|
When you're GPU bottlenecked on Ryzen, RAM speed won't matter that much, but if you're CPU bottlenecked, then RAM speed can seriously shift the bottleneck back to the GPU. IIRC the numbers testing Deus Ex Mankind Divided on an R7 1700 with a GTX 1080 had LegitReviews (I think it was them) at 95 FPS with 2133 and 110 FPS at 3200 when testing at 1920x1080, but they were GPU bottlenecked at 2560x1440.
|
#
?
Sep 17, 2017 00:55
|
|
|
|
| # ? Apr 25, 2024 20:19 |
|
|
So I I'm pairing the 1600 with a 1070, so go for the good ram? Edit: at 1440
|
|
#
?
Sep 17, 2017 01:12
|
|
|
NomNomNom posted:I just bought a 1600, I ordered 2133 ram before looking at performance impact, someone tell me how screwed I am and if so what I should get instead. Return the ram
|
|
#
?
Sep 17, 2017 03:34
|
|
|
NomNomNom posted:So I I'm pairing the 1600 with a 1070, so go for the good ram? Its only at 1080p you run into some CPU significant bottlenecking with DDR4-2133. You don't actually need DDR4-3200 to address this for the most part either. DDR4-2933/3000 can get you most of the gains that DDR4-3200 will and runs around $130-ish for 16GB of it going by newegg while 16GB of DDR4-2133 goes for around $120-ish. A ~$10 price difference is worth it. You'll probably have to do a little finangling of course to get it work at those speeds with Ryzen but unless you're after the lowest possible timings for each setting this isn't all that hard to do. FWIW I used some DDR4-3000 that was meant for Intel X99 and X100 platforms and it worked fine at DDR4-2933 with the XMP settings in the BIOS after bumping up the DRAM volts to 1.4 and increasing the SoC volts to ~1.1. Took a few minutes to do those changes in the BIOS and to reboot and its been fine for months now. Some reported they had to switch to T2 command rates to get it to work at those speeds with Ryzen but that still isn't all that bad for non-B die stuff. A few have gotten it to work at DDR4-3200 but don't buy it expecting that. It was this stuff here I used. Not worth buying right now if you're on a tight budget since the price went up to $150 for 16GB but it was $132 when I bought it a few months ago.
|
|
#
?
Sep 17, 2017 08:15
|
|
|
AMD stated that TR had glass spacers to physically support the package, Der8auer murdered a threadripper and it apparently turns out there's 4 full dies there. Now AMD has retreated to "well 2 of them are broken dies", but someone made an interesting comment: remember how Epyc is on a new stepping? It kinda looks like TR is actually a straight-up Epyc engineering sample, and they found some faults and decided to move Epyc to a new stepping instead.
Paul MaudDib fucked around with this message at 21:02 on Sep 17, 2017 |
|
#
?
Sep 17, 2017 19:19
|
|
|
Super weird.
|
|
#
?
Sep 17, 2017 20:48
|
|
|
I mean, when you think about it, glass and dead CPU dies are basically the same thing, both being made from SiO2 as raw materials and all. I'm betting a marketing drone Telephoned something in their head from something an engineer said. Probably what the engineer said was that these dies were completely dead-dead, as in, not "fused-off-so-you-might-be-able-to-do-shenanigans-like-go-from-Phenom II-X4-to-X6-dead", and that they might as well be glass, and then the marketing drone went "right, glass." SwissArmyDruid fucked around with this message at 22:10 on Sep 17, 2017 |
|
#
?
Sep 17, 2017 22:07
|
|
|
PC LOAD LETTER posted:If you're gaming at 1440p then you'll be GPU limited nearly always for any newer game. Thanks, I was looking at that exact RAM, ordered. I'll let you guys know in a few days how it all turns out
|
|
#
?
Sep 17, 2017 23:02
|
|
|
There's a new Ryzen bios out that seems to be aimed at improving memory compatibility.
|
|
#
?
Sep 17, 2017 23:06
|
|
|
PC LOAD LETTER posted:
what no it isn't. (or i misread and thought you meant saving 10 bux would be worth it." (edit: he didn't*. drat.) Eyochigan fucked around with this message at 08:00 on Sep 18, 2017 |
|
#
?
Sep 18, 2017 02:07
|
|
|
The whole segfault thing has got people tracking datecodes and there's one thing that I don't think has been noticed yet: AMD is producing the B1 and B2 steppings in parallel. Why? The assumption was that AMD was just burning off their stock of B1 dies and that Ryzen would eventually switch to B2 when they exhausted their supply of B1. But that doesn't seem to be happening. Is there a reason they are producing B1 and B2 in parallel instead of going to fully B2 stepping?
|
|
#
?
Sep 18, 2017 02:54
|
|
|
Paul MaudDib posted:The whole segfault thing has got people tracking datecodes and there's one thing that I don't think has been noticed yet: AMD is producing the B1 and B2 steppings in parallel. Why? Possible issues with the fabs, with only a limited amount able to produce the more flawless B2 compared to B1? And that the expense to retool to B2 spec is mitigated by the incoming (14nm+) retool to a better process anyway?
|
|
#
?
Sep 18, 2017 04:15
|
|
|
remember how epyc has mirrored dies?
|
|
|
#
?
Sep 18, 2017 07:23
|
|
|
If Epyc has both stereoisomers wouldn't they need to have both, otherwise two of the dies would be broken? Plausible though, I hadn't thought about that
|
|
#
?
Sep 18, 2017 08:49
|
|
|
http://techreport.com/news/32558/rumor-amd-raven-ridge-apu-scores-appear-on-geekbench Look! It's the NewFatMike signal! Finally, we're one step closer to that HSA future that AMD has been promising us for close to a decade now!
|
|
#
?
Sep 18, 2017 17:25
|
|
|
SwissArmyDruid posted:http://techreport.com/news/32558/rumor-amd-raven-ridge-apu-scores-appear-on-geekbench quote:In every game-like test, the Ryzen 7 2700U posts double or more the performance of the Bristol Ridge chip. That's despite the fact that the two processors have similar scores in the "ALU 2" GPU compute test.  Construction cores have been holding back the iGPU since forever, there is no way to make that kind of generational leap otherwise, not from what we've at all seen from Vega.
|
|
#
?
Sep 18, 2017 18:00
|
|
|
I want to believe so bad.
|
|
#
?
Sep 19, 2017 03:21
|
|
|
can't wait for a bastardized 4-6 core version of one of those to show up in the PS5 and Xbone Z
|
|
#
?
Sep 19, 2017 03:25
|
|
|
Anime Schoolgirl posted:can't wait for a bastardized 4-6 core version of one of those to show up in the PS5 and Xbone Z Consoles wouldn't go back from the 8 cores they have now. At the very least the Xbox Aleph-4 is going to keep an 8 core minimum cuz Microsoft is all about the backwards compatibility stuff. And then Sony would be hard pressed to keep parity with a 4-6 core variant of the same hardware.
|
|
#
?
Sep 19, 2017 03:30
|
|
|
Well... now that AMD has come back to the land of sanity and re-embraced SMT, I'm thinking a 4c/8t or 6c/12t part would still be a net step up. Remember: Construction cores may have been polished to a mirror sheen, but they are still garbage.
SwissArmyDruid fucked around with this message at 04:06 on Sep 19, 2017 |
|
#
?
Sep 19, 2017 04:03
|
|
|
Like, let me reiterate, double the performance of Bristol Ridge A12-9800E (about on par with desktop A8-9600) is mindblowing. TechEpiphany has a playlist of a A8-9600 running a variety of games. Based on a quick run through, even demanding modern games would run on Low, 1080p60 for Raven Ridge, you might get away with medium on some less demanding ones and older games should run fluently. That's also not even the highest end SKU AMD will be offering, it's the midrange one. I think the biggest issue will remain though, and that's the occasional stutter and screen tearing, so a quality Freesync implementation would be a must.
|
|
#
?
Sep 19, 2017 04:07
|
|
|
SwissArmyDruid posted:Well... now that AMD has come back to the land of sanity and re-embraced SMT, I'm thinking a 4c/8t or 6c/12t part would still be a net step up. Remember: Construction cores may have been polished to a mirror sheen, but they are still garbage. Yeah but the games and especially the emulation packages for 360 titles rely on having 8 full cores (or rather 6 or 7 exclusive cores with the main system holding a core for its own) to work against, and trying to shove that into hyperthreading is unlikely to work out too well for that. If you're OK with breaking compatibility then sure you could do a faster system on your computer 4c/8t, but the compatibility issue would make it a no go for MS.
|
|
#
?
Sep 19, 2017 04:10
|
|
|
SwissArmyDruid posted:http://techreport.com/news/32558/rumor-amd-raven-ridge-apu-scores-appear-on-geekbench I came as quick as I could. 💦 Just some of the stuff that's cropping up around the net is the bench reports clocks at boot, so expecting much higher results based on clocks is probably misguided. That said, hopefully some of my suspicions about Vega (designed for mobile hence weird bottlenecks) hold true, and that we'll be able to get an enterprise grade laptop with 1080p30+ performance, Freesync 2, and have a number pad. And really goosing the voltage in a SFF case might be my favorite  for a little while, especially since dGPUs are so drat pricey these days.
|
|
#
?
Sep 19, 2017 05:45
|
|
|
NewFatMike posted:an enterprise grade laptop with ... a number pad. Oh poo poo you're one of "those people" Just get a USB numpad for those rare occasions you actually need it and don't poo poo up your keyboard with that crap Paul MaudDib fucked around with this message at 06:30 on Sep 19, 2017 |
|
#
?
Sep 19, 2017 06:24
|
|
|
Paul MaudDib posted:Oh poo poo you're one of "those people" I mean, on a 15" it doesn't require much space and that's kinda my preferred form factor. The numpad is so good for 3D modeling and that's kinda "the thing" I need my computer to do.
|
|
#
?
Sep 19, 2017 07:07
|
|
|
NewFatMike posted:I mean, on a 15" it doesn't require much space and that's kinda my preferred form factor. The numpad is so good for 3D modeling and that's kinda "the thing" I need my computer to do. 
|
|
#
?
Sep 19, 2017 07:48
|
|
|
A coalition forms! There are at least TWO people in this niche!
|
|
#
?
Sep 19, 2017 10:31
|
|
|
I just put together the following: Ryzen 1600 MSI B350M Mortar (not my first choice, but basically 1 of 2 MATX boards out there with SPDIF - come on board makers step it up!) G.Skill 3200mhz CAS 14 RAM (whatever the cheapest was) Build came together nicely, though I haven't worked on a CPU with pins in a while (probably not since my old Phenom II X2!) so stress-free tasks like spreading the TIM made me nervous to bend a pin. Everything booted on the first shot, so I installed Windows. The install was ridiculously fast, maybe 10 minutes or so. I loaded some updates and grabbed the newest BIOS for the motherboard. Flashing that went well too - the stock BIOS was from March and didn't detect XMPP for the RAM properly. The new BIOS from late July found the correct profile so I dialed it in, booted up and ran 30 mins of Prime95 without any errors. Later today I'll try my hand at overclocking. Some observations: 1) Ryzen is crazy fast. I was worried because the 3770k that I was replacing technically has better single-thread performance, but Ryzen feels quite a lot snappier. It isn't just a placebo because I did a clean install of Win 10 on the 3770k about a month ago. Ryzen feels faster at stock speeds than the 3770k felt overclocked to 4.4ghz. 2) Ryzen runs insanely cool. I have a Corsair H80i V2 that used to keep the 3770k at 40C idle / 85C at worst under Prime95. At stock clocks the Ryzen idles at 31C and hit 51C at worst on Prime95. I'm eager to see how far I can push it on stock-ish volts! bobfather fucked around with this message at 11:34 on Sep 19, 2017 |
|
#
?
Sep 19, 2017 11:32
|
|
|
I bent a few pins installing my Ryzen 3, despite taking quite a bit of care doing so. Not sure when it happened, easy enough to fix with a scalpel though.
|
|
#
?
Sep 19, 2017 12:31
|
|
|
bobfather posted:Build came together nicely, though I haven't worked on a CPU with pins in a while (probably not since my old Phenom II X2!) so stress-free tasks like spreading the TIM made me nervous to bend a pin. It's possible - likely, even - that I'm a dipshit who does stupid wrong stuff but if I have to apply thermal compound to a CPU with pins I do it when it's clamped in the socket. quote:The install was ridiculously fast, maybe 10 minutes or so. I know right? Remember when that'd take over an hour? Mine was USB3 to NVMe and finished copying while I was downstairs grabbing a beer. Couldn't have been more than five minutes.
|
|
#
?
Sep 19, 2017 16:03
|
|
|
Munkeymon posted:It's possible - likely, even - that I'm a dipshit who does stupid wrong stuff but if I have to apply thermal compound to a CPU with pins I do it when it's clamped in the socket. I think people forgot how to apply TIM to PGA CPUs, but that was the only way I've ever done it or heard of doing it going back to the mid 90s.
|
|
#
?
Sep 19, 2017 16:05
|
|
|
I want to know if anyone applies paste while it's not in the socket.
|
|
#
?
Sep 19, 2017 16:12
|
|
|
GRINDCORE MEGGIDO posted:I want to know if anyone applies paste while it's not in the socket. Ok, that's what it meant. I was really, really confused. I'm still confused.
|
|
#
?
Sep 19, 2017 16:14
|
|
|
Munkeymon posted:It's possible - likely, even - that I'm a dipshit who does stupid wrong stuff but if I have to apply thermal compound to a CPU with pins I do it when it's clamped in the socket. It is literally inconceivable to me, a fellow frittata, that anyone would apply thermal compound to a CPU that is not firmly in its home.
|
|
#
?
Sep 19, 2017 16:53
|
|
|
I apply TIM to a CPU not in the socket. But I also don't just put a rice-sized blob of TIM in the middle of the heatspreader and hope the heatsink will spread the TIM for me. My steps to applying TIM: 1. Remove CPU from socket 2. Clean old TIM with a lint-free cloth with a bit of rubbing alcohol 3. Apply rice-sized (or for Ryzen more like pea size) blob of TIM onto middle of heatspreader 4. Use a plastic spreader (I like to use old credit cards) to spread an even layer across the entire heatspreader 5. Install in socket, install heatsink Edit: the only reason I do it this way is that I don't trust the heatsink to spread the TIM for me. If anyone can convince me this is a fallacy and my thinking is wrong, I'd have no trouble switching to just putting a blob of TIM on a socketed CPU.
|
|
#
?
Sep 19, 2017 17:32
|
|
|
bobfather posted:I apply TIM to a CPU not in the socket. Why don't you try the heatsink spreading method on a chip, let it do a few hot/cold cycles, and then take the heatsink off and look at the paste
|
|
#
?
Sep 19, 2017 17:50
|
|
|
According to https://www.pugetsystems.com/labs/articles/Thermal-Paste-Application-Techniques-170/ the X shape is the best CPU paste spreading method. But always with the CPU in the socket.
|
|
#
?
Sep 19, 2017 18:22
|
|
|
Volguus posted:According to https://www.pugetsystems.com/labs/articles/Thermal-Paste-Application-Techniques-170/ the X shape is the best CPU paste spreading method. But always with the CPU in the socket. 
|
|
#
?
Sep 19, 2017 18:32
|
|
|
Volguus posted:According to https://www.pugetsystems.com/labs/articles/Thermal-Paste-Application-Techniques-170/ the X shape is the best CPU paste spreading method. But always with the CPU in the socket.  
|
|
#
?
Sep 19, 2017 18:46
|
|
|
|
| # ? Apr 25, 2024 20:19 |
|
|
Booyah the smooth spread is close to the best, with the best (X) being something insane that nobody has ever really recommended ever. Out of curiosity, what are the justifications for applying paste while in the socket? I can see a few immediate advantages: 1. No risk of bent pins (not applicable to any Intel made in the last 10 years) 2. No risk of dropping the processor I can also see some limitations: 1. Small cases are tough to work in 2. No ability to get a nice smooth layer (somewhat obviated by the fact that the X shape seems to work so well) 3. Most TIM is semi-conductive, so getting it on motherboard surfaces can be harmful if you don't realize you've made the mess
|
|
#
?
Sep 19, 2017 18:48
|
|