|
BIG HEADLINE posted:Yet Intel can continue to eat poo poo because they have 15 fabs (8 major ones): https://www.intel.com/content/www/us/en/support/articles/000015142/programs.html Assembly/test is not fab. The facilities are built to very different specs. # of locations isn't really a great proxy for fab capacity either. Not all locations do logic. Of those that do, not all are on the latest node. Lastly, there is a vast difference in wafer volume at each facility. Fun fact: at one point, Intel CPUs were the single largest export from the state of New Mexico. Now it's just "computer chips" as a category 
|
#
?
Dec 17, 2020 21:39
|
|
|
|
| # ? May 31, 2024 21:48 |
|
|
I'm debating upgrading from a 7700K a 10900K (or maybe a 10850K), but I've read that the Z490 Mobos may not offer "full PCIe Gen 4 support", whatever that means. Are we even at a point where PCIe Gen 4 is actually used outside of gaming/general video editing? I'm wondering if I should wait on the 500 series mobo, or get the above now (if there's stock)? Edit: VVV thanks, I was leaning that way, but needed some confirmation. VVV Edit 2: vvv thanks for the additional replies. I might bite the bullet and upgrade now, as with how everything has gone the past year (Nvidia crap gpu stock, CP2077 a mess, supply issues for many pc components/consoles) waiting might just be more frustrating, and I’ll gain very little from it if I’m able to even find the new cpu/mobo in the new year. Kragger99 fucked around with this message at 23:17 on Dec 18, 2020 |
|
#
?
Dec 18, 2020 18:01
|
|
|
PCIe 4 is not really stressed by anything and is gonna have an extremely short life with PCIe 5 coming this year. If you are on the fence about upgrading, I would wait a few weeks for Rocket Lake.
|
|
#
?
Dec 18, 2020 18:42
|
|
|
The 10850K and 10900K don’t support PCIE 4 anyway. Some Z490 boards have PCIE 4 components but you’ll still need 11th gen Intel which will have PCIE 4 support.
|
|
#
?
Dec 18, 2020 22:35
|
|
|
Kragger99 posted:Are we even at a point where PCIe Gen 4 is actually used outside of gaming/general video editing? I'm wondering if I should wait on the 500 series mobo, or get the above now (if there's stock)? Short answer is: no. Even in gaming, PCIe 4 vs 3 is like a 1% difference at 4k, and 0% below that. Hell, there's usually less than a 5% difference dropping from PCIe 4 x16 down to PCIe 3 x8, which is 1/4th the bandwidth. As for SSDs, the difference would be 4GB/s vs 8GB/s, except that there literally aren't any SSDs that can do 8GB/s yet (7GB/s seems the cap for now), and it almost certainly wouldn't matter anyhow, even for video editing, because your processing capacity is likely going to be considerably lower than 7GB/s (or even 4GB/s) anyhow. Plus the 7GB/s drives are silly expensive so if you're already thinking with a budget you're gonna end up with a 3-4GB/s drive anyhow. Most people cannot tell the difference between a 500MB/s SATA SSD and a 4GB/s NVMe SSD, and most games don't load appreciably faster on them, either. The effective difference between PCIe 3 and PCIe 4 right now for anyone with a single SSD is basically zero. With multiple SSDs, it only could possibly matter a little if you were doing bulk data transfer from one SSD to the other with minimal processing between them, which isn't a particularly common daily task. tl;dr don't worry about PCIe 4. It's pretty pointless, and stands to remain pretty pointless for most home users for a good while.
|
|
#
?
Dec 18, 2020 22:47
|
|
|
How does the overall number of lanes compare in similar mobos with PCIe 3 and 4? Is the count similar, so you could in theory stick more devices without compromising their bandwidth; or is the total transfer rate the same? Or something in between maybe?
|
|
#
?
Dec 18, 2020 23:58
|
|
|
DrDork posted:Short answer is: no. Even in gaming, PCIe 4 vs 3 is like a 1% difference at 4k, and 0% below that. Hell, there's usually less than a 5% difference dropping from PCIe 4 x16 down to PCIe 3 x8, which is 1/4th the bandwidth. In the HPC/enterprise world, they hang a few hundred NVMe disks off SAS controllers, so they need all the bandwidth they can get.
|
|
#
?
Dec 19, 2020 00:25
|
|
|
mmkay posted:How does the overall number of lanes compare in similar mobos with PCIe 3 and 4? Is the count similar, so you could in theory stick more devices without compromising their bandwidth; or is the total transfer rate the same? Or something in between maybe? The lane counts from the cpu pretty much stay the dame by type (consumer=16-20, enterprise/workstation=40-128) regardless of speed. So the gen4 capable cpus just have twice the thruput capacity.
|
|
#
?
Dec 19, 2020 00:26
|
|
|
BlankSystemDaemon posted:All of this is for the consumer/prosumer/commercial side of things, but you're right. Agreed, but if you're trying to decide between a 10850k and an 11850k you're not a HPC/enterprise user to begin with. PCIe 4 seems to very much be a "if you actually need it, you already know you need it; if you have to ask, you don't."
|
|
#
?
Dec 19, 2020 01:42
|
|
|
DrDork posted:As for SSDs, the difference would be 4GB/s vs 8GB/s, except that there literally aren't any SSDs that can do 8GB/s yet (7GB/s seems the cap for now) Since it sounds like you think this might be a drive limitation, fyi it's not. There's several types of required overhead eating into that nominal 8 GB/s: the 128b/130b line coding, packet headers, and link layer maintenance packets which don't carry user data. It's enough to reduce the real max throughput of gen4 x4 to about 7 GB/s, assuming 256 byte packets (a reasonably common PCIe max packet size, at least on Intel systems). All these overheads are the same at gen3 speeds, so a gen3 x4 SSD shouldn't score higher than ~3.5 GB/s.
|
|
#
?
Dec 19, 2020 13:54
|
|
|
128b/130b line coding is better than 8b/10b line coding which we were stuck with for decades.
|
|
#
?
Dec 19, 2020 17:20
|
|
|
BlankSystemDaemon posted:128b/130b line coding is better than 8b/10b line coding which we were stuck with for decades. PCIe 3.0 actually also had 128b/130b encoding. So yeah, while a 4.0 drive will never be able to reach a full 8GB/s (or, technically, 7,880MB/s), between NVidia and Microsoft working on direct transfers from SSDs straight to GPU memory, and hardware decompression blocks being used in the next-gen consoles, it's possible we'll see some drives eventually that leverage that to allow for >7GB/s effective data transfer, which is what I meant by "for now." Ain't gonna see it anytime soon, though, if ever.
|
|
#
?
Dec 19, 2020 18:04
|
|
|
DrDork posted:PCIe 3.0 actually also had 128b/130b encoding. I don't think I ever saw anything outside of HPC gear that actually used it.
|
|
#
?
Dec 19, 2020 18:28
|
|
|
BlankSystemDaemon posted:Sure, it supported it. Your eyes were lying to you. 128b130b is the only supported line coding for a link operating in Gen3 mode. 8b10b has some positive things going for it, btw. The code word substitution design makes it trivial for the transmitter to ensure it sends an equal number of ones and zeroes, and to put tight bounds on run length (number of consecutive ones or zeroes). Long-term DC balance is important to maintain error free data transmission, and run length limits help receiver design (receivers have to recover the transmitter's clock from transitions in the received signal; the fewer transitions the receiver sees, the harder its job is). 64b66b and 128b130b don't have algorithmically guaranteed DC balance, just statistically mostly-guaranteed. They also permit much longer worst-case run lengths. These were undesirable tradeoffs for Gen1 and Gen2, when you could just bump the line rate up a bit to get the throughput you wanted and keep everything else simple and guaranteed. However, because physics, easy line rate scaling died at Gen3 and above, so it became worthwhile to put in the work to engineer a much more complicated line coding scheme with low overhead but only statistical guarantees of DC balance. Anyways. I ended up effortposting about line coding when I really wanted to say that for PCIe Gen3 onwards, packet overhead is far more significant than line coding overhead. I just glanced at some specs and, if I counted things up, total header/footer size is about 26 bytes for a TLP with a 64-bit address. Assuming a 256 byte payload, 256 / (256 + 26) gives us an efficiency of ~0.908. Compare this to the efficiency of the line coding: 128 / 130 = ~0.985.
|
|
#
?
Dec 19, 2020 23:11
|
|
|
https://videocardz.com/newz/intel-11th-gen-core-series-to-feature-rocket-lake-s-and-comet-lake-s-refresh-cpus Looks like both the i9-11900k and i7-11700k are 8/16 parts, with the differences being clockspeed and the i9 part having TVB. The i5's are all 6/12 with clock differences. But no plans for a 4 core Rocket Lake, as the "11th Gen" i3s/Pentiums/Celeries are all gonna be based on a Comet Lake Refresh. Looks like both Intel and AMD have decided that sub $200 CPUs are no longer worthy of having current gen IPC performance.
|
|
#
?
Dec 19, 2020 23:40
|
|
|
BobHoward posted:Your eyes were lying to you. 128b130b is the only supported line coding for a link operating in Gen3 mode. I'm sure you know this, but in case someone doesn't, most of the hardware in a modern PC, from the PHY/MAC, over DIMMs, the PCI bus, CPU, and too many other things to list, use training (aka firmware talking to other firmware) to find out the optimum values ot can operate at, while remaining compatible with what it's plugged into.
|
|
#
?
Dec 19, 2020 23:47
|
|
|
Cygni posted:https://videocardz.com/newz/intel-11th-gen-core-series-to-feature-rocket-lake-s-and-comet-lake-s-refresh-cpus In Intel's case, continually cranking the power to keep up with AMD was always a short-term measure, and 95% of consumers in that price bracket would still be well-served by chips going as far back as Skylake/Zen 1 anyways. I have noted that it is awfully suspicious that AMD has no Zen 3-derived 4 core parts, though. Maybe they're saving them for APUs, though considering how desperate people are for silicon, that doesn't seem wise. Either that, or their yields must be *amazing* or be an inverted bell curve when graphing number of usable parts against core count. SwissArmyDruid fucked around with this message at 00:18 on Dec 20, 2020 |
|
#
?
Dec 19, 2020 23:47
|
|
|
BlankSystemDaemon posted:If 128b/130b encoding is the only supported mode in PCI-Ex 3.0, it wouldn't be backwards compatible - which I don't need to remind you, it is. My dude... one of us knows what the acronym LTSSM means off the top of his head, one of us doesn't. You might wanna stop. I will repeat, with emphasis and more detail: 128b130b is the only supported mode for Gen3 (and Gen4) line rate. When negotiation decides to shift up from the initial 2.5GT/s (links always come up in Gen1 x1 and then enter negotiation), if the negotiated speed is 8 GT/s (or 16 GT/s for Gen4), both sides must switch to 128b130b line coding. No other encoding scheme is permitted at 8 GT/s. Conversely, even if both ends of the link are Gen3 capable, while the link is running at Gen2 or Gen1 speed it uses 8b10b coding. You don't get to mix and match line rates and line coding in PCIe.
|
|
#
?
Dec 20, 2020 00:10
|
|
|
Cygni posted:https://videocardz.com/newz/intel-11th-gen-core-series-to-feature-rocket-lake-s-and-comet-lake-s-refresh-cpus There’s no such thing as mobile rocket lake, right? What are ~$600-800 laptops that previously had i3 going to be packing? Or is the real answer for sub $1k laptops just “buy a Mac” until Microsoft’s ARM chips are here?
|
|
#
?
Dec 20, 2020 01:03
|
|
|
Twerk from Home posted:There’s no such thing as mobile rocket lake, right? What are ~$600-800 laptops that previously had i3 going to be packing? Isn't rocket lake the backport of the current gen 10nm mobile chip? Then there isn't much point in using that as mobile instead.
|
|
#
?
Dec 20, 2020 01:23
|
|
|
Twerk from Home posted:There’s no such thing as mobile rocket lake, right? What are ~$600-800 laptops that previously had i3 going to be packing? Mobile Rocket Lake is Tiger Lake, though, I can't in good conscience tell anyone to get any chip with UHD graphics these days. Hurry the gently caress up, Koduri.
|
|
#
?
Dec 20, 2020 01:33
|
|
|
P5800x can do 7.2GB/s on sequential so obviously everyone needs gen 4 now
|
|
#
?
Dec 20, 2020 01:41
|
|
|
Also woah are we mixing circuit/phy tuning or link training with encoding???
WhyteRyce fucked around with this message at 01:50 on Dec 20, 2020 |
|
#
?
Dec 20, 2020 01:44
|
|
|
Disappointed that we won't be seeing 4/8 i3's or 2/4 Pentiums with Xe IGPs, at least based on these current rumors
|
|
#
?
Dec 20, 2020 01:57
|
|
|
Is there an info on whether intel will make a workstation version of rocket lake, and if so when it will be out?
|
|
#
?
Dec 20, 2020 21:48
|
|
|
Ika posted:Is there an info on whether intel will make a workstation version of rocket lake, and if so when it will be out? There's no real indication that Intel will update their HEDT line for a verrrry long time, I'd either consider Threadripper or get a Xeon/EPYC.
|
|
#
?
Dec 20, 2020 21:54
|
|
|
Ika posted:Is there an info on whether intel will make a workstation version of rocket lake, and if so when it will be out? I imagine any HEDT part, if that even continues to be a product line, will be based on Ice Lake and not Rocket Lake. The last Intel roadmap leak had no HEDT update at all, still X299 as far as the eye can see. Rocket Lake is very much looking like a stopgap meant for desktop gamers only until Alder Ridge is out this time next year (maybe?). So no Mobile, no HEDT, no low end desktop.
|
|
#
?
Dec 20, 2020 22:34
|
|
|
Cygni posted:https://videocardz.com/newz/intel-11th-gen-core-series-to-feature-rocket-lake-s-and-comet-lake-s-refresh-cpus Sucks that the Xe graphics aren't filtering down to the low end chips, since better IGP performance is the most interesting thing to me.
|
|
#
?
Dec 20, 2020 23:30
|
|
|
FuturePastNow posted:Sucks that the Xe graphics aren't filtering down to the low end chips, since better IGP performance is the most interesting thing to me. right? getting a 4.0 GHz Pentium with an IGP that's competitive with (if not better than) an Athlon 3000G's Vega 3 graphics would be a powerful little machine
|
|
#
?
Dec 21, 2020 02:10
|
|
|
https://twitter.com/OneRaichu/status/1340715682695024640 597 still seems quite a bit less than Ryzen 5000s getting over 600 on single-thread perf
|
|
#
?
Dec 21, 2020 04:08
|
|
|
gradenko_2000 posted:https://twitter.com/OneRaichu/status/1340715682695024640 Looks like an ES with a 4.5ghz single core max. The 11900k is apparently going to go to 5.3ghz single core with TVB. Thats why they are talkin about a potential 700+ single thread score with retail, which would be pretty impressive.
|
|
#
?
Dec 21, 2020 04:21
|
|
|
MaxxBot posted:There's no real indication that Intel will update their HEDT line for a verrrry long time, I'd either consider Threadripper or get a Xeon/EPYC. Probably will go with i7/i9 or 5800x/5900x, depending on what things look like in the spring. I want to stick with DDR4 for now, or rather upgrade to DDR4 since I'm still on a 4790k and its constantly costing me a couple of seconds here and there which do add up.
|
|
#
?
Dec 21, 2020 16:43
|
|
|
What are the specific reasons that some motherboards allow for a higher overclock on the same chip over other motherboards? I just had to swap out my 6700K from a Gigabyte Z170 Gaming to an ASRock Z270 Fatality and have managed to eek out another 100 mhz on my CPU at a lower voltage than before (4.5 @ 1.350 versus 4.6 @ 1.325). For both I'm keeping cache ratio at 42x with a -100mv offset, and ram is XMP3000 which I've had to a little juice to VCCIO and SA to get stability. Is it better power? A better bios? Edit: Come to think of it I also moved from a gold psu to a platinum one (Corsair RM750 to Silverstone SX700-PT). Could that have made a difference? Shrimp or Shrimps fucked around with this message at 01:45 on Dec 22, 2020 |
|
#
?
Dec 22, 2020 01:40
|
|
|
Shrimp or Shrimps posted:What are the specific reasons that some motherboards allow for a higher overclock on the same chip over other motherboards? There are a bunch of things that all can work in conjunction together to provide a better overclocking platform: -Better PCBs (more layers, more copper) allowing for traces to incur less resistance and/or less EMI from other bits of the board. -More/better VRMs providing stronger, "cleaner" power lines, which can directly impact stability at iffy overclocks. -BIOS options exposed to let you fine tune things more, or enable overclocking-friendly options/modes. E.g., allow for fine-tune voltage adjustments. All sorts of other smaller things, too, but I think those are some of the bigger ones. PSUs can also have an impact, as if the power it's supplying isn't stable (google for "power ripple" if you want more info), that can also play havoc with a chip's ability to hold a high overclock.
|
|
#
?
Dec 22, 2020 01:49
|
|
|
DrDork posted:There are a bunch of things that all can work in conjunction together to provide a better overclocking platform: Thanks for the explanation!
|
|
#
?
Dec 22, 2020 01:51
|
|
|
Software voltage readings are not necessarily directly comparable between motherboards, not even ones on the same chipset.
|
|
#
?
Dec 23, 2020 00:55
|
|
|
The spirit of Mike Tsai the Intern lives on. Someone on reddit just posted the snaps his buddy who "works in the automotive industry" sent him... with full serial number   Now that he realizes he's hosed over his friend for fake internet points he's deleted it. But I'm curious what to make of it. S-spec QS15, produced 33W 2019, given the lack of "wings" it's probably a BGA package? Supposedly on a board with an Altera V and 26GB of GDDR6. I'm curious what "someone in the automotive industry" would be doing with it. The Reddit discussion raised the idea that it's C3000 type thing in an autonomous driving type role. Or it could be some kind of system-on-module for parallel computing with an attached FPGA?
|
|
#
?
Dec 23, 2020 08:02
|
|
|
Paul MaudDib posted:The spirit of Mike Tsai the Intern lives on. Someone on reddit just posted the snaps his buddy who "works in the automotive industry" sent him... with full serial number "Altera V"... do you mean a Cyclone V, or Stratix V? Been awhile since I've used Altera parts but 26GB of GDDR6 doesn't make me think of the low-end part (Cyclone), makes me think of a higher end part like an Arria or Stratix. It's probably some embedded part as the control plane / running the software part while the FPGA accelerates image processing / some massively parallel thing. Why the separate CPU entirely vs. a RISC-V softcore on the FPGA, or usage of a MPSoC type device that glues ARM cores w/ fabric all in one package... no clue.
|
|
#
?
Dec 23, 2020 08:17
|
|
|
"Altera V" is what the OP said, don't think they knew. I agree probably a higher end part if it's got 26GB of GDDR6. Any idea what generation it is? Is this like a new embedded chip just being initially sampled, or just an older ES that hadn’t caught the internet’s radar yet? Again, it's produced mid-2019. Are they going to update the segment that Denverton currently fills anytime soon, is that on any of the leaked roadmaps? Paul MaudDib fucked around with this message at 08:31 on Dec 23, 2020 |
|
#
?
Dec 23, 2020 08:19
|
|
|
|
| # ? May 31, 2024 21:48 |
|
|
Stratix Vs and Arria Vs are really old, like, 28nm. That doesn't seem like something you'd want to be using in some kind of self-driving application, unless they're intentionally cutting corners by using a chip that can barely handle 10 Gbps chip-to-chip bandwidth. Which sounds like an incredibly Tesla thing to do mind you.
|
|
#
?
Dec 23, 2020 08:45
|
|
