|
Lambert posted:The exploits are relevant because all the necessary mitigations reduce the performance of the CPU a great deal. You can turn the mitigations off, even with them on AMD isn't catching up on single thread performance, and browsers have internal patches for the very small number of cases where theoretically you could use one of these exploits (in practice you straight up cannot use them). They're a huuuge deal for datacenters, and Rome is going to take a giant poo poo on Intel in that segment, but for home power users it's still going to be Intel on top and AMD producing a cheaper, slower, viable alternative. Based on what we've seen so far, if you're in the market for a 9900k you should still buy one, if you're looking below that you should be waiting for real Zen2 benches/real world prices to decide. K8.0 fucked around with this message at 23:36 on May 27, 2019 |
# ? May 27, 2019 23:33 |
|
|
# ? May 17, 2024 15:41 |
|
K8.0 posted:Based on what we've seen so far, if you're in the market for a 9900k you should still buy one, if you're looking below that you should be waiting for real Zen2 benches/real world prices to decide. Correct me if I'm wrong, but haven't we seen that the 3800x matches the 9900k in single thread and the 3900x beats it by 1%? The 9900ks will be another story, but if you don't already have a 9900k, it seems like the 3900x is the better option.
|
# ? May 27, 2019 23:51 |
|
But I want a pcie 4 M2 drive.
|
# ? May 27, 2019 23:52 |
|
Dramicus posted:Correct me if I'm wrong, but haven't we seen that the 3800x matches the 9900k in single thread and the 3900x beats it by 1%? The 9900ks will be another story, but if you don't already have a 9900k, it seems like the 3900x is the better option. Yeah but AMD picks pretty favorable benchmarks for their comparisons, it looks really good but I'd wait for reviews before making any purchasing decisions. I could see the 3900X beating the 9900k in ST but there's still questions about how the dual chiplet design affects gaming performance and such.
|
# ? May 27, 2019 23:53 |
|
GRINDCORE MEGGIDO posted:But I want a pcie 4 M2 drive. I want one that uses the 21100 form factor, so I can get me a 2+ TB PCI3 4 drive.
|
# ? May 28, 2019 00:04 |
|
GRINDCORE MEGGIDO posted:But I want a pcie 4 M2 drive. https://www.youtube.com/watch?v=nkrHGnSGrk4
|
# ? May 28, 2019 00:08 |
|
K8.0 posted:With the enormity of the spike in motherboard prices, AMD will only really be beating Intel as a value proposition by having more plentiful and appealing midrange options. Only true if require PCIE 4.0. You can ostensibly go buy a $70 B450 mobo and drop a 3000 series proc in it on 7/7/19 if you'd like. It'll be interesting to see if there's any gotchas in this scenario. Also, I'll be interested to see how X570 motherboard street prices look once the initial market demand it met.
|
# ? May 28, 2019 00:12 |
|
Any x570 mini itx boards with 10 Gb ethernet spotted? Really looking forward to upgrading my power hungry i7 5960x itx build...
|
# ? May 28, 2019 00:34 |
|
Deathreaper posted:Any x570 mini itx boards with 10 Gb ethernet spotted? Really looking forward to upgrading my power hungry i7 5960x itx build... 10 gbit Ethernet is still extremely rare on consumer motherboards. As in, there are only like three boards I know of that have it at all, and the only one that is even remotely reasonably priced is the Asrock Taichi Ultimate. There's a handful of boards with 2.5 or 5 gbit, but I tried looking just now and could not find even a single consumer ITX board with anything beyond 1 gbit on the market today. Maybe this will change with PCIe 4 but it's too early to tell. TheFluff fucked around with this message at 01:01 on May 28, 2019 |
# ? May 28, 2019 00:57 |
|
PC LOAD LETTER posted: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. Wait, what now? I thought Rome was confirmed to support AVX-512. It's half-rate support (like how Zen1 emulated AVX2 with two uop'd instructions) but it supports it in the sense that you can run instructions. Which is actually an important thing since AVX-512 fills in some holes in the existing sets (eg AVX2 has a "gather" instruction but no "scatter" instruction) as well as uop masking which allows a GPU-like coding approach. Right now it's not widely used in consumer code but that's largely because only Skylake-X and that one oddball cannon lake NUC actually support it. No sense writing code paths nobody can execute. Once it hits the desktop in Ice Lake it will start getting used over time... just like a lot of games nowadays won't run without at least AVX1 support. Is there confirmation that this is not in the consumer processors?
|
# ? May 28, 2019 01:00 |
|
OhFunny posted:Can you drop a Ryzen 3000 into a B450 or X470 and update the BIOS so it works or do you need a Ryzen 1000 or 2000 series CPU in there first? Many board manufacturers have already shipped BIOS that supports these chips at least in some basic form already. I will say though that the last time we were in this spot that ASRock/Newegg beat everybody in getting stock to shelves.
|
# ? May 28, 2019 01:09 |
|
NewFatMike posted:The thing I'm really curious about with PCIe 4.0 is how or whether it will impact IOMMU groupings. This of course comes with the caveat that you only get 4.0 speeds if you buy a 4.0 GPU... right now they don't exist and until NVIDIA catches up the only GPU line on the market that supports 4.0 will top out at 2070 performance. Which would have been fine with 3.0x8 speeds anyway (as would pretty much any other GPU that exists short of a 2080 Ti). I don't know if the switch chips exist yet but I'd like to see a Supercarrier style motherboard with fuckloads of PEX/PLX switch chips on it. Or, y'know, just Threadripper. At least Su confirmed it's coming at some point here. GRINDCORE MEGGIDO posted:But I want a pcie 4 M2 drive. A while ago I realized the CPU-direct M.2 slot would be pretty cool for a U.2 Optane SSD. Those really don't like being behind a PCH and right now you are eating 8 (or 4+4) lanes from your GPU. Which isn't the end of the world in reality but kinda sucks a bit if you're trying to do the cool high-end 9900K build or whatever. Also, it means that a 3-slot GPU is pressing right up against whatever's in the next slot. Consumer platform just doesn't have great expansion, unfortunately. M.2 to U.2 kinda gets around that by giving you another slot that doesn't gently caress with airflow too much. (I guess the other option would be to pony up and go custom loop on the GPU so that airflow doesn't matter) Paul MaudDib fucked around with this message at 01:25 on May 28, 2019 |
# ? May 28, 2019 01:11 |
|
So this chart from AMD: Is that supposed to mean B450 and X470 boards don't need a Bios update to run Ryzen 3000?
|
# ? May 28, 2019 02:33 |
|
Drakhoran posted:So this chart from AMD: No, it means not all 350/70 boards are guaranteed to have Zen2 support.
|
# ? May 28, 2019 02:37 |
|
I was independently curious and found this. https://wccftech.com/amd-ryzen-3000-series-support-on-existing-am4-boards-confirmed-by-asrock/
|
# ? May 28, 2019 03:13 |
|
Paul MaudDib posted:This of course comes with the caveat that you only get 4.0 speeds if you buy a 4.0 GPU... right now they don't exist and until NVIDIA catches up the only GPU line on the market that supports 4.0 will top out at 2070 performance. Which would have been fine with 3.0x8 speeds anyway (as would pretty much any other GPU that exists short of a 2080 Ti). Threadripper 3000 series completely submerged in mineral oil. It's the only way to get the form factor smaller. Riser cables are probably geficht for PCIe 4.0, right?
|
# ? May 28, 2019 03:15 |
|
Paul MaudDib posted:Wait, what now? I thought Rome was confirmed to support AVX-512. AMD's not saying until Rome launch. From an interview with Mark Papermaster last November (link 2 pages back):
|
# ? May 28, 2019 03:16 |
|
Will see any third-party benchmarks before the launch? And I wonder if Microsoft is going to unveil a Zen 2 powered Xbox at E3.
|
# ? May 28, 2019 03:39 |
|
15GB/s
|
# ? May 28, 2019 03:48 |
|
spasticColon posted:Will see any third-party benchmarks before the launch? Cerny came out and said that ps5 will be a bespoke 8 core Zen 2 chip with Navi based video that supports ray tracing, with 3d positional audio to boot. I assume that the same chip will be in the Xbox as well. https://www.wired.com/story/exclusive-sony-next-gen-console/
|
# ? May 28, 2019 03:49 |
|
NewFatMike posted:15GB/s One of the speculated reasons for all the boards having fans is that running nvme raid at those speeds might make things toasty.
|
# ? May 28, 2019 04:04 |
|
spasticColon posted:I'm sorry but this feels like a wet fart to me or maybe I just bought into the hype too much. Speaking of which, I have a 2700X paired with a high-end MSI X470 mobo. News so far of the current clocks is a bit of a letdown, as I was hoping for at least one of those Zen 2 chips to hit 5.0Ghz. Moving from a 2700X to a 3800X doesn't seem worthwhile, but a 3900X seems like a different story. I'll probably just wait until the pricing comes down a little, because that poo poo's verging on Threadripper territory atm.
|
# ? May 28, 2019 04:19 |
|
Is AMD going to stick with the current socket for Zen3/4...?
|
# ? May 28, 2019 04:26 |
|
Ragingsheep posted:Is AMD going to stick with the current socket for Zen3/4...? I believe they've said "through 2020" which would indicate that we perhaps get a 4000 series Zen 2+ on AM4 and then Zen 3 is AM5 or whatever.
|
# ? May 28, 2019 04:34 |
|
I think the general consensus is that we will get a new consumer socket when DDR5 is ready for consumers. If that happens in 2020 then this will probably be the last generation of AMD CPUs on AM4. If DDR5 is not ready until 2021 then we will probably get one more generation. That said, having the I/O die separate from the cores does give AMD some flexibility so we may get a transitional CPU generation released for both AM4 and AM5 but I wouldn't count on that.
|
# ? May 28, 2019 04:51 |
|
NewFatMike posted:15GB/s That 100 GbE looking like a bottleneck instead of a feature, all of a sudden. SwissArmyDruid fucked around with this message at 05:17 on May 28, 2019 |
# ? May 28, 2019 04:57 |
|
GRINDCORE MEGGIDO posted:But I want a pcie 4 M2 drive. https://pcper.com/2019/05/corsair-force-series-mp600-pcie-gen4-nvme-ssd/ Hell yeah, matches my case, cooling and PSU.
|
# ? May 28, 2019 04:57 |
|
Is there a common use case where current NVMe drives are a major bottleneck?
|
# ? May 28, 2019 05:06 |
|
Dramicus posted:I don't think there's a reason to get concerned over VRM's yet. If you're not overclocking and keep things stock with PBO off yeah I'd think you'd be fine on almost (there were some really cheap rear end ones that struggled with the 1800X or 2700X at stock from what I recall) any AM4 mobo even with the 16C Zen2 since it'll probably still have a stock 105W TDP. Many people here seem to be interested in either overclocking the 8C, 12C, or 16C Zen2's and the VRM is gonna matter with those though and that is why you're seeing more posts about it right now. So right now its looking like anyone with a high(er) end X370/X470 mobo should be fine after a BIOS update. Its the people who got some of the mid-range-ish B350/B450 mobo's who might run into some real issues if they want to OC 8C, 12C, or 16C Zen2. BeastOfExmoor posted:Is there a common use case where current NVMe drives are a major bottleneck? I don't think they really bottleneck so much as the random read and write speeds, especially at low QD, still aren't all that great and RAID'ing m.2 SSD's is potentially a way to get you some significant improvement there. Dramicus posted:One of the speculated reasons for all the boards having fans is that running nvme raid at those speeds might make things toasty. That is if you're running RAID off of the PCIe 4.0 lanes coming from the X570 chipset. That thing has to be ran off of the CPU PCIe lanes otherwise it can't boot. As Steve mentions its probably a special use case sort of thing and probably better to use on something like TR3. CaptainSarcastic posted:Edit: Also, how much performance difference would there be on the Ryzen 2 CPUs between 2400 and 3200 speed RAM? Just make sure the latency is at least somewhat low too otherwise you don't benefit as much. So something like CL14 DDR4 3200 is pretty much ideal if you don't mind spending more although OK latency DDR4 3000 can get you most of the benefits for a nice bit less cash. PC LOAD LETTER fucked around with this message at 05:38 on May 28, 2019 |
# ? May 28, 2019 05:26 |
|
CaptainSarcastic posted:I have wanted to buy or build a new system for a long time, but don't have much money to spare. I've got a parts list on NewEgg that is just over $900 built around a 2600x, and I see a Dell system built around 2700x that is on sale for about the same price. Either way I'll be moving my GeForce 1060 6gb from my current machine to serve as the GPU. I don't think that prices for the 2000 series will get dumped immediately after the 3 launch. There are already good sales going on (right now a plain 2600 has a -$20 code bringing it to just $145 on newegg). It wasn't until the past 6-9 months that we saw some crazy closeout deals on the last of the 1000 inventory where you were basically getting a free CPU with RAM and mobo purchase. So that probably won't happen again with the current stuff until zen 3 is around the corner. quote:Edit: Also, how much performance difference would there be on the Ryzen 2 CPUs between 2400 and 3200 speed RAM? Faster ram is kinda important for Ryzen 1 & 2, due to its internal hypertransport bus running at the same clock as ram. e: BeastOfExmoor posted:Is there a common use case where current NVMe drives are a major bottleneck? Common use case for consumer use? No. Most of the time NVMe isn't even that different from a sata SSD. For many people the only way they can see their NVMe perform better is running benchmarks. Common use case for high end work? Sure, big databases will eat all the IOps you serve them. Dunno who is running that on their desktop, but if they are then raiding up some NVMe sounds good. Klyith fucked around with this message at 05:44 on May 28, 2019 |
# ? May 28, 2019 05:31 |
|
Klyith posted:I don't think that prices for the 2000 series will get dumped immediately after the 3 launch. There are already good sales going on (right now a plain 2600 has a -$20 code bringing it to just $145 on newegg). It wasn't until the past 6-9 months that we saw some crazy closeout deals on the last of the 1000 inventory where you were basically getting a free CPU with RAM and mobo purchase. So that probably won't happen again with the current stuff until zen 3 is around the corner. I lucked into 80 off the 2700, so your mileage may vary. It's what pushed me over to AMD for sure over the i5, and I don't regret my choice.
|
# ? May 28, 2019 05:32 |
|
PC LOAD LETTER posted:I don't think they really bottleneck so much as the random read and write speeds, especially at low QD, still aren't all that great and RAID'ing m.2 SSD's is potentially a way to get you some significant improvement there. I asked this a while ago on some thread and I never really got a straight answer, but does dropping to (say) x2 or x1 affect IOPS that substantially? If the limitation is that the controller can only dial up the flash so quickly then shouldn't running it on fewer lanes not really affect performance? I probably would never notice anyway but I've seen boards with x2 slots (C232/C236 seem to have a lot of them) and I always wondered how badly that affects performance. tbh the fact that they're PCH slots probably affects things more... I have heard optane performs quite a bit worse on PCH than direct to CPU. Paul MaudDib fucked around with this message at 06:44 on May 28, 2019 |
# ? May 28, 2019 06:36 |
|
Paul MaudDib posted:I asked this a while ago on some thread and I never really got a straight answer, but does dropping to (say) x2 or x1 affect IOPS that substantially? My understanding is that IOPS is mostly limited by the flash controller of the SSD and perhaps to some extent the flash itself and not really the PCIe bandwidth. (edit) At least for most work loads. Stuff like booting is one of those things where peak stream bandwidth does actually matter a bunch. PC LOAD LETTER fucked around with this message at 07:08 on May 28, 2019 |
# ? May 28, 2019 07:06 |
|
https://www.pcworld.com/article/3397335/intel-boldly-claims-its-ice-lake-integrated-graphics-are-as-good-as-amds.html It should be noted that the article makes reference to a "3700W". No such product exists, and is probably a typo or misheard bit of information meant to refer to the 3700U, which is, as the article says, a 15W part. So, are AMD integrated graphics bad now, or did Intel just crank up the power to the point that whatever cooling they were using couldn't keep up and the AMD chip started throttling to make their own chips look good? :hmm: SwissArmyDruid fucked around with this message at 11:30 on May 28, 2019 |
# ? May 28, 2019 11:28 |
|
Wasn't that test where they gave the AMD platform like DDR4 2400 and the Intel one like DDR4 3700 or some poo poo too? I think the article mentions they did allow the AMD part to have a 25W TDP (I guess its variable for that part? not sure) at least but if they fudged on the RAM for the AMD system then I'd wait for more and better benchmarks before getting worked up. edit: yeah they used much faster RAM (LPDDR4X-3733) on the Intel Icelake U system: https://www.anandtech.com/show/14405/intel-teases-ice-lake-integrated-graphics-performance I'm shocked it didn't flat out curbstomp the AMD system more (it had 56% more bandwidth) with a disparity in system bandwidth like that for iGPU performance. PC LOAD LETTER fucked around with this message at 11:47 on May 28, 2019 |
# ? May 28, 2019 11:43 |
|
I'm having trouble finding concrete information on that end, actually, I was hoping you guys had seen better reporting. AMD is sucking all of the oxygen out of Computex so hard, every bit of Intel reporting I've seen is mostly, "....oh, and btw, Intel has something that does 5 GHz, and there might be some new graphics?"
|
# ? May 28, 2019 11:45 |
|
3700u officially only supports up to DDR4-2400.
|
# ? May 28, 2019 11:46 |
|
Arzachel posted:3700u officially only supports up to DDR4-2400. edit: Remember they were testing iGPU performance and bandwidth has a huge impact on that. Come to think of it testing with DDR4 2400 for both systems would've made a lot more sense anyways since that sort of cheap stuff is what you tend to get in the cheap and/or low power systems that the 25-15W APU's go in anyways. You'll probably never see those systems come with something like LPDDR4 3700 stock. edit: AMD pulled some stupid benchmarking shenanigans too (that synthetic PCIe 4.0 bandwidth test vs a 2080ti) so they're not really coming off all that clean this round either but this is clearly some BS on Intel's part. PC LOAD LETTER fucked around with this message at 11:57 on May 28, 2019 |
# ? May 28, 2019 11:48 |
|
SwissArmyDruid posted:So, are AMD integrated graphics bad now, or did Intel just crank up the power to the point that whatever cooling they were using couldn't keep up and the AMD chip started throttling to make their own chips look good? :hmm: I'm in the camp of people who expect Intel to drop some unbelievable products over the next few years as they get onto actual new processes. I'm pretty sure zen2+navi igpu will out perform that ice lake igpu by quite a bit. PC LOAD LETTER posted:edit: AMD pulled some stupid benchmarking shenanigans too (that synthetic PCIe 4.0 bandwidth test vs a 2080ti) so they're not really coming off all that clean this round either but this is clearly some BS on Intel's part. Khorne fucked around with this message at 13:48 on May 28, 2019 |
# ? May 28, 2019 13:41 |
|
|
# ? May 17, 2024 15:41 |
|
PC LOAD LETTER posted:Wasn't that test where they gave the AMD platform like DDR4 2400 and the Intel one like DDR4 3700 or some poo poo too? Hahaha, I thought you were exaggerating but they really did use LPDDR4X-3733. If that's the best they can do with 50% more bandwidth then rip Intel.
|
# ? May 28, 2019 14:33 |