir and an appropriate water-temperature based fan curve, it takes a long while for the system to saturate and allows for turboing longer. Combat Pretzel fucked around with this message at 15:43 on Apr 6, 2020 |
|
#
?
Apr 6, 2020 15:39
|
|
|
Arzachel posted:If you want to drive a 240hz (or even 144hz, depending on the game) monitor, there's a pretty good argument for going with a 9900k over Zen 2. "Dynamic Super Resolution renders a game at a higher, more detailed resolution and intelligently shrinks the result back down to the resolution" Arzachel posted:I agree but competitive games are a big reason behind the 144hz push and you want the framerate to be rock solid for that. Ironically though, CSGO is one of the games where it's a wash between AMD and Intel for some reason.
|
|
#
?
Apr 6, 2020 17:03
|
|
|
Resolution doesn't change the CPU side of the equation neither do most graphic settings so lowering settings wouldn't matter all that much for AMD vs Intel in CSGO a game that isn't that hard to 300+ fps even at 1440p. Now keeping above 300 in a firefight that's a different story but your goal is going to be 144fps so you got plenty of wiggle room.
|
|
#
?
Apr 6, 2020 17:16
|
|
|
https://www.youtube.com/watch?v=PIeMGobEFzA The Athlon 3000G is the best potato ever.
|
|
#
?
Apr 7, 2020 06:01
|
|
|
taqueso posted:I haven't seen it but that doesn't mean it isn't a thing of course. Generally you want the impedance to be controlled and matched, using vias adds a lot of chaos to that. Plus the distance on the PCB is usually much much shorter than a cable. Its theoretically possible but PCBs are on an exponential scale cost wise for layers. In a previous life anything we did that was more than 10/12 layers had to be full-custom/customer paying NRE insane poo poo. IBM themselves do come to mind idk if they ever figured out we didn't actually support USB3 to the front panel
|
|
#
?
Apr 7, 2020 15:11
|
|
|
Quaint Quail Quilt posted:Turning the settings down like 20% from max gets me to 144 fps in most games, other older games I pretty up with DSR at the expense of some fps. when i was playing fps competitively, i lowered settings primarily to see poo poo better because it was less flashy. can't hide in the darkness/fog/grass if there's none of those on my screen, sucker. the fps boost is a nice addition tho.
|
|
#
?
Apr 7, 2020 15:59
|
|
|
Truga posted:when i was playing fps competitively, i lowered settings primarily to see poo poo better because it was less flashy. can't hide in the darkness/fog/grass if there's none of those on my screen, sucker. More or less this. PubG had to do some pretty impressive raster tricks in order to get people who set the game to 'running on half of a potato and a wedge of lemon' graphics quality from winning every bush/field snipe fight.
|
|
#
?
Apr 7, 2020 17:28
|
|
|
https://www.tomshardware.com/news/amd-ryzen-4000-desktop-cpus-release-date Apparently Ryzen 4K was to be announced at Computex. Now they'll be announced in September, whether Computex actually happens or not. Also, apparently 600-series chipsets will be coming, so I guess stop waiting on the B550 to happen?
|
|
#
?
Apr 8, 2020 19:43
|
|
|
I am not sure what they could do with the 4,000 series that the 3,000 series can't. Higher clock speeds/single threaded performance?
|
|
#
?
Apr 8, 2020 19:50
|
|
|
ratbert90 posted:I am not sure what they could do with the 4,000 series that the 3,000 series can't. Higher clock speeds/single threaded performance? I'm pretty certain single thread perf per clock is already better than Intel, they basically just need to keep working on lowering latencies and bump clocks a bit
|
|
#
?
Apr 8, 2020 19:55
|
|
|
Got my Sliger SM560 yesterday and crammed my new 3700x system into it. It's amazing the amount of performance I have concentrated in something smaller than a shoebox. 8c/16t, 32gb of ram, 2.75tb of storage, and a GTX1080. Fired up furmark alongside prime95 and my GPU kept max TDP and only ramped the fan to about 67% and the CPU settled in around 73c. Also all on a 450w PS. Amazing time we live in. bull3964 fucked around with this message at 20:20 on Apr 8, 2020 |
|
#
?
Apr 8, 2020 20:16
|
|
|
bull3964 posted:Got my Sliger SM560 yesterday and crammed my new 3700x system into it. It's amazing the amount of performance I have concentrated in something smaller than a shoebox. 8c/16t, 32gb of ram, 2.75tb of storage, and a GTX1080. What cpu cooler are you using?
|
|
#
?
Apr 8, 2020 20:38
|
|
|
ratbert90 posted:I am not sure what they could do with the 4,000 series that the 3,000 series can't. Higher clock speeds/single threaded performance? All we know, in the form of actual AMD statements, is * Mark Papermeister saying to expect "similar IPC improvements" to the previous gen. * Confirmation that the process will still be 7nm * So improvements are architectural, not process driven * Edit: A unified L3 cache model
|
|
#
?
Apr 8, 2020 21:08
|
|
|
B-Mac posted:What cpu cooler are you using? Noctua NH-L9a
|
|
#
?
Apr 8, 2020 21:14
|
|
|
mdxi posted:All we know, in the form of actual AMD statements, is So assuming a small frequency bump from the refined 7nm process, we're talking maybe 15% faster than Zen2. Not bad, but it will be interesting to see if they can improve the cache latency to a significant degree.
|
|
#
?
Apr 8, 2020 22:27
|
|
|
also haven't they got it to an 8 core CCD as well? So no CCX poo poo for a Ryzen 4700/4800
|
|
#
?
Apr 8, 2020 22:32
|
|
|
HalloKitty posted:I'm pretty certain single thread perf per clock is already better than Intel, they basically just need to keep working on lowering latencies and bump clocks a bit In Cinebench and productivity, yeah. In games, no. The difference is that the core chiplet incurs some memory latency going to the IO die, so things that are memory latency dependent are less happy. There are a few articles out there about it that i remember reading, but here is the one that came up first when i googled just now: https://www.techspot.com/article/1876-4ghz-ryzen-3rd-gen-vs-core-i9/ e: i reread your post and thats probably what you meant anyway but oh well i already made the post and we are all stuck with it now!! Cygni fucked around with this message at 22:45 on Apr 8, 2020 |
|
#
?
Apr 8, 2020 22:39
|
|
|
If AMD can do another 10-15% gain in IPC that will put them basically on par with Intel’s gaming performance. I doubt there will be huge leaps in clockrate but they might get another couple percent. (“IPC” as a measurement is really a top line figure, it includes things like cache improvements, not just pure theoretical algebraic/FP throughput. So when Papermaster says it’ll be around X% then don’t mentally add “and then cache improvements” on top of that. Of course, things like cache improvements can have more or less impact on various workloads, IPC is a very fuzzy term.) Basically in total I expect them to come in roughly on par with a 5 GHz Intel in gaming performance, 0-5% ahead. But at much better power/temps. Still, nothing to upgrade over if you have been enjoying a Coffee Lake for the past couple years, and people seem to have agreed this level of performance difference over Zen2 is insignificant and unnoticeable, so no reason for those users to upgrade either. In particular though it’ll be hilarious to see what CS:GO can do with a unified 32MB cache. It really seems to love the cache on Zen2, it is already significantly faster than on Intel. Paul MaudDib fucked around with this message at 00:08 on Apr 9, 2020 |
|
#
?
Apr 8, 2020 23:39
|
|
|
uhhhhahhhhohahhh posted:also haven't they got it to an 8 core CCD as well? So no CCX poo poo for a Ryzen 4700/4800 Yeeeeeeeeees, but it's less of an 8-core chipet and still two 4-core chiplets on the same piece of silicon. Like, the Zen/Zen+ CCXes were side-by-side, right? And Zen 2 moved them to be over/under, while also stripping out the IO bits off to their own die. Zen+:  https://commons.wikimedia.org/wiki/File:Ryzen_5-2600.jpg Zen 2:  https://commons.wikimedia.org/wiki/File:Zen2_Matisse_Ryzen_7nm_Core_Die_shot.jpg But even though they're "over-under" now, and on the same piece of silicon, they still have separate L3 caches, instead of one shared across both quadcore complexes. I believe the L3 cache is still 16-way associative, independent between "halves", the next step would probably to get the two L3 caches talking to eachother? And then from there, properly, eight-core chiplets instead of roommates. SwissArmyDruid fucked around with this message at 01:24 on Apr 9, 2020 |
|
#
?
Apr 9, 2020 01:15
|
|
|
Some leaked slides a few months ago showed Zen 3 having a unified L3 per chiplet. Rumor obvi, so take it for what it is worth. e: found em:  
Cygni fucked around with this message at 02:17 on Apr 9, 2020 |
|
#
?
Apr 9, 2020 02:15
|
|
|
the really interesting question that arises is, what is the consequence for the CCX? Is this effectively a move to an 8-core CCX, and what are the implications for cross-core latency (not memory latency)?
|
|
#
?
Apr 9, 2020 02:27
|
|
|
Paul MaudDib posted:Basically in total I expect them to come in roughly on par with a 5 GHz Intel in gaming performance, 0-5% ahead. But at much better power/temps. Still, nothing to upgrade over if you have been enjoying a Coffee Lake for the past couple years, and people seem to have agreed this level of performance difference over Zen2 is insignificant and unnoticeable, so no reason for those users to upgrade either. Should be a decent enough upgrade from the 2000 series though, shouldn't it? I think Zen+ was still a bit behind similar Intel CPUs at the time (except in price and maybe price/perf) so going from that to something that has parity with a 5Ghz Intel would be pretty good, no?
|
|
#
?
Apr 9, 2020 14:39
|
|
|
If they can pull off 15% improvement that is a fairly significant upgrade.
|
|
#
?
Apr 9, 2020 14:48
|
|
|
The pro move was obviously to get a 3600 and then wait and see what the new console ports require, then get a 4xxx if necessary when the time comes 
|
|
#
?
Apr 9, 2020 14:58
|
|
|
Icept posted:The pro move was obviously to get a 3600 and then wait and see what the new console ports require, then get a 4xxx if necessary when the time comes The ultimate placeholder CPU is clearly a 1600 As gently caress
|
|
#
?
Apr 9, 2020 15:03
|
|
|
Athlon 1600+ e: no regrets on my 3950x tho, holy moly is this expensive piece of poo poo good.
|
|
#
?
Apr 9, 2020 15:11
|
|
|
Icept posted:The pro move was obviously to get a 3600 and then wait and see what the new console ports require, then get a 4xxx if necessary when the time comes This is exactly what I did. I also did the same buying a 5700 for £300 but that hasn't been as successful.
|
|
#
?
Apr 9, 2020 17:05
|
|
|
Hi do I get a 3600 or a 10600k? Only for gaming at 1080p 144hz.
|
|
#
?
Apr 9, 2020 20:59
|
|
|
I’m currently running a Kaby-lake system and I’m hearing a lot of buzz about AMD once again being the superior CPU for gaming applications. Can anyone explain why? I’m looking at building a new gaming system but feel hesitant about moving to AMD. The last time I had an AMD system I had all sorts of issues making my RAM work properly without crashing my system randomly. With intel everything is plug and play on my ROG motherboard. I looked at benchmarks and intel still seems to edge out AMD for gaming applications. Having said that it seems the new consoles are going to be taking advantage of multi threading in ways previous ones have not. I imagine this means AMDs core advantage will actual make the 3000 and future 4000 series better than Intel? TLDR: longtime intel gamer planning a new system build and debating getting a 3900X or 4000 series. Good idea or bad idea?
|
|
#
?
Apr 9, 2020 20:59
|
|
|
Kraftwerk posted:debating getting a 3900X or 4000 series. Good idea or bad idea? The 4X00s won't even be announced until September, so you can't make any meaningful decision there. I don't personally know why you'd want a 3900X to play video games on. You're paying a huge premium over a 12 or 16 thread CPU, in exchange for lower clocks. The XBX and PS5 CPUs are going to look a lot like the 3700, if that helps any. In my view, the 3900 and 3950 are quasi-workstation CPUs, for people who do want/need more cores but don't need a hundred PCI lanes and a terabyte of RAM.
|
|
#
?
Apr 9, 2020 23:34
|
|
|
|
| # ? May 17, 2024 09:30 |
|
|
Kraftwerk posted:I’m currently running a Kaby-lake system and I’m hearing a lot of buzz about AMD once again being the superior CPU for gaming applications. mdxi posted:The 4X00s won't even be announced until September, so you can't make any meaningful decision there. Clocks are actually a bit higher and it's not that much of a premium for the 3900X anymore as it's selling for $400-420. Still doesn't make sense for just gaming and nothing else though, in that case the 3600 or 3700X makes the most sense. There were some quirks with the Ryzen 3000 parts at launch but they're long since ironed out, I got the 3950X at launch time and have had no issues. The main advantage of AMD for gaming is saving a lot of money while only being 5-10% slower than Intel worst case, especially if you get the 3600. The CPUs themselves are cheaper than their Intel counterparts and also they come with bundled coolers of decent quality for additional cost savings. MaxxBot fucked around with this message at 23:50 on Apr 9, 2020 |
|
#
?
Apr 9, 2020 23:44
|
|
