|
repiv posted:https://twitter.com/VideoCardz/status/1528440602223947776 Hopefully that's 15% over Zen3D, otherwise Zen4 is a dud.
|
#
?
May 22, 2022 21:06
|
|
|
|
| # ? May 31, 2024 02:19 |
|
|
K8.0 posted:A 3900X is resalable. They appear to be going for $250+ on ebay. It's still a reasonable choice for someone with high threadcount workloads where single core performance isn't the focus. For ~$225 net a 5800X3D would be a pretty killer upgrade that actually makes a lot of sense if you're mostly gaming. Not really losing much if anything in multithreaded stuff you might do, and a very significant uplift for games. looks like MSRP availability is still limited
|
|
#
?
May 22, 2022 22:05
|
|
|
Twerk from Home posted:What do you do with these? I'm genuinely curious, not heckling. Volunteer scientific computing; mostly biomedical research. I'm sure I've bored this thread with it enough over the years, so head to the BOINC thread if you want more details. quote:My understanding is that for long term high load computing desktop chips are pretty inefficient compared to server chips, which are at much lower clocks. Ryzen has a really good efficiency curve until you get to the top of the frequency range. I use PPT limits to keep things cooler and more efficient in terms of compute/watt. No need to buy server parts just to run slower than default  My serious answer to the comedy option of Epyc is that, yes, that would be pretty sweet, but I'd be paying a fuckton of money for heavy-duty IO capabilities that I'd never use (same for Threadripper). My machines are just a high-core-count CPU; a decent (B550) mobo; one small, cheap, NVMe SSD; a decent PSU and HSF, and a GPU.
|
|
#
?
May 22, 2022 23:14
|
|
|
 does "everywhere" really mean 5.0 on all the PCIe/m.2 slots? How much is that going to cost? lmao I guess it'll be a good future proofed board that you'll hopefully be able to keep around for several generations. edit: and a link to the stream with the times it's going live: https://twitter.com/AMD/status/1528042995483828225 edit 2: and to be clear, there will probably be no point to having 5.0 for several more years at least. nothing uses it yet, and even once stuff does use it, you won't actually see any benefit from it. the 3090 ti still works the same on 3.0 as 4.0, and none of the poo poo gamers or other home computer users do benefits from pcie 4.0 nvme speeds, aside from basic file transfers. Dr. Video Games 0031 fucked around with this message at 23:40 on May 22, 2022 |
|
#
?
May 22, 2022 23:31
|
|
|
ConanTheLibrarian posted:Hopefully that's 15% over Zen3D, otherwise Zen4 is a dud. Probably not, but I'm betting the 5800X3D will probably become unavailable about when Zen 4 is coming out. (As I've mentioned before ITT, it's simple math that AMD makes way more selling these chiplets as Epyc server CPUs.) Then they'll release the X3D version of Zen 4 six months later and the e: Dr. Video Games 0031 posted:edit 2: and to be clear, there will probably be no point to having 5.0 for several more years at least. nothing uses it yet, and even once stuff does use it, you won't actually see any benefit from it. the 3090 ti still works the same on 3.0 as 4.0, and none of the poo poo gamers or other home computer users do benefits from pcie 4.0 nvme speeds, aside from basic file transfers. Hell, none of the poo poo gamers do benefits from PCIe 3 in any significant way as of right now. Klyith fucked around with this message at 00:53 on May 23, 2022 |
|
#
?
May 23, 2022 00:48
|
|
|
So I guess a big conference (Computex) is in a few hours? Any chance we get concrete info the next gen AMD?
|
|
#
?
May 23, 2022 01:05
|
|
|
Klyith posted:Hell, none of the poo poo gamers do benefits from PCIe 3 in any significant way as of right now. AMD seemingly prioritizing 5.0 storage over 5.0 graphics is likely them betting that the storage will be more consequential due to DirectStorage. I do wonder how much that will change things, though I find it hard to imagine 5.0 giving much extra performance over a fast 4.0 drive. Hughmoris posted:So I guess a big conference (Computex) is in a few hours? Any chance we get concrete info the next gen AMD? Yes, that's what we've been talking about today.
|
|
#
?
May 23, 2022 01:17
|
|
|
Maybe you bifurcate it to 2 full speed pcie 4 speed NVMEs. That’s the only interesting thing about “PCIE 5 everywhere” - just think of how much storage you can hang off the chipset at full speed!
|
|
#
?
May 23, 2022 01:31
|
|
|
Dr. Video Games 0031 posted:AMD seemingly prioritizing 5.0 storage over 5.0 graphics is likely them betting that the storage will be more consequential due to DirectStorage. I do wonder how much that will change things, though I find it hard to imagine 5.0 giving much extra performance over a fast 4.0 drive. Doh! I've been all over the place today. Excited to see what comes out of it.
|
|
#
?
May 23, 2022 01:35
|
|
|
hobbesmaster posted:Maybe you bifurcate it to 2 full speed pcie 4 speed NVMEs. populating the second 5.0 x16 slot with 8 4.0 NVMes running in raid 0 at full bandwidth
|
|
#
?
May 23, 2022 01:44
|
|
|
Dr. Video Games 0031 posted:populating the second 5.0 x16 slot with 8 4.0 NVMes running in raid 0 at full bandwidth In theory that could match the performance of fast, tuned DDR4. In theory.
|
|
#
?
May 23, 2022 02:39
|
|
|
hobbesmaster posted:In theory that could match the performance of fast, tuned DDR4. In bandwidth maybe, but DDR4 will crush it in latency by a huge margin.
|
|
#
?
May 23, 2022 02:50
|
|
|
Indiana_Krom posted:In bandwidth maybe, but DDR4 will crush it in latency by a huge margin. 
|
|
#
?
May 23, 2022 03:17
|
|
|
Dr. Video Games 0031 posted:AMD seemingly prioritizing 5.0 storage over 5.0 graphics is likely them betting that the storage will be more consequential due to DirectStorage. I do wonder how much that will change things, though I find it hard to imagine 5.0 giving much extra performance over a fast 4.0 drive. I think it's mainly prioritizing a number that's one higher for marketing. "Intel has some PCIe 5, so we're going to have all PCIe 5!" (And Intel jumping to PCIe 5 was because they got beat to 4 and wanted to beat AMD to 5, even though it was utterly pointless.) Anticipating that PCIe 5 DirectStorage will be a big thing in the near future, when no DirectStorage game of any sort has yet shipped, is like... hoo boy. Lemme just lol a bit at the people who bought PCIe 4 SSDs for DirectStorage, who will probably upgrade to PCIe 5 before a game has ever actually used it.
|
|
#
?
May 23, 2022 04:10
|
|
|
PS5 XBSX uses PCIe 4 NVMe SSDs for their storage, right?
|
|
#
?
May 23, 2022 04:17
|
|
|
ConanTheLibrarian posted:Hopefully that's 15% over Zen3D, otherwise Zen4 is a dud. I expect it would be that, on some computational benchmark suite, a Zen4 core averages 15% IPC uplift over a Zen3 core. That's how every other generational uplift number for Ryzen has been figured; not through the lens of specific product vs specific product. That said, "Zen3D" is just a portmanteau/nickname made up for the products with vcache. The cores are still Zen3. Whatever the real numbers turn out to be, I'm curious how much of the uplift will be coming from DDR5. It's fair no matter what the number is, because DDR5 is baked into the platform -- but I'm still curious. gradenko_2000 posted:PS5 XBSX uses PCIe 4 NVMe SSDs for their storage, right? I don't remember what the current xbox uses. The ps5 does, but it has a custom controller that has decompression baked into the silicon, which gives it ludicrous effective read speeds.
|
|
#
?
May 23, 2022 04:53
|
|
|
I got my R5 5600 delivered and installed. Haven't had a chance to play a game or so anything particularly CPU hungry, but absolutely amazed how noticeable the difference is over the 2600 in just general browsing, office apps etc. I wasn't sure I'd see the difference at all. Boot times are down a bit, I feel the 2600 was blunting the impact of moving from the EVO 950 to the WD 570 NVME SSD a bit, definitely seems snappier all round. Seeing as I have a ridiculously over-specced 360mm AIO on this thing, I was wondering if it's worth playing with overclocking at all? I can't actually find any OC testing and benchmarks for the non-x chip, but I understand that there may be a couple reasons for a chip to get binned down, and I did notice when looking for info on overclocking that they dropped the L3 cache by 1mb, presumably so they can sell chips with very minor faults on that RAM. Obviously I didn't buy this expecting it to be a Sandy Bridge affair, and if I recall from the initial reviews, Zen3 chips are binned pretty close to their max anyway - but I'm wondering knowledgable goons, if it's worth taking the time to see if this thing can hit higher max clocks or not, seeing as I have the cooling and a decent board (MSI B550 Mortar)?
|
|
#
?
May 23, 2022 06:43
|
|
|
It's starting https://www.youtube.com/watch?v=BRtBB2VnF8M
|
|
#
?
May 23, 2022 07:00
|
|
|
I'm remembering why I quit watching these things (from anyone, not just AMD). I wanna fast forward so bad.
|
|
#
?
May 23, 2022 07:09
|
|
|
Awful lot of time being spent on how everything is basically the same. e: oh wow that was that I guess Alchenar fucked around with this message at 07:33 on May 23, 2022 |
|
#
?
May 23, 2022 07:25
|
|
|
currently on a big 5900x rig but im intrigued by a 16+core 7000 series setup, though i want to see what numbers the gen5 ssds can do first as i'm a disk transfer boy. other'n that it looks good.
|
|
#
?
May 23, 2022 07:36
|
|
|
No increase in core count: "...up to two core chiplets, each with eight Zen4 cores..." New socket limit of 170W though. I'm guessing those 5.5GHz clocks don't come cheap. Not needing a spare GPU to do builds and troubleshooting is nice though. Edit: 1MB of L2 per core is also nice. mdxi fucked around with this message at 07:42 on May 23, 2022 |
|
#
?
May 23, 2022 07:36
|
|
|
So yeah, basically they said nothing that wasn't already leaked. So the 15% single-threaded uplift claim is based on cinebench scores compared to a 5950X. Considering the CPU should be clocked at least 12% higher (5.5GHz was shown during the quick Ghostwire Tokyo demo), that's not leaving much room left for IPC gains (just a 2.5% IPC uplift?). This is a pretty disappointing number, since this will leave Zen 4 behind Alder Lake in this particular test, let alone Raptor Lake. Gaming performance will differ though, considering the 5800X3D gets more performance in single-threaded games than the 5950X despite scoring worse in Cinebench. How this 16-core Zen4 CPU does in comparison, i don't think anyone can say until it's in reviewers' hands. Dr. Video Games 0031 fucked around with this message at 08:49 on May 23, 2022 |
|
#
?
May 23, 2022 07:42
|
|
|
the x3d chips almost seem like a threat to this one, people cannot say enough good poo poo about it. interesting
|
|
#
?
May 23, 2022 07:44
|
|
|
manufacturers are now sticking m.2 slots next to the RAM. that's new to me. motherboards are going to perpetually grow in size to accommodate more and more m.2 slots until they eventually take up the entire length and height of the corsair 7000D Dr. Video Games 0031 fucked around with this message at 08:05 on May 23, 2022 |
|
#
?
May 23, 2022 07:59
|
|
|
I presume there's a reason they mount them parallel to the board and don't just replace a PCI3 slot with a 4xm2 bank
|
|
#
?
May 23, 2022 08:02
|
|
|
edit: removed the previous post. I'm getting mixed messages from MSI's marketing materials and AMD's marketing materials. MSI advertises 28 gen 5 lanes from Raphael CPUs in this graphic, while AMD advertised 24 gen 5 lanes in their "Total AM5 platform I/O" slide from their presentation. I don't know what exactly this means, if one figure is chipset IO and the other is CPU IO, or maybe MSI just had the wrong number. MSI is also advertising three x16 5.0 slots in their high-end X670E motherboards. I imagine there must be a lot of bifurcation happening when using all three, but it's not clear. If the total lanes are 28 + 24, then it could theoretically work without bifurcation (though you'd surely run into bandwidth limitations in the link between the chipsets and the CPU)
Dr. Video Games 0031 fucked around with this message at 08:34 on May 23, 2022 |
|
#
?
May 23, 2022 08:06
|
|
|
hobbesmaster posted:Maybe you bifurcate it to 2 full speed pcie 4 speed NVMEs. mdxi posted:I expect it would be that, on some computational benchmark suite, a Zen4 core averages 15% IPC uplift over a Zen3 core. That's how every other generational uplift number for Ryzen has been figured; not through the lens of specific product vs specific product. One thing that is interesting is that doubling the size of the L2 cache hasn't made a huge difference to IPC. Iirc, that's something Intel will be doing with their 13th gen as well. If they don't get much of a bump out of it too, we may have found the point where the diminishing returns on cache size increases aren't worth it anymore. ConanTheLibrarian fucked around with this message at 08:52 on May 23, 2022 |
|
#
?
May 23, 2022 08:44
|
|
|
Gordon and Ian shooting the poo poo post-keynote: https://www.youtube.com/watch?v=UdTGshQ0A_0&t=5229s
|
|
#
?
May 23, 2022 09:27
|
|
|
GN Summary of the Keynote https://www.youtube.com/watch?v=QZdScv4zxXM Assepoester fucked around with this message at 11:54 on May 23, 2022 |
|
#
?
May 23, 2022 11:41
|
|
|
are there any new and fun use cases for the igpu, now that amd and (maybe) intel are using proper graphics chips beyond just accelerating rendering and encoding?
|
|
#
?
May 23, 2022 12:36
|
|
|
kliras posted:are there any new and fun use cases for the igpu, now that amd and (maybe) intel are using proper graphics chips beyond just accelerating rendering and encoding? I think the big thing is that with the advent of FSR (and maybe Xe) as image upscalers that can work without need for tensor cores, you can stretch iGPU even harder That said, according to that GN vid the Ryzen 7000 CPUs are still being marketed as CPUs and not APUs, so presumably the iGPU on these things is going to be even worse than the Radeon graphics we see today, and a "real APU" with a big boy iGPU is going to be a separate line altogether
|
|
#
?
May 23, 2022 12:45
|
|
|
gradenko_2000 posted:PS5 XBSX uses PCIe 4 NVMe SSDs for their storage, right? They both use PCIe4 but only the PS5 uses a full four lanes, the Xboxes use just two lanes so their actual throughput is more like a PCIe3 x4 drive As mentioned both have extra secret sauce for hardware accelerated decompression that doesn't (yet) exist on PC though
|
|
#
?
May 23, 2022 13:26
|
|
|
kliras posted:are there any new and fun use cases for the igpu, now that amd and (maybe) intel are using proper graphics chips beyond just accelerating rendering and encoding? I think it is folly to expect igpu on desktop to be anything other than just another bit of fixed-function acceleration silicon. You might never touch it, but the silicon and the instructions exist. Of course, AMD needs to make their fixed-function silicon worth a drat, I don't think anyone would dare use it for encode/decode, but that might change in the future. Now, that being said, AMD and Intel are not on equal footing here. While Intel's iGPUs necessitate coming at the expense of die area because they are still making monolithic dies, they HAVE to come up with poo poo for it. for AMD, they could come back with an -F equivalent SKU three, four months down the line by swapping in an I/O die that doesn't have the relevant iGPU silicon, and say, has a bulked-up memory controller instead, for those people who really don't care and are chasing every bit of perf.
|
|
#
?
May 23, 2022 13:47
|
|
|
Don Dongington posted:Seeing as I have a ridiculously over-specced 360mm AIO on this thing, I was wondering if it's worth playing with overclocking at all? I can't actually find any OC testing and benchmarks for the non-x chip, but I understand that there may be a couple reasons for a chip to get binned down, and I did notice when looking for info on overclocking that they dropped the L3 cache by 1mb, presumably so they can sell chips with very minor faults on that RAM. There's probably almost no performance gain from just turning on PBO and giving it more watts, thought you can try. If you have more time to experiment (and validate stability), the new hotness for OCing zen 3 is using an undervolt offset + the curve optimizer to try to get more MHz with less volts. The downside is that stability testing is much harder -- in this method it's possible to be stable running a stress test and occasionally unstable at idle. The 5600 & 5600X have the same amount of L3, which is 1mb less than the 8-core chips. I dunno if that's because they can block out a section of flawed silicon, or if some of the L3 is kinda "attached" to the cores that get disabled. IIRC the L3 is generally the least likely place to have flaws because it's way less complex than anything else. kliras posted:are there any new and fun use cases for the igpu, now that amd and (maybe) intel are using proper graphics chips beyond just accelerating rendering and encoding? If you do multi-OS there are really cool things you can do now with hardware pass-through to VMs.
|
|
#
?
May 23, 2022 13:54
|
|
|
Valve Steam Deck posted:GN Summary of the Keynote Micro Star International Micro Star International Extreme Gaming X670 Extreme Ace Asus Republic of Gamers Crosshair X670 Extreme Extreme Gigabyte X670 Extreme Aorus Xtreme edit: had to rewind twice to make sure I heard his Alienware joke correctly hobbesmaster fucked around with this message at 16:03 on May 23, 2022 |
|
#
?
May 23, 2022 15:47
|
|
|
micro star international makes them sound like the villains in a cyberpunk novel, which is to say a much better name
|
|
#
?
May 23, 2022 15:57
|
|
|
Klyith posted:There's probably almost no performance gain from just turning on PBO and giving it more watts, thought you can try. If you have more time to experiment (and validate stability), the new hotness for OCing zen 3 is using an undervolt offset + the curve optimizer to try to get more MHz with less volts. The downside is that stability testing is much harder -- in this method it's possible to be stable running a stress test and occasionally unstable at idle. Yeh, I use this machine primarily for work, so losing stability on idle sounds like a bit of a shitshow, but the wattage thing was something I was considering. I did a few stress tests using prime95 small FFTs on both single and all core, and while all core hit 4.3ghz and 82c in a few seconds, with one thread it sat on 4.45 with PBO enabled, at about 60c. Given it wasn't thermally constrained I'd have to guess it was hitting its power budget.
|
|
#
?
May 23, 2022 15:58
|
|
|
Don Dongington posted:Yeh, I use this machine primarily for work, so losing stability on idle sounds like a bit of a shitshow, but the wattage thing was something I was considering. I did a few stress tests using prime95 small FFTs on both single and all core, and while all core hit 4.3ghz and 82c in a few seconds, with one thread it sat on 4.45 with PBO enabled, at about 60c. Given it wasn't thermally constrained I'd have to guess it was hitting its power budget. Add a +200 auto OC , it’ll probably hang out at 4.65GHz all day. My wife’s 5600x has a 240mm aio and it’ll cleanly stay at the +200 boost for quite some time while in anything except a stress test.
|
|
#
?
May 23, 2022 16:01
|
|
|
|
| # ? May 31, 2024 02:19 |
|
|
https://www.theverge.com/2022/5/23/23137502/amd-mendocino-zen-2-rdna-2-6nm-battery-life-price-laptop So am I reading this right that they die-shrank Zen 2? They've got new silicon Zen2 6nm monolithic chips to push into cheap laptops? What's the reasoning behind this? I'm guessing that Zen 2 on 6nm is somehow more cost-effective per die area than Zen 3 on 7nm or Zen 4 on 5nm?
|
|
#
?
May 23, 2022 16:04
|
|
