|
surf rock posted:Thank you for any clarity you can provide! The high end coffee lake Intel CPUs have a competent boost as well and don't have to be overclocked. They do benefit a bit more from overclocking than AMD if you want every last ounce of performance. As far as difficulty, overclocking isn't hard and will eat up 1-4 hours of your time. And then it's set and forget. Overclocking will be:
From those steps you can say "Great, I got 90% of the value of overclocking", set the settings you used in your bios, and you're done. Or you can repeat the last two steps by tinkering with the settings to try and consume a bit less power or go a bit faster. If it's not stable at safe settings, which can be for any number of reasons, you either dial back the settings slightly or throw a little more voltage at it until it is stable. There's also the option to overclock memory. To avoid doing it, you just enable the XMP profile in your bios to use the manufacturer overclock. If you want to tinker, I'd honestly just use a tool for this. You input the CPU, motherboard, and memory modules you have into the tool. It gives you settings to use. You set those settings in your bios. You run memtest. If your computer is stable then you're done. If it's not, you need to either use slightly more voltage or slightly more lenient timings. Khorne fucked around with this message at 14:39 on May 27, 2019 |
#
?
May 27, 2019 14:34
|
|
|
|
| # ? Apr 26, 2024 01:26 |
|
|
12 core is still pretty cool. I still feel dumb for buying the 9900k after all of the exploits. I mean I might have to turn off hyper threading. I just wished the lower end models hit 5 ghz
|
|
#
?
May 27, 2019 14:36
|
|
|
You all rock, thank you for sharing your knowledge! Khorne, is there any program you'd recommend for doing the CPU overclocking test of stability and temperature?
|
|
#
?
May 27, 2019 14:39
|
|
|
Also, I'm defiantly going to buy a few hundred shares of AMD.
|
|
#
?
May 27, 2019 14:39
|
|
|
Cygni posted:Two Ryzen 5's are in AMD's press release, AnandTech has some details on them: So wait, what's the difference between 2600 and 3600 (non-X and X)? 2600 : 3.4 Ghz base, 3.9 Ghz boost 3600 : 3.6 Ghz base, 4.2 Ghz boost 2600X : 3.8 Ghz base, 4.2 Ghz boost 3600X : 3.9 Ghz base, 4.4 Ghz boost same number of cores. The clockspeed improvement looks marginal at best (and the 3600 seems to be even better value than the 2600, vs. each one's X counterpart) are those IPC improvements going to make the new ones really competitive? If it has 30% more IPC, will it actually be 30% faster at the same speed (so considering the extra 100-200Mhz, say 35% faster)? If it's less than that I think I'll be keeping my 2600X  or maybe go for the 3700X or 3900X when prices start coming down if I feel the need for extra cores
|
|
#
?
May 27, 2019 15:07
|
|
|
Interesting that AMD's site has the 3700X listed as using the big Prism cooler, despite the 65W TDP. Either it's going to be much closer to 100W in practice or they'll be using the extra thermal headroom to run the fans at a more tolerable speed under load.
|
|
#
?
May 27, 2019 15:11
|
|
|
Kinda disappointed in the clockspeed bumps but I guess we'll have to see how it shakes out in real world benchmarks from third parties.
|
|
#
?
May 27, 2019 15:18
|
|
|
TorakFade posted:are those IPC improvements going to make the new ones really competitive? If it has 30% more IPC, will it actually be 30% faster at the same speed (so considering the extra 100-200Mhz, say 35% faster)?
|
|
#
?
May 27, 2019 15:21
|
|
|
Guys even AMD said Zen2 was ~15% more IPC over Zen+. https://hothardware.com/photo-gallery/NewsItem/48222?image=big_amd-zen-2-ipc-lift.jpg&tag=popup I don't really know where 30%+ more IPC came from. Is it you're confusing Zen2 numbers with the "RDNA"/Navi numbers? I think that is supposed to be a ~25% IPC increase. 30% more IPC over Zen+ would be friggin' amazing for a x86 CPU. For reference I think Sandybridge was about ~20% more IPC over Westmere and most people at the time were pretty wowed by that. edit: or do you mean application specific IPC?
|
|
#
?
May 27, 2019 15:32
|
|
|
PC Load Letter pretty much covered most of this, but I'll just add some thoughts. quote:So, I've been pestering the Intel thread with super-basic questions because I'm probably going to get a new desktop this year. I'm anticipating that I'll pay my brother's friend to build it. I have no idea what your physical abilities, etc. are, but if its at all possible for you to build it yourself, possibly with a little bit of supervision from your brother's friend or whatever, I'd encourage you to do it. It's a fairly straightforward process. There are a few parts that can be a little confusing or challenging the first time, but it's really empowering and educational to know how everything's connected in your system. quote:4.) There are so many performance variations to compare with processors... IPC, clock speed, core count, simultaneous multi-threading or not (and whether or not that's a feature or a bad thing in the first place) are all stuff I've come across. Is there a good resource out there that says "For this set of use cases, clock speed is the thing that matters most, and for this set of cases, you really want simultaneous multi-threading..." and breaks it down like that? This can get really confusing, but you essentially want to avoid comparing clock speed between different types (intel Coffee Lake vs Zen 2) or even generations (Zen vs Zen 2) of chips because it's not a particularly meaningful comparison. You want to find real world benchmarks examples of performance. There are tons of websites that will have dozens of benchmarks for new chips when they're released. It can be really easy to get bogged down in minutia with these, but in general you're looking for broad generalizations of trends of where each chip falls in your use case scenario. The PC building thread will be your best friend here. Lurk it to try to understand why people suggest buying the parts they do and ask a ton of questions when you're putting together your build.
|
|
#
?
May 27, 2019 15:34
|
|
|
Are we to assume no major difference in AVX support and performance beyons pure IPC gains?
|
|
#
?
May 27, 2019 16:01
|
|
|
leaks said avx2 support e: avx-512? whatever the next level is..
|
|
#
?
May 27, 2019 16:10
|
|
|
Zen2 is supposed to properly support 256 bit AVX2 instructions (so no more breaking up the 256 bit operations between 2 FPU's) and it won't have any special clock speed limitations associated with it either. I don't believe Zen2 is supposed to support the 512 bit AVX instructions though. So a nice improvement there but still not on par with Intel. Personally I don't care too much about the 256 or 512 bit AVX support because there is still heaps of stuff that doesn't use either in the desktop space and that doesn't seem to be changing much for a while yet. If you're in the HPC market its a different story. Those 512 bit AVX instructions help lots for those work loads.
|
|
#
?
May 27, 2019 16:12
|
|
|
Yeah Zen2 doesn't have AVX-512 but neither does Intel in the consumer space, Coffee Lake doesn't have it. It's only on Skylake-X for now and eventually Cannon Lake/Ice Lake.
|
|
#
?
May 27, 2019 16:18
|
|
|
ufarn posted:Are we to assume no major difference in AVX support and performance beyons pure IPC gains? To the contrary Zen2 AVX performance is 2X over Zen+. This was plainly stated at the Rome announcement, and not-so-plainly stated again last night. It's what this slide means:  I don't remember the exact wording, but last night Dr. Su used one of the more polite wads of marketspeak I've ever heard in order to say "We unfucked AVX performance, so yay" without using any of those words. EDIT: here's Ian Cuttress and Mark Papermaster talking about Zen 2's FP units, back in November, at the Rome reveal  Source: https://www.anandtech.com/show/13578/naples-rome-milan-zen-4-an-interview-with-amd-cto-mark-papermaster mdxi fucked around with this message at 16:59 on May 27, 2019 |
|
#
?
May 27, 2019 16:37
|
|
|
surf rock posted:Khorne, is there any program you'd recommend for doing the CPU overclocking test of stability and temperature? I only got into OC a year ago and have only OCed two chips, but I use ROG RealBench, personally. There are more strenuous tests designed to send you chip into a meltdown, and I know my 1600 on it's air cooler couldn't operate under those, but they don't simulate real world conditions at all.
|
|
#
?
May 27, 2019 16:41
|
|
|
PC LOAD LETTER posted:Guys even AMD said Zen2 was ~15% more IPC over Zen+. No I'm just dumb  and heard that 30% figure somewhere... maybe a slide earlier on? v vanyway say it's +15% IPC, if that's actually 15% faster and has a few extra Mhz on top (and perhaps more headroom to overclock than the current 2000 series which is rather bad at that) even going from a 2600x to a 3600x could be a decent upgrade, but I guess we have to see it in real world situations to say.
|
|
#
?
May 27, 2019 17:39
|
|
|
Has AMD said anything about changes infinity fabric for zen 2? It was IF that caused problems with RAM speeds on zen, right? Something about the processor or something being locked to IF speeds? I don't know much about these smaller details but all this amd stuff got me to remembering what problems zen faced before. Did that get fixed with zen+? Separate dumb questions: Since I gather IPC and clock speeds are different, would you be able to figure out equivalent GHz of an older generation by multiplying the IPC gain percent by the clock speed of the newer generation? Or are there smaller details in there that would make that wildly inaccurate? I would assume it would be or else everyone would have been spouting off numbers otherwise.
|
|
#
?
May 27, 2019 17:41
|
|
|
TorakFade posted:No I'm just dumb I believe you're thinking of this slide where the 3800X is 30%+ in some games over the 2700X. We need to see some more methodology here though to know how accurate that is vs marketing hype. 
|
|
#
?
May 27, 2019 17:45
|
|
|
Haven't had the time to go through the material, is there any word on how the CCX distribution is done on the announced models? 4+4 or 8+0 for the 8 core? 6+6 or 8+4 for the 12 core?
|
|
#
?
May 27, 2019 17:51
|
|
|
This is really dumb math that's way too trusting of marketing numbers, but if I make a percentage of the clockspeed increase first-gen Ryzen and third-gen Ryzen, and then slap on a flat 15% increase for IPC and multiply, a 3600 is roughly where a 1600 would be at 4.7ghz (keep in mind that you rarely see higher than 3.9 on a 'real' 1600.) A 3600X would be 4.9 ghz, and the 3700X at 1700 clockspeeds should be 5ghz. This is just a dumb exercise and not buying advice. Wait for actual reviews.
|
|
#
?
May 27, 2019 17:53
|
|
|
Broose posted:Has AMD said anything about changes infinity fabric for zen 2? It was IF that caused problems with RAM speeds on zen, right? Something about the processor or something being locked to IF speeds? I don't know much about these smaller details but all this amd stuff got me to remembering what problems zen faced before. Did that get fixed with zen+? I don't know where this info came from but according to Wikichip the bandwidth per link is doubled (it says 2.3x but I think that it is assuming a higher memory clock speed). IF is still locked at RAM speeds but with double the bandwidth and faster memory support I would assume this would be less of an issue. https://en.wikichip.org/wiki/amd/microarchitectures/zen_2#Key_changes_from_Zen.2B Craptacular! posted:This is really dumb math that's way too trusting of marketing numbers, but if I make a percentage of the clockspeed increase first-gen Ryzen and third-gen Ryzen, and then slap on a flat 15% increase for IPC and multiply, a 3600 is roughly where a 1600 would be at 4.7ghz (keep in mind that you rarely see higher than 3.9 on a 'real' 1600.) A 3600X would be 4.9 ghz, and the 3700X at 1700 clockspeeds should be 5ghz. Yeah this slide was the most impressive marketing tidbit they showed IMO showing the sheer ST performance improvement from 1st gen Ryzen to 3rd gen Ryzen, hopefully other workloads see this type of gain. 
MaxxBot fucked around with this message at 18:00 on May 27, 2019 |
|
#
?
May 27, 2019 17:53
|
|
|
Broose posted:Since I gather IPC and clock speeds are different Yes. IPC is Instructions Per Clock, which means exactly what it says: per tick of the clock, X many instructions are handled by a core. IPC is important because (broadly speaking) it tells you about speed-ups due to architectural changes. Clock speed is: how fast are the ticks of the clock? Clock speed improvements (again, generally speaking) tell you about process improvements. You can have clock speed improvements wihout IPC improvements (usually described as a "speed bump" by the press). You can have IPC improvements without clock speed improvements. In each step, from Zen to Zen+, and from Zen+ to Zen 2, we've gotten both. quote:would you be able to figure out equivalent GHz of an older generation by multiplying the IPC gain percent by the clock speed of the newer generation? In a rough, "on paper" fashion, sure. This is exactly what AMD did to get the "2X performance between 1800X and 3900X" number that they put up on a slide last night. code:quote:Or are there smaller details in there that would make that wildly inaccurate? I would assume it would be or else everyone would have been spouting off numbers otherwise. Also yes, but that's the real world for you. Every workload varies. Boost varies with available voltage and thermal overhead. Chips within a batch vary tiny bits, even if binned into the same bucket. You can't meaningfully account for everything, and you sure can't market products that way. So manufacturers make generalizations (sometimes fair, sometimes bullshit) and call it a day. Then reviewers and/or invested users do focused benchmarks for the things they're more specifically interested in. I'm planning a couple weeks of testing to get a full characterization of all 3 generations of Ryzen CPUs in scientific computing, once I have my hands on a 3900X.
|
|
#
?
May 27, 2019 17:54
|
|
|
https://www.youtube.com/watch?v=yDA0zBR6MgA Looks like X570 boards are going to be more expensive than X470 and X370 boards were. GloFo is also making the mobo chips (did AMD end up designing the chipset?), so I wonder if that's how they're planning on satisfying the WSA through the next year.  the highest-end Gigabyte X570 is $600. Only the Gaming X board is under $200 ($170). The mITX board is $220, though, so good thing SFF isn't getting left behind even with the increased challenges PCIe 4.0 brings. e: 11-15W for chipset consumption confirmed, hence the fans.
|
|
#
?
May 27, 2019 17:54
|
|
|
|
|
#
?
May 27, 2019 17:56
|
|
|
Icept posted:Haven't had the time to go through the material, is there any word on how the CCX distribution is done on the announced models? 4+4 or 8+0 for the 8 core? 6+6 or 8+4 for the 12 core? https://www.anandtech.com/show/14407/amd-ryzen-3000-announced-five-cpus-12-cores-for-499-up-to-46-ghz-pcie-40-coming-77 According to Anandtech the 3900X is 6+6 and the 3800X/3700X are 8+0.
|
|
#
?
May 27, 2019 17:58
|
|
|
NewFatMike posted:https://www.youtube.com/watch?v=yDA0zBR6MgA Is this the kind of thing that does, in time, show up in every board? Hopefully excluding the fans someday? I’m one of those people that only notices changes in technology when the actual slot physically changes, so since AGP was replaced with PCIE I haven’t paid a whole lot of attention. AMD seemed to be very carefully saying little and showing only what made the tech look like Magic Juice that turns midrange graphics cards into premium supermen. I know that’s bullshit but am I really going to be paying $100 more for my motherboards in the future for this or what.
|
|
#
?
May 27, 2019 18:07
|
|
|
Broose posted:It was IF that caused problems with RAM speeds on zen, right? Zen+ was better than Zen1 at memory overclocking but still not that good. Zen2 is supposed to be better than Zen+ but I wouldn't expect it to overclock RAM as well as Intel either. Honestly though you don't need to OC the RAM that much for any of the Ryzen's so far. CL14 DDR4 3200 gets you about the peak benefit that RAM overclocking can give you with Zen1/+ for instance and that probably won't change much for Zen2. And cheaper CL15 DDR4 3000 gets you nearly all the benefit that CL14 DDR4 3200 would get you anyways BTW. Broose posted:would you be able to figure out equivalent GHz of an older generation by multiplying the IPC gain percent by the clock speed of the newer generation? Icept posted:Haven't had the time to go through the material, is there any word on how the CCX distribution is done on the announced models? 4+4 or 8+0 for the 8 core? 6+6 or 8+4 for the 12 core? NewFatMike posted:GloFo is also making the mobo chips (did AMD end up designing the chipset?), so I wonder if that's how they're planning on satisfying the WSA through the next year. Asmedia couldn't get PCIe 4.0 working in time so AMD rolled their own. Its supposed to be pretty much some IO stuff that was on die for Epyc shoehorned into a chipset. So far the word is that it only uses peak power if you try RAID'ing m.2 SSD's so maybe the fan will be off most of the time or at least barely spinning. OhFunny posted:According to Anandtech the 3900X is 6+6 and the 3800X/3700X are 8+0. Zen1/+ had 4 cores per CCX and inter CCX latency and bandwidth could be a issue for some work loads which is why a 8+0 CCX would've been real nice. PC LOAD LETTER fucked around with this message at 18:57 on May 27, 2019 |
|
#
?
May 27, 2019 18:49
|
|
|
NewFatMike posted:https://www.youtube.com/watch?v=yDA0zBR6MgA I bought one of the more expensive x470 boards and those prices are ridiculous.
|
|
#
?
May 27, 2019 18:50
|
|
|
fknlo posted:I bought one of the more expensive x470 boards and those prices are ridiculous.
|
|
#
?
May 27, 2019 18:59
|
|
|
Its there any advantage outside of PCI4 over using a 4XX one?I dont want to deal with the heat/extra wattage and price if there isn't more tangible benefits
|
|
#
?
May 27, 2019 19:02
|
|
|
What are you missing out on if you pair the 3900x w/ an x470 board? Seems like x570’s are going to be too rich for my blood. Shame they keep the processor costs down but inflate the mobo prices.
|
|
#
?
May 27, 2019 19:04
|
|
|
Cao Ni Ma posted:Its there any advantage outside of PCI4 over using a 4XX one?I dont want to deal with the heat/extra wattage and price if there isn't more tangible benefits PCIe lanes and perhaps somewhat better memory overclocking maybe? If you're just running 1x m.2 SSD and 1x video card with maybe a sound card and are fine with DDR4 3000/3200 I don't think you need to care much about that though. xgalaxy posted:What are you missing out on if you pair the 3900x w/ an x470 board? xgalaxy posted:Shame they keep the processor costs down but inflate the mobo prices. There are some good reasons to increase the mobo price (expensive chipset, expensive materials for PCIe 4.0, fan (lol)) but as Steve mentioned in the GN video demand is starting to play a factor too. Some of the high end Z390 mobo prices have been like this for a while now too. edit: interestingly at the end of this video here Buildzoid mentions that overclocking Zen2 will be more like overclocking the old AM3+ CPU's. I'm guessing he means the memory controller/uncore will need to be overclocked as well if you want to get a nice performance improvement but I'm not sure: https://www.youtube.com/watch?v=ZbFQWwKEDec PC LOAD LETTER fucked around with this message at 19:25 on May 27, 2019 |
|
#
?
May 27, 2019 19:09
|
|
|
Cao Ni Ma posted:Its there any advantage outside of PCI4 over using a 4XX one?I dont want to deal with the heat/extra wattage and price if there isn't more tangible benefits xgalaxy posted:What are you missing out on if you pair the 3900x w/ an x470 board? Some websites tested B350/X370 vs. B450/X470 performance when the 2000 series CPUs was released, I assume there will be performance comparisons with the new CPUs as soon as they're out.
|
|
#
?
May 27, 2019 19:15
|
|
|
I'm sorry but this feels like a wet fart to me or maybe I just bought into the hype too much. If the Zen 2 launch pushes the 2700X down to ~$200 I'm just going to get a 2700X and a B450 or X470 board. I don't need the latest and greatest for gaming anyway. I'm still gaming on a older-than-dirt 2500k.
|
|
#
?
May 27, 2019 19:20
|
|
|
As it stands right now, PCIe 4.0 is really a shot at Intel more than anything. Unless the upcoming batch of GPUs from AMD are also PCIe 4.0 (which is likely, to be fair), it's them shoving it in Intel's face that they have it implemented already. That said, the purpose of it being solely a shot at Intel won't last for long, because you know nVidia (and AMD, natch) will also have PCIe 4.0 ready GPUs around the bend now. It's easier to swap your GPU than your CPU, and given we're starting to see raytracing ones come out, well there you go.
|
|
#
?
May 27, 2019 19:21
|
|
|
orcane posted:I guess (some) X470 boards might lack the VRM setup to reach high boost/OC clocks with the 3900X but it's too early to tell. For reference according to Buildzoid (skip to around 7min) my X370 Taichi's VRM can do something silly like over 300A if you've got the cooling (LN2 probably required) for it for instance. iospace posted:As it stands right now, PCIe 4.0 is really a shot at Intel more than anything. For servers its a big deal.
|
|
#
?
May 27, 2019 19:22
|
|
|
For those concerned about the prices of x570 boards, just wait for the b550 boards to come out. They will absolutely be able to run all the chips to max and will certainly be cheaper. The X boards are for oc enthusiasts and have a corresponding price. The B series are for those of us who just want to run the chip and use xfr.
|
|
#
?
May 27, 2019 19:30
|
|
|
I think the motherboard vendors are going kinda overkill on the high end boards by dimensioning them for exotic-coolant overclocking of a 16C/32T part. For a more reasonable-for-home-use 6- or 8-core on ambient cooling you should still be fine with the $100-$150 X470 boards. The new fancy PCIe stuff is expensive too of course. e: also that ^^^^^^ spasticColon posted:I'm sorry but this feels like a wet fart to me or maybe I just bought into the hype too much. Wait for the benchmarks, as always.
|
|
#
?
May 27, 2019 19:32
|
|
|
|
| # ? Apr 26, 2024 01:26 |
|
|
Dramicus posted:For those concerned about the prices of x570 boards, just wait for the b550 boards to come out. If you can wait that long then fine. Otherwise a X470 isn't a bad option so long as you don't care much about PCIe 4.0 or need more lanes. TheFluff posted:I think the motherboard vendors are going kinda overkill on the high end boards by dimensioning them for exotic-coolant overclocking of a 16C/32T part. Guys high end AMD mobos are just getting the same treatment high end Intel boards have been getting for a while now. There are Z390 mobo's with $700+ "open box deal" price tags!! I don't think they're worth it but apparently someone is buying them.... PC LOAD LETTER fucked around with this message at 19:39 on May 27, 2019 |
|
#
?
May 27, 2019 19:36
|
|