|
eames posted:We'll find out soon enough I guess. Wouldn't this be sort of related to the Navi technology they are working on on the GPU side?
|
![]() |
|
![]()
|
# ? Jan 17, 2021 04:11 |
|
eames posted:We'll find out soon enough I guess. I'm not sure how I feel about this... I was hoping they'd be binned Zeppelin dies, since Ryzen 7 yields seem to be topped out. ....unless Zeppelin are actually four Ryzen 7 dies with some extra sauce, then I have no loving idea how to assess parts anymore.
|
![]() |
|
Risky Bisquick posted:Wouldn't this be sort of related to the Navi technology they are working on on the GPU side? Yup, AMD is all-in on Infinity Fabric. Their plan for Navi to do the same thing on the GPU side that they're doing on the CPU side right now. If they can manage to glue together several under-volted Polaris/Vega cores and have them function as a single unified GPU I think they'll be very successful. I'm interested to see what the die layout looks like for the 16 core Threadripper part. They showed off the 32 core layout and it's 4 8-core modules arranged in a square. I'm wondering if the 16 core will be; 2 modules on one side of the chip? diagonal? centered? Or (worst case) will it be some sort of conglomeration of various combinations of modules with some of their cores disabled. I really can't imagine them doing that last one; but it's AMD.
|
![]() |
|
That's probably exactly how they're going to do it - gives them a way to recoup as many defective dies as possible. They did it for their 4 & 6 core products, probably just scaling the same thing up.
|
![]() |
|
If they can get the infinity fabric fast enough and low enough latency, then it shouldn't be too much of an issue, as long as they address the NUMA issues and make sure that all the context switching costs are known/available to the OS. It would be a great way to salvage marginally performing chips, or chips that are super great except for that huge defect in core 3.
|
![]() |
|
Methylethylaldehyde posted:If they can get the infinity fabric fast enough and low enough latency, then it shouldn't be too much of an issue, as long as they address the NUMA issues and make sure that all the context switching costs are known/available to the OS. It would be a great way to salvage marginally performing chips, or chips that are super great except for that huge defect in core 3. Since its single cpu NUMA doesn't come into play with infinity fabric, do note that 1700-1800x parts don't do numa and had other fixes come in without having to use NUMA as a scheduler.
|
![]() |
|
Krailor posted:Yup, AMD is all-in on Infinity Fabric. Their plan for Navi to do the same thing on the GPU side that they're doing on the CPU side right now. If they can manage to glue together several under-volted Polaris/Vega cores and have them function as a single unified GPU I think they'll be very successful. There are probably more optimal arrangements, it's kind of an interesting question. Are there any model systems (toy or real-world) where both the logical design (eg VHDL) and the design constraints on the fab process are known, ideally along with a datasheet on the process with something like "thermal output per mm^2" according to some clock/duty cycle rating? (I doubt it but someone please surprise me) Worse comes to worst you should be able to discern a lot of this experimentally - delid that bitch, toss some thermal-sensitive laquer on there, and record it as it goes through a test suite designed to gently caress with each particular stage or execution unit. I'm 100% sure one of the reviewers did this for Ryzen? Paul MaudDib fucked around with this message at 01:07 on May 31, 2017 |
![]() |
|
Anybody watching the AMD thing? Got a dying cellphone battery, can't watch or listen while I'm working on this machine.
|
![]() |
|
SwissArmyDruid posted:Anybody watching the AMD thing? Got a dying cellphone battery, can't watch or listen while I'm working on this machine. Anandtech are liveblogging: http://www.anandtech.com/show/11476...-starts-10pm-et Nothing interesting so far, just rehashed slides and OEM talk.
|
![]() |
|
SwissArmyDruid posted:Anybody watching the AMD thing? Got a dying cellphone battery, can't watch or listen while I'm working on this machine. Aye. They are currently parading a bunch of OEMs with increasingly thick accents, I'm not sure I've understood much of the past 10 minutes. Edit: Apparently the ROG Ryzen 7 based laptop is 60% faster than any other laptop. Needs qualification I think. Anarchist Mae fucked around with this message at 02:31 on May 31, 2017 |
![]() |
|
No word on Vega, what a let down. Still time I guess
|
![]() |
|
Ryzen/Vega APU something something
|
![]() |
|
Ryzen Mobile is a CPU and Vega GPU on one die, and it's loving tiny. I was expecting separate dies. 50% better CPU performance, 40% better GPU performance and 50% of the power requirements than their previous APUs.
|
![]() |
|
Nice, all Threadripper parts will have 64 PCI-E lanes. No segmentation there.
|
![]() |
|
16C/32T 64 PCIE 3.0 Lanes for Threadripper. RIP MY THREADS LISA SU
|
![]() |
|
Good god Threadripper is the size of an old school hard drive
|
![]() |
|
Threadripperrrrrr![]()
|
![]() |
|
Someone quote the Threadripper comic.
|
![]() |
|
No Vega stuff until end of July. RIP AMD shares.
|
![]() |
|
![]()
|
![]() |
|
Still no news about threadripper prices or clock speeds oh dear
|
![]() |
|
By AMD standards, that was a pretty good event I reckon. Okay demo'ing loving Prey with dual AMD Titans/16 cores and have frame tearing was a chuckle ![]()
|
![]() |
|
Measly Twerp posted:Ryzen Mobile is a CPU and Vega GPU on one die, and it's loving tiny. I was expecting separate dies. 50% better CPU performance, 40% better GPU performance and 50% of the power requirements than their previous APUs. That's supposed to be 704 Vega cores as well, so why even the gently caress is Big Vega approaching 550mm˛? That kind of performance should be pretty incredible for gaminmg even, like 1080p medium-high on even the most recent titles, if the performance of the A12-9800 is anything to go by. repiv posted:Nice, all Threadripper parts will have 64 PCI-E lanes. No segmentation there. No pricing though, but I don't think it's going to be like Intel's i9s and having 64s lanes on even the cheapest Threadripper CPU could make for a strongly compelling option despite Ryzen trailing in performance overall.
|
![]() |
|
Maybe cut down CUs due to non compute stuff?
|
![]() |
|
https://twitter.com/Flandre_1507/st...663506743914497
|
![]() |
|
Measly Twerp posted:Ryzen Mobile is a CPU and Vega GPU on one die, and it's loving tiny. I was expecting separate dies. 50% better CPU performance, 40% better GPU performance and 50% of the power requirements than their previous APUs. Holy crap, you weren't kidding. ![]()
|
![]() |
|
![]() I feel like AMD is killing it right now except for the impending desktop Vega disaster. Based on the Ryzen TDP scaling we've seen so far they're in a really good position for mobile. Doubly so if they can pair the APU with a few gigabytes of HBM.
|
![]() |
|
Geez if they can just put a "lousy" 1GB of HBM1 on their APU's it'd be a huge win for iGPU performance. It could probably compete fairly well with some of the mid-ish range dGPU's at 1080p or less resolution with that. Seems like that isn't gonna happen though. At least not for 2017.
|
![]() |
|
So without availability date, I suppose there won't be early samples going to reviewers any time soon?sauer kraut posted:Good god Threadripper is the size of an old school hard drive
|
![]() |
|
PC LOAD LETTER posted:Geez if they can just put a "lousy" 1GB of HBM1 on their APU's it'd be a huge win for iGPU performance. It could probably compete fairly well with some of the mid-ish range dGPU's at 1080p or less resolution with that. Seems like that isn't gonna happen though. At least not for 2017. As much as I'd like to see a zen quad with a RX480-level GPU on die paired to a stack of HBM, but it doesn't make any sense for Raven Ridge when and the memory controller can push 30-40 GB/s on DDR4 at average clocks. The GPU part is absolutely tiny. Arzachel fucked around with this message at 11:00 on May 31, 2017 |
![]() |
|
wargames posted:Since its single cpu NUMA doesn't come into play with infinity fabric, do note that 1700-1800x parts don't do numa and had other fixes come in without having to use NUMA as a scheduler. Yeah it will, look up Intel's Cluster-On-Die architecture. You can have multiple numa nodes defined on a single socket. You need something to indicate to the OS where the crossbar/whatever is and to schedule threads around it unless you want even more erratic performance.
|
![]() |
|
PCPer did some tests on the ROTR Ryzen update. It looks like a pretty solid improvement ![]()
|
![]() |
|
Arzachel posted:As much as I'd like to see a zen quad with a RX480-level GPU on die paired to a stack of HBM, but it doesn't make any sense for Raven Ridge when and the memory controller can push 30-40 GB/s on DDR4 at average clocks. The GPU part is absolutely tiny.
|
![]() |
|
The Raven Ridge platform is my most hype Vega product. Hopefully there are bigger ones than the demo unit at the show for mobile platforms. I'd take upscaled 900p@60 or with Freesync 2 in a thin and light, or 1080p Freesync 2 in an XPS 15 style computer. The Threadripper platform is looking super neat, too. Since overclocking won't get it to Skylake-E performance, I wonder if all those PCIE lanes and soldered IHS are going to pull folks to AMD.
|
![]() |
|
Is AMD managing to go from CPU = BAD and GPU = GOOD to vice versa in one generation?
|
![]() |
|
![]()
|
![]() |
|
Wistful of Dollars posted:Is AMD managing to go from CPU = BAD and GPU = GOOD to vice versa in one generation? That's definitely a summary of current events, yes
|
![]() |
|
![]()
|
![]() |
|
Vega is a concept, an ideal we strive towards. There is no Vega, because Vega represents our yearnings and struggles toward the future. Vega is pure, and will forever be beyond reach.
|
![]() |
|
![]()
|
# ? Jan 17, 2021 04:11 |
|
Really though, is there any reason to actually be concerned about them using two Vega cards for the Prey demo? Everyone seems to have forgotten the "drivers still aren't ready and they're just emulating fiji" defense that was so popular when those janky numbers leaked
|
![]() |