|
buglord posted:I read that AMD's iGPU solution with the 2200G/2400G blew intel's iGPU offerings out of the water, as far as graphics performance is concerned. Are there any indications of where this is going in the future? Even though I have a GTX 1070, it's really cool to see that you can actually game without a GPU and without relegating yourself to Flash-like steam games. It's going to be tough to go any farther using system memory, there just isn't a lot of bandwidth there to play with (and getting significant gains from delta compression is hard). AMD is apparently developing their own equivalent to Kaby Lake-G called Fenghuang, which is a 28-CU APU with HBM2 (probably for embedded, not socketed).
|
# ? May 2, 2018 04:03 |
|
|
# ? May 17, 2024 19:47 |
|
Embedded being like, baked in and not something you can swap in/out?
|
# ? May 2, 2018 04:15 |
|
It's looking like a BGA Could be interesting in a tiny game box. Maybe a NUC with a glowing monster truck. GRINDCORE MEGGIDO fucked around with this message at 04:34 on May 2, 2018 |
# ? May 2, 2018 04:27 |
|
I mean, quadchannel is a thing but the additional die space for the IMC, additional traces, and thus added expense don't really add up to a product that properly targets it's market segment anymore, and iGPU will at best only ever tickle the underside of midrange dGPUs. It's why AMD is researching "Near Memory" basically low power, easy to add in HBM for CPUs/APUs that will alleviate the bottleneck cheaply. Even 1GB of oldschool HBM would have been like , +30% performance, +/-5%, for Ravens iGPU. Which means Vega 8 would consistently beat the GT 1030 by about 10-15%, and Vega 11 would probably overtake the RX460 with an overclock. It's more also that Near Memory probably solves Zens biggest weakness, it's absolute hunger for high bandwidth, low latency memory. Again, even 1GB of old school HBM as L4 would probably fix it.
|
# ? May 2, 2018 05:25 |
|
AMDs plans leaked: TR+ in August, whatever the hell Z490 chipset is (integrated wifi? rebranding to keep up with the joneses and Z390?) in Summer. https://www.anandtech.com/show/12711/german-it-distributor-publishes-amd-intel-roadmaps-z490-z390-8core-cfl-in-q4
|
# ? May 3, 2018 17:36 |
|
4 months for them to harvest up all the best 12nm dies. 32 threads @ 4.3-4.4 along with the latency improvements should OWN HARD.
|
# ? May 3, 2018 18:11 |
|
Someone needs to jail the necromancer who demands 760G around. Like why even the gently caress dude, upgrade already goddamn.
|
# ? May 3, 2018 19:13 |
|
FaustianQ posted:Someone needs to jail the necromancer who demands 760G around. Like why even the gently caress dude, upgrade already goddamn. I bet there's a lot of unsold AM3(+) inventory for some reason
|
# ? May 3, 2018 19:54 |
|
Munkeymon posted:I bet there's a lot of unsold AM3(+) inventory for some reason It's going to be a literal decade because it launched in 2009 and it looks like the socket will stick around until well into 2019 at a minimum. Khorne fucked around with this message at 20:17 on May 3, 2018 |
# ? May 3, 2018 20:15 |
|
Not cool dudes, no complaining about long-lived backward/forward compatibility please. Take it to the Intel thread.
|
# ? May 3, 2018 20:32 |
|
Munkeymon posted:I bet there's a lot of unsold AM3(+) inventory for some reason It's really too bad AM3+ never really got a last hurrah in like, 2015. But 760G is some ancient poo poo, and so is whatever you can stick in AM3/+. Like, if you wanted to continue to support that old platform at this point why not just make a new chipset so they're full and properly featured? Most 760G boards lack the VRMs necessary to push Phenoms and Visheras properly, so gently caress why? And then it's just wasted money anyway, just let them die, reprocess them or something. Make a deal for some low end NUClikes with Kaveri or Carizzo as they still have potential but lmao Bulldozer is dead even the lowest end SKU of Ryzen absolutely crushes it on performance and price.
|
# ? May 3, 2018 20:35 |
|
I really want to know what the roadmap is for a Zen based successor to the G-series embedded SoCs is like, myself.
|
# ? May 3, 2018 20:38 |
|
Eight new security vulnerabilities affecting Intel chips found. Details are currently under a 90 day embargo, ending May 7th. In case you were wondering, AMD processors do not appear to be affected by these exploits, (hi paul) but they have found some ARM chips that are vulnerable. https://www.reuters.com/article/us-cyber-intel/next-generation-flaws-found-on-computer-processors-magazine-idUSKBN1I42BZ
|
# ? May 3, 2018 20:40 |
|
Bloody Antlers posted:AMD processors do not appear to be affected by these exploits So why did you post here EDIT: In more relevant news, AMD is going to release their patches soon.
|
# ? May 3, 2018 20:44 |
|
Dual channel memory 2700U laptop review, something I know some folks were waitin for: https://www.anandtech.com/show/12709/the-acer-swift-3-sf31541-review-ryzen-meets-laptop
|
# ? May 3, 2018 20:49 |
|
Craptacular! posted:So why did you post here
|
# ? May 3, 2018 21:12 |
|
Craptacular! posted:So why did you post here Because I beat him to the Intel thread
|
# ? May 3, 2018 21:16 |
|
Craptacular! posted:So why did you post here cool, AMD finally took a minute to deal with that super serious big deal problem, where you could totally pwn an AMD system after convincing the owner to flash their BIOS with a third party crafted malicious image. Oh, also they are patching that other thing where if you have physical access to the system, you could in theory do some bad stuff. Thanks for the heads up.
|
# ? May 3, 2018 21:26 |
|
Bloody Antlers posted:cool, AMD finally took a minute to deal with that super serious big deal problem, where you could totally pwn an AMD system after convincing the owner to flash their BIOS with a third party crafted malicious image. Oh, also they are patching that other thing where if you have physical access to the system, you could in theory do some bad stuff. Thanks for the heads up. It's pretty cool pr for AMD tbh. It's all good.
|
# ? May 3, 2018 22:58 |
|
Kazinsal posted:I really want to know what the roadmap is for a Zen based successor to the G-series embedded SoCs is like, myself. Well, the 2200GE and 2400GE are coming soon, both 35W SKUs and both basically screaming to be included in a NUClike. Honestly, ASRock just needs to get off their rear end an produce a Mini-STX board for AM4. I wonder if the upcoming Atari is using the full quad channel capability of Bristol Ridge? My understanding is that the memory controllers between Steamroller and Excavator are very similar, so Kaveri had the option of running 256 bit DDR3 (and supporting up to DDR3 2400Mhz) could improve performance substantially. If such an option exists for Bristol Ridge, could Atari have some kind of custom GDDR5+DDR4 solution? They were saying their Bristol was semi-custom, so this is what I'd assume, it'd allow AMD to just dump all their inventory and the Atari would have just sub Xbox One performance.
|
# ? May 3, 2018 23:23 |
|
Wait, wait, wait: Is AMD seriously about to do 7nm with GloFo & IBM before Intel unfucks their 10nm gamble? Because if so that's pretty LOL worthy.
|
# ? May 3, 2018 23:30 |
|
Fabulousity posted:Wait, wait, wait: Is AMD seriously about to do 7nm with GloFo & IBM before Intel unfucks their 10nm gamble? Yes, but not GloFo, seems Ryzen 3000 will be TSMC because GloFo is delayed. They'll instead be handling 12nm until mid next year if my understanding is correct.
|
# ? May 3, 2018 23:39 |
|
FaustianQ posted:Yes, but not GloFo, seems Ryzen 3000 will be TSMC because GloFo is delayed. They'll instead be handling 12nm until mid next year if my understanding is correct. TSMC and Gloflo are apart of the same research group so their 7nm stuff isn't too different, just gloflo being gloflow here.
|
# ? May 4, 2018 01:30 |
|
Craptacular! posted:So why did you post here To get a row out of you I guess.
|
# ? May 4, 2018 01:30 |
|
wargames posted:TSMC and Gloflo are apart of the same research group so their 7nm stuff isn't too different, just gloflo being gloflow here. TSMC isn’t a member of IBMs research alliance I don’t think.
|
# ? May 4, 2018 01:56 |
|
Cygni posted:TSMC isn’t a member of IBMs research alliance I don’t think. Could've sworn they joined.
|
# ? May 4, 2018 02:49 |
|
Craptacular! posted:EDIT: In more relevant news, AMD is going to release their patches soon. I remember having an argument with someone on Ars about this back in March, they were defending the scammers with nonsense such as: quote:They still have a fair chance of fixing it and having patches ready . The difference is that they have to start fully working from day one thanks to public pressure. Providing a 90 period to disclose anything doesn't make too much sense since you don't know exactly how much is going to take . Those standard periods are just arbitrary . The only reasonable period to provide fixes and mitigations is the shortest one, which could be shorter or longer than the arbitrary 90 days. My head still hurts from the mental gymnastics.
|
# ? May 4, 2018 08:58 |
|
Measly Twerp posted:I remember having an argument with someone on Ars about this back in March, they were defending the scammers with nonsense such as: Started to reply but then the apathy set in, you know exactly how it is.
|
# ? May 4, 2018 12:52 |
GRINDCORE MEGGIDO posted:Started to reply but then the apathy set in, you know exactly how it is. That feeling when you write most of a reply to someone being dumb and then just delete it.
|
|
# ? May 4, 2018 12:58 |
|
https://www.techspot.com/news/74441-tsmc-stacked-wafer-tech-could-double-power-nvidia.htmlquote:The Taiwan Semiconductor Manufacturing Company, better known as TSMC, is holding its 24th annual Technology Symposium in Santa Clara right now, and it’s just unveiled a process that could spell a revolution for graphics cards: Wafer-on-Wafer (WoW) technology. Or, ostensibly, also future CPUs.
|
# ? May 4, 2018 16:56 |
|
My understanding was that die size itself was a limiting factor because fault rates increased along with die area. Does WoW get around that by mixing and matching layers, or is (probably) my understanding out of whack? For some applications the physical size of the package is an issue, obviously, but not for high end parts really.
|
# ? May 4, 2018 16:59 |
|
Have they figured a way to avoid cooking the bottom die, because that's always been the biggest issue with stacking dies.
|
# ? May 4, 2018 17:02 |
|
Unless some additional details were in the presentation that haven't been covered in the articles, I think its use will be strictly-limited to high-yield processes, i.e., 16nm and larger.
|
# ? May 4, 2018 17:02 |
|
Arzachel posted:Have they figured a way to avoid cooking the bottom die, because that's always been the biggest issue with stacking dies.
|
# ? May 4, 2018 17:12 |
|
Arzachel posted:Have they figured a way to avoid cooking the bottom die, because that's always been the biggest issue with stacking dies. The way it looks in the docs, they are proposing essentially spinning off everything low power draw and not super heat sensitive into the lower die (FIVR, I/O, memory controller, L3 cache, special decode engines etc), while the top die becomes the high power stuff (processing cores or shaders or whatever) Seems like another proposal for dealing with a world where process improvement slows to a crawl, and pretty similar to the original concept docs that lead to HBM.
|
# ? May 4, 2018 17:17 |
|
Moore's Law, he dead
|
# ? May 4, 2018 19:35 |
|
What if we doubled the cores every 1.5 years?
|
# ? May 4, 2018 20:36 |
|
Increasingly desperate workarounds and countermeasures don't make Moore's Law not dead
|
# ? May 4, 2018 20:38 |
|
I'm sure the answer is "We don't know", but is there any sense of how much of an improvement Ryzen 2 might be over Ryzen+? The R7 2700X is enticing...
|
# ? May 5, 2018 18:57 |
|
|
# ? May 17, 2024 19:47 |
|
Rastor posted:Moore's Law, he dead Is it 2012?
|
# ? May 5, 2018 19:04 |