|
Im a bad reader. I saw “RDNA 2 CUs” and read it as 2 CUs. My bad! In that case the clock differences and memory bandwidth differences are gonna be interesting. But yeah, not coming to a socket in all likelihood.
|
#
?
Oct 10, 2021 04:01
|
|
|
|
| # ? Jun 9, 2024 11:48 |
|
|
While the Steam Deck is neat hardware, I'd rather have that APU in a conventional thin and light laptop.
|
|
#
?
Oct 10, 2021 06:32
|
|
|
FuturePastNow posted:While the Steam Deck is neat hardware, I'd rather have that APU in a conventional thin and light laptop. This is my feeling too, I have no idea why nobody has tried to make an iGPU based gaming PC before. It's not going to max stuff out, but with fast RAM, tight timings, and smart cost cutting I feel like there's room to undercut the Razer Stealth and other premium thin and light gaming laptops. I guess there's no market for it, you and I are the only ones that would buy it. I guess we'll see how the Steam Deck does.
|
|
#
?
Oct 10, 2021 06:46
|
|
|
I've started pondering a Steam Deck as a Chromebook + Switch replacement for travel. That pondering is currently about one level above "random shower thought" though, and I'm definitely not going to be an early adopter.
|
|
#
?
Oct 10, 2021 07:16
|
|
|
You'd have to haul around a keyboard, mouse, and dock too. Add that on top of the unit itself and you've got an awkward traveling package.
|
|
#
?
Oct 10, 2021 07:49
|
|
|
AMD Announces Ambitious Goal to Increase Energy Efficiency of Processors Running AI Training and High Performance Computing Applications 30x by 2025
|
|
#
?
Oct 10, 2021 07:50
|
|
|
is this another one of their initiatives where they define “efficiency” as the ratio of load power to idle power divided by the phase of the moon?
|
|
#
?
Oct 10, 2021 08:02
|
|
|
Paul MaudDib posted:is this another one of their initiatives where they define “efficiency” as the ratio of load power to idle power divided by the phase of the moon? https://www.tomshardware.com/news/amd-increase-efficiency-of-chips-thirtyfold-by-2025 quote:"We used internal AMD lab measurements of MI50 paired with an AMD EPYC 7742 CPU which produced 5.26TF per MI50 on 4k matrix DGEMM with trigonometric data initialization, and 21.6TF of FP16 on 4k matrices. Summing up the rated power for 4 MI50s (300W TBP), the 225W TDP for EPYC and include 100W for DRAM, plus power conversion losses and overheads to get 1582W for the compute node.
|
|
#
?
Oct 10, 2021 08:08
|
|
|
At least on their last one, when you drilled down into the actual metrics, they needed to do 1.5x the performance at half the power and they called that a 25x increase in efficiency. https://www.anandtech.com/show/11965/amds-progress-on-its-25x20-goal-the-task-ahead Shouldn’t be too hard if they can use the same kind of math here. Paul MaudDib fucked around with this message at 09:00 on Oct 10, 2021 |
|
#
?
Oct 10, 2021 08:42
|
|
|
Twerk from Home posted:This is my feeling too, I have no idea why nobody has tried to make an iGPU based gaming PC before. It's not going to max stuff out, but with fast RAM, tight timings, and smart cost cutting I feel like there's room to undercut the Razer Stealth and other premium thin and light gaming laptops. the Vega-based APUs already are the "high-performance" part of that particular niche, compared to, say, Intel's pre-Xe/Iris offerings there's no reason to use anything better than Vega because you can't squeeze out any more bandwidth out of DDR4, and if you were to do something like on-chip HBM or GDDR5 or even GDDR6, that just pushes you into console territory... which is already what the consoles are, or what this Steam Deck is look at the Aya Neo or the GPD Win series of Windows-based handhelds as an example of what you can do with current technology: lots of indie titles can work well on it, and even more demanding games can scrape-by on 720p-low heck, a Ryzen APU is still plenty capable with the right expectations: https://www.youtube.com/watch?v=tXlqh9p1Jkk&t=395s https://www.youtube.com/watch?v=lmroXDF4rVQ&t=179s we're definitely going to see a generational improvement with DDR5 approaching GDDR5 levels of bandwidth, coupled with APUs getting outfitted with RDNA2, but my opinion is that this space didn't get pushed any farther than it did because that was already the reasonable limit of what was available gradenko_2000 fucked around with this message at 10:34 on Oct 10, 2021 |
|
#
?
Oct 10, 2021 10:32
|
|
|
RDNA2 has better delta compression and better bandwidth efficiency in general, switching to RDNA2 is the only way to get more out of DDR4.
|
|
#
?
Oct 10, 2021 11:21
|
|
|
Twerk from Home posted:This is my feeling too, I have no idea why nobody has tried to make an iGPU based gaming PC before. It's not going to max stuff out, but with fast RAM, tight timings, and smart cost cutting I feel like there's room to undercut the Razer Stealth and other premium thin and light gaming laptops. I mean, it depends on how you define iGPU. There was that Intel chip a couple years back that had an AMD GPU strapped on, so it could run on the Intel iGPU for low power and switch to the AMD for performance. There was at least one thin-and-light laptop put out using it. Unsurprisingly, Intel quietly axed the line without living up to their promises of long term driver support.
|
|
#
?
Oct 10, 2021 11:28
|
|
|
Also I'm not sure Vega iGPUs are all that bottlenecked by DDR4 anyway. https://www.youtube.com/watch?v=mi4eAWWWxGc The benefits from using faster RAM were fairly minimal in these benchmarks. It makes me think that there's still some room to improve the speed of iGPUs on current platforms. Dr. Video Games 0031 fucked around with this message at 11:40 on Oct 10, 2021 |
|
#
?
Oct 10, 2021 11:36
|
|
|
Vega has some hosed-up performance characteristics, so what it's limited by in any particular case is pretty variable. If the application can keep the shaders full, it needs a ton of memory bandwidth -- thus the HBM on the original Vega design. But that's a huge if, because keeping the shaders full was one of the achilles heels of later GCN chips and it reached ludicrous inefficiency with Vega. I remember back when rdna came out there was a vid with Ian Cutress saying Vegas generally ran at like 50% of their theoretical peak in games or something crazy like that. edit: this is your monthly reminder to lol at stadia Klyith fucked around with this message at 00:24 on Oct 12, 2021 |
|
#
?
Oct 12, 2021 00:22
|
|
|
Klyith posted:edit: this is your monthly reminder to lol at stadia i haven't kept up with stadia, are they still not talking about gen2 hardware? their value proposition is more questionable than ever when the new consoles run circles around their gen1 nodes
|
|
#
?
Oct 12, 2021 00:31
|
|
|
Always surprised Stadia is still going tbh. The Xbox cloud thing works really well though and it’s great for my kid to play Subnautica and other stuff. It’s definitely got some latency going on there but for the right kind of games it’s not a huge deal and incredibly convenient.
|
|
#
?
Oct 12, 2021 00:46
|
|
|
Klyith posted:Vega has some hosed-up performance characteristics, so what it's limited by in any particular case is pretty variable. If the application can keep the shaders full, it needs a ton of memory bandwidth -- thus the HBM on the original Vega design. It sounds like you know something about this, was Vegas problem like the Pentium 4 where it has a super long pipeline? But GPUs can't branch, so how would that work?
|
|
#
?
Oct 12, 2021 01:08
|
|
|
Twerk from Home posted:It sounds like you know something about this, was Vegas problem like the Pentium 4 where it has a super long pipeline? But GPUs can't branch, so how would that work? found it, it was David Kanter not Ian Cutress https://www.youtube.com/watch?v=V7wwDnp8p6Y I'm not an expert, just watched a technical vid and know enough to understand it. but in fact the tl;dr is that the P4 analogy isn't terribly off-base: GCN ran instructions every 4 cycles, but was pipelined 4 deep to keep the units running all the time. Once upon a time this was fine, early GCN GPUs were quite good. (I had a 7870 that was the longest-in-use video card I've ever had.) But as time went on it was hard to make it faster, AMD had to settle for making it wider. It didn't work well, the overhead got higher, the penalties for cache misses got functionally worse. By Vega even HBM couldn't save it. OTOH the Vega in a Ryzen iGPU is a lot less stupid wide. They also may have made some tweaks to it to keep it semi-decent. I dunno why they haven't put RDNA (1 even) into the mainstream Ryzen iGPUs before though.
|
|
#
?
Oct 12, 2021 01:48
|
|
|
2022 Rembrandt APU with RDNA2 graphics and DDR5 might be a pretty good jump
|
|
#
?
Oct 12, 2021 02:39
|
|
|
https://www.youtube.com/watch?v=yE9PsKWYYXA AMD mentions their plans for the next year here. Zen 3 with 3D V-Cache in early 2022, Zen 4 "later in 2022." So no real surprises. Zen 3+ is indeed dead. Also new notebook CPUs coming focusing on power efficiency.
|
|
#
?
Oct 12, 2021 16:30
|
|
|
Dr. Video Games 0031 posted:AMD mentions their plans for the next year here. Zen 3 with 3D V-Cache in early 2022, Zen 4 "later in 2022." So no real surprises. Zen 3+ is indeed dead. So this means that the next APUs getting mainstream release are just Zen 3 + RDNA2, or not until a year from now with Zen 4?
|
|
#
?
Oct 12, 2021 17:37
|
|
|
ayyyyyy scored a 3600 for my rig and a x570 and cooler to resell for Ł120 total. six cores here i come
|
|
#
?
Oct 12, 2021 18:52
|
|
|
Dr. Video Games 0031 posted:https://www.youtube.com/watch?v=yE9PsKWYYXA Also, confirmation of AM4 coolers being forward-compatible with AM5. I'm super stoked about not having to replace my NH-C14Ses.
|
|
#
?
Oct 13, 2021 05:21
|
|
|
mdxi posted:Also, confirmation of AM4 coolers being forward-compatible with AM5. I'm super stoked about not having to replace my NH-C14Ses. does that mean no LGA socket? drat
|
|
#
?
Oct 13, 2021 07:30
|
|
|
Pretty sure it means same mounting, z-height, and more-or-less the same heatspreader size
|
|
#
?
Oct 13, 2021 07:36
|
|
|
With all the various leaks, it is very very very probable that AM5 is LGA. Specifically one that looks like this:
|
|
#
?
Oct 13, 2021 09:24
|
|
|
How is AM4 motherboard availability and pricing, of late? I only have a B350 in storage, and the recent confirmation of Zen3D has got me thinking if I should snap-up a B550
|
|
#
?
Oct 13, 2021 13:04
|
|
|
gradenko_2000 posted:How is AM4 motherboard availability and pricing, of late? Generally fine.
|
|
#
?
Oct 13, 2021 20:45
|
|
|
Cygni posted:With all the various leaks, it is very very very probable that AM5 is LGA. Specifically one that looks like this:
|
|
#
?
Oct 13, 2021 21:41
|
|
|
Dr. Video Games 0031 posted:https://www.youtube.com/watch?v=yE9PsKWYYXA Isn't Zen 3 with 3D cache essentially Zen3+, though, or am I missing something? I mean, it's a significant refresh with a performance uplift. I'm thinking a 5600X Super or whatever would be a reasonable upgrade as a last AM4 CPU replacing my current 3600X.
|
|
#
?
Oct 14, 2021 08:54
|
|
|
CaptainSarcastic posted:Isn't Zen 3 with 3D cache essentially Zen3+, though, or am I missing something? I mean, it's a significant refresh with a performance uplift. I'm thinking a 5600X Super or whatever would be a reasonable upgrade as a last AM4 CPU replacing my current 3600X. I think some people were expecting a "Zen 3+" as in another round of performance improvements as a last hurrah to AM4, as opposed to "Zen3D" being "just" Zen 3 with the additional 3D-stacked cache
|
|
#
?
Oct 14, 2021 08:57
|
|
|
gradenko_2000 posted:I think some people were expecting a "Zen 3+" as in another round of performance improvements as a last hurrah to AM4, as opposed to "Zen3D" being "just" Zen 3 with the additional 3D-stacked cache Fair enough - I can see that. For me I see the quoted 15% uplift as a big enough jump for me to consider going to that to ride this system out more comfortably for a few more years. It does just fine now, but that extra headroom would be nice.
|
|
#
?
Oct 14, 2021 08:59
|
|
|
It'll be more beneficial for people on previous gens. For something like a 3600, stepping up to a 5800XT would be a significant enough jump to be worthwhile.
|
|
#
?
Oct 14, 2021 09:09
|
|
|
CaptainSarcastic posted:Isn't Zen 3 with 3D cache essentially Zen3+, though, or am I missing something? No, again, Zen3+ is Zen3 on DDR5, it was intended to be the leading product on AM5 (old product on new socket/memory, like Bristol Ridge) but is essentially scrubbed at this point except for Rembrandt, which is an APU with Zen3/DDR5/RDNA2, and probably will mostly be a BGA product (may get a socketed release, but not necessarily) for laptops, all in ones, etc. That likely leaves a Zen3XT refresh (no vcache, just binned) either very late this year/early next year, Zen3D I assume at the end of Q2 (supposedly 1H and that always means Q2 or else they would have said Q1, plus XT will take up the first bit of the year), then Zen4 has slid to 2023 at this point (certainly 2023 for consumers, possibly Q4 2022 for Epyc and probably sampling for hyperscalers in Q4 at the latest, if not earlier). An HP all in one just leaked with a “Ryzen 7000U series” processor and that’s a distinct possibility of being Rembrandt. So it’s likely 6000 is desktop Zen3XT and or Zen3D and APUs jump the numbering scheme again, probably Zen4 desktop will be 7000 series desktop, meaning we’re back to mixed generations within a numeric lineup. Oh AMD, never change… Paul MaudDib fucked around with this message at 09:27 on Oct 14, 2021 |
|
#
?
Oct 14, 2021 09:22
|
|
|
I mean, if we're taking what they said in the video that came out just a few days ago at face value, we should expect the 3D vcache CPUs to be the only new AM4 product for desktops before the switch to AM5, no additional refresh. And they said it's coming in "early 2022", not "first half," for whatever that's worth. They also clearly said that AM5 is launching in 2022, though there was no specific time frame given. That could just mean for servers only I guess, but I don't think we should trust any rumors saying desktop CPUs are definitely not coming out next year. I'm not sure where you're getting these estimates from. edit: also the "leak" just said "Ryzen 7000 series" which is a very small but important distinction. Though I also don't know how much stock we should put in that listing. Dr. Video Games 0031 fucked around with this message at 09:35 on Oct 14, 2021 |
|
#
?
Oct 14, 2021 09:33
|
|
|
Me actually watch the video!? Heck no! Yeah I mean I was a little dubious about slotting in an XT release in there too, that’s an awfully fast product life cycle with Zen3D coming so soon, but it’s potentially possible to do it quick seeing as it’s just binned SKUs, no microcode, so very little motherboard work/etc, and we don’t know how they’re going to do the branding here. Maybe Zen3D gets the 5000 XT branding (although rumors said they were different) and the cheap 5600/5700X SKUs finally get released, maybe there is a whole new 6000 series with just vcache SKUs and the two lineups coexist, maybe 6000 is a mixture of rebrands and vcache, etc. This is all just complete supposition and the only people who know work at AMD. Remember to parse the language carefully, if they said “AM5 is 2022” that doesn’t necessarily mean Zen4 is 2022, they could lead off with socketed Rembrandt or other Zen3+, and if they said “Zen4 is 2022” that could mean server only. Or maybe its plain language and we do see consumer Zen4 in 2022. It’s hard to say. I’m real dubious consumer Zen4 is a 2022 product at this point, I’ve read a number of rumors saying it’s slid to 2023, but OK I guess. I could easily see servers sliding in earlier, especially sampling and not a full release, but I think consumer may be a bit later. But I guess maybe AM4 and AM5 could coexist for a while too. An AIO computer getting a performance laptop chip makes perfect sense to me though and AMD needs to pipe clean their memory controller regardless of what happens with the actual products themselves, so that one is very believable to me. Zen3+ and a full AM5 launch was supposed to happen not too much later anyway, the timing is about right. Paul MaudDib fucked around with this message at 10:12 on Oct 14, 2021 |
|
#
?
Oct 14, 2021 10:02
|
|
|
doing a Bristol Ridge-esque line of not-Zen-4 parts on AM5 would be a gutsy move in the face of Alder Lake, but I guess it's not outside the realm of possibility
|
|
#
?
Oct 14, 2021 10:06
|
|
|
gradenko_2000 posted:doing a Bristol Ridge-esque line of not-Zen-4 parts on AM5 would be a gutsy move in the face of Alder Lake, but I guess it's not outside the realm of possibility Rumors implied (or at least didn’t rule out) that it wasn’t just going to be Rembrandt, there were going to be enthusiast chips, and I can only imagine that’s what their market research told them. Yeah, you get the old architecture but pay for a whole new motherboard and ram that is triple the cost of ddr4 right now, and get the super lovely first gen memory controller and ram, imagine what reviews would have been like on that. There is zero appetite for purchasing a bunch of extra components right now in order to get an early adopter experience with minimal to no performance advantage. Unless DDR5 alder lake unexpectedly dominates the DDR4 variant to a massive degree Intel isn’t going to do well on that either. It’s a weird time, both these platforms were supposed to do battle on DDR5 but I think ddr4 has become unexpectedly important to the buying public, or at least the DIY market. Rembrandt, for the near future, will only be sold in whole systems (laptops/AIOs/etc) I’d imagine - being primarily BGA - and that probably will be the most popular way to consume DDR5 alder lake too, packaged with a whole system. Same for servers, Dell doesn’t want to reuse old ddr4 sticks, they sell whole systems. Those are going to be the markets that lead on DDR5 rather than the DIY market, and even then the timeline may be 6 months or so behind the original plans due to general shortages and delays. The supply chain issues are loving everyone massively right now. Paul MaudDib fucked around with this message at 10:32 on Oct 14, 2021 |
|
#
?
Oct 14, 2021 10:17
|
|
|
Paul MaudDib posted:Remember to parse the language carefully, if they said “AM5 is 2022” that doesn’t necessarily mean Zen4 is 2022, they could lead off with socketed Rembrandt or other Zen3+, and if they said “Zen4 is 2022” that could mean server only. Or maybe its plain language and we do see consumer Zen4 in 2022. It’s hard to say. They said the "new platform" with DDR5 and PCIe Gen 5 is coming 2022, though not in the same breath that they mentioned Zen 4 I guess. A lot of the rumors have not really panned out so far, and so the most reliable information we have in my opinion is the video AMD just released where they talked about this stuff (in fairly casual, non-specific terms). If they're saying they're going to release the "next gen platform" in 2022, then I'm inclined to believe that this is something they're seriously trying to do. Whether they'll actually accomplish it or not is another question entirely. Dr. Video Games 0031 fucked around with this message at 10:35 on Oct 14, 2021 |
|
#
?
Oct 14, 2021 10:33
|
|
|
|
| # ? Jun 9, 2024 11:48 |
|
|
At the timings we're seeing, I wouldn't be surprised if DDR5 alder lake was actually a performance downgrade in games.
|
|
#
?
Oct 14, 2021 10:51
|
|
