|
quote:We have investigated reports alleging incorrect thread scheduling on the AMD Ryzen™ processor. Based on our findings, AMD believes that the Windows® 10 thread scheduler is operating properly for “Zen,” and we do not presently believe there is an issue with the scheduler adversely utilizing the logical and physical configurations of the architecture.
|
#
?
Mar 13, 2017 23:12
|
|
|
|
| # ? May 29, 2024 18:05 |
|
|
Well, there you have it. "Working As Intended". The absolute and utter _lack of hedge_ has me suspicious. I definitely don't like this, and I don't like how there is not even a whiff of "We are working on BIOS updates" or "We are continuing to work with Microsoft in order to ensure the best-possible performance". There is something going on that the more-technically-minded in the review industry sees as a problem, and I'm not sure why AMD doesn't.
|
|
#
?
Mar 13, 2017 23:16
|
|
|
SwissArmyDruid posted:Well, there you have it. "Working As Intended". There is some amount of hedge since the scheduler statement appears to only address proper differentiation between physical and logical (virtual) cores as well as assignment based on that characterization. It does not, however, deny the inter CCX latency/assignment issue, and in fact seem to discuss this issue specifically later on in the update. quote:Going forward, our analysis highlights that there are many applications that already make good use of the cores and threads in Ryzen, and there are other applications that can better utilize the topology and capabilities of our new CPU with some targeted optimizations. quote:We have already identified some simple changes that can improve a game’s understanding of the "Zen" core/cache topology, and we intend to provide a status update to the community when they are ready. kirtar fucked around with this message at 23:32 on Mar 13, 2017 |
|
#
?
Mar 13, 2017 23:17
|
|
|
https://www.pcper.com/news/General-Tech/AMD-Running-out-Intel-Sheckels-Renews-Contract-Defame-Own-Products
|
|
#
?
Mar 14, 2017 00:12
|
|
|
Also from AMD https://community.amd.com/community/gaming/blog/2017/03/13/amd-ryzen-community-update?sf62107357=1Reagrding Windows 7 and Windows 10 performance delta posted:Any differences in performance can be more likely attributed to software architecture differences between these OSes. Did AMD just slam Windows 10? Why Ryzen runs hot posted:Specifically, the AMD Ryzen™ 7 1700X and 1800X carry a +20°C offset between the tCTL° (reported) temperature and the actual Tj° temperature. Offset exists for the 1800X and 1700X to force fans to run harder, why I'm not sure - maybe they expect idiot to kill their CPU's without it. Ryzen's reported 70-80°C temps are just 50-60°C temps. Why turning SMT seems to net some gains posted:For the remaining outliers, AMD again sees multiple opportunities within the codebases of specific applications to improve how this software addresses the “Zen” architecture. We have already identified some simple changes that can improve a game’s understanding of the "Zen" core/cache topology, and we intend to provide a status update to the community when they are ready. Software problem, sounds similar to why they think games overall are performing worse and SMT just aggravates it (game developers are lovely at their jobs apparently)
|
|
#
?
Mar 14, 2017 00:18
|
|
|
So is AMD going to double down on the split L3 fabbed with the cores in Zen+ or will they move to a unified L3
|
|
#
?
Mar 14, 2017 01:09
|
|
|
Zero Gravitas posted:LIke I said, this machine is intended as a toy machine to test out some workflow processes and whether its possible to do this idea i have (however slowly) on a shitbox frankenstein machine thrown together from spare parts I have lying around and some from the bargain basement bin instead of dropping serious coin on it. I know from work that OpenFOAM doesnt play particularly nice with hyperthreading so I was trying to avoid that. I appreciate the nod towards ex-server hardware, but I'm really not looking for that kind of stuff right now. I went to one of OpenFOAM's trainings and they said it was fine with hyperthreading, it just didn't do anything except use a bit of overhead in parallelization. Although based on that I haven't bothered trying it and can't speak for any benchmarks. From what I've read, the optimum OpenFOAM computer minimizes threading and maximizes memory bandwidth. I dunno what physics you're simulating, but if you're doing transient, turbulent, or really any flow that takes longer than a day to converge, I would really recommend ECC RAM. Without ECC you will not get repeatable results, and the odds of experiencing an error increase proportionally to the length of calculation.
|
|
#
?
Mar 14, 2017 01:14
|
|
|
Risky Bisquick posted:So is AMD going to double down on the split L3 fabbed with the cores in Zen+ or will they move to a unified L3 No one knows, but if they were I'm not sure why they wouldn't have in the original design. In theory there is nothing wrong with the design as long as the Infinity Fabric is running fast enough - 40ns inside the cores, 140ns between CCXs @ 22GB/s, and Infinity Fabric speed is directly tied to the IMC clock at a half rate. The best change they could do is 1) Improve the speeds at which memory can run, 3) Make the IF and IMC run at better than 1:2. Changes in design would mean optimizations for Zen wouldn't carry over to Zen+ and newer iterations.
|
|
#
?
Mar 14, 2017 01:18
|
|
|
Did anyone mention this absurdity yet? The -X SKUs are intentionally adding 20°C to their reported temperature so fans will run faster. Incredibly ghetto "fix", and it means people think their Ryzens are running way hotter than they are. Presumably newer software can hard-code (ugh) a -20°C offset to compensate.
|
|
#
?
Mar 14, 2017 15:24
|
|
|
HalloKitty posted:Did anyone mention this absurdity yet? The -X SKUs are intentionally adding 20°C to their reported temperature so fans will run faster. Incredibly ghetto "fix", and it means people think their Ryzens are running way hotter than they are. Presumably newer software can hard-code (ugh) a -20°C offset to compensate. Put another tally in the "1700 non-X is the only Ryzen worth buying right now" column.
|
|
#
?
Mar 14, 2017 15:29
|
|
|
HalloKitty posted:Did anyone mention this absurdity yet? The -X SKUs are intentionally adding 20°C to their reported temperature so fans will run faster. Incredibly ghetto "fix", and it means people think their Ryzens are running way hotter than they are. Presumably newer software can hard-code (ugh) a -20°C offset to compensate. Wait what, WTF is the point of that? What is it trying to fix?
|
|
#
?
Mar 14, 2017 16:13
|
|
|
Don Lapre posted:Wait what, WTF is the point of that? What is it trying to fix? I have no idea at all. None. They're practically sabotaging their own product launch. vvvvv kirtar posted:Makes fans run faster on the 1700X and 1800X Yeah, but an offset on the reported temp is a stupid way to do so. Why not make the mobo manufacturers enforce a different fan speed policy based on SKU instead of breaking a standard feature? HalloKitty fucked around with this message at 16:24 on Mar 14, 2017 |
|
#
?
Mar 14, 2017 16:14
|
|
|
Don Lapre posted:Wait what, WTF is the point of that? What is it trying to fix? Makes fans run faster on the 1700X and 1800X
|
|
#
?
Mar 14, 2017 16:16
|
|
|
kirtar posted:Makes fans run faster on the 1700X and 1800X Yes but why? Is there a thermal problem that damages the cpu over a certain temp?
|
|
#
?
Mar 14, 2017 16:17
|
|
|
Would this not cool the chip more and therefore allow it more headroom to turbo?
|
|
#
?
Mar 14, 2017 16:21
|
|
|
At the expense of noise. Again, why not just report the correct temperature and have board manufacturers adjust their bios.
|
|
#
?
Mar 14, 2017 16:22
|
|
|
I would love it if they did that to fix an issue with their RGB stock cooler and gave no fucks about any other cooling solution.
|
|
#
?
Mar 14, 2017 16:24
|
|
|
Don Lapre posted:At the expense of noise. Again, why not just report the correct temperature and have board manufacturers adjust their bios. Presumably because they didn't trust board partners to define an aggressive enough default fan curve.
|
|
#
?
Mar 14, 2017 16:26
|
|
|
Apparently the default asus curve runs fan at 100% at 75c so its really loud https://www.reddit.com/r/Amd/comments/5z9weg/amd_confirms_20c_offset_thermal_reading_bug/dewmne6/ and really its only at 55c Guess this wont help the old amd runs hot and loud.
|
|
#
?
Mar 14, 2017 16:30
|
|
|
Isn't that kinda what Intel Core2Quad Yorkfields were doing a bit there? The Temps were off by like 5-10C from the likes of the Q6600 and you had to manually adjust to get the correct reading? Not to the extent of 20C, but I know I have heard of this before intentionally or otherwise.
|
|
#
?
Mar 14, 2017 16:36
|
|
|
HalloKitty posted:Did anyone mention this absurdity yet? The -X SKUs are intentionally adding 20°C to their reported temperature so fans will run faster. Incredibly ghetto "fix", and it means people think their Ryzens are running way hotter than they are. Presumably newer software can hard-code (ugh) a -20°C offset to compensate.  FaustianQ posted:Also from AMD https://community.amd.com/community/gaming/blog/2017/03/13/amd-ryzen-community-update?sf62107357=1
|
|
#
?
Mar 14, 2017 17:39
|
|
|
EdEddnEddy posted:Isn't that kinda what Intel Core2Quad Yorkfields were doing a bit there? The Temps were off by like 5-10C from the likes of the Q6600 and you had to manually adjust to get the correct reading?
|
|
#
?
Mar 14, 2017 17:51
|
|
|
I can think of some explanations for the offset hack, i.e. rushing CPUs out the door and not wanting to rely on mainboard manufacturers for some last minute fix or wanting to prevent early adopter overclockers from pushing the CPUs too deep into unsafe Vcore voltages at seemingly safe temperatures. Though how do they explain 1800X idle temperatures of 35°C with aircooling at 21°C ambient? Is the offset fixed or a curve? :iaam:
|
|
#
?
Mar 14, 2017 18:09
|
|
|
eames posted:Though how do they explain 1800X idle temperatures of 35°C with aircooling at 21°C ambient? Is the offset fixed or a curve? :iaam: ... The plot thickens. What on earth has AMD done? This doesn't feel like a rushed launch at all, no sir
|
|
#
?
Mar 14, 2017 18:20
|
|
|
E- it really doesn't seem worth the effort of misreporting it.
GRINDCORE MEGGIDO fucked around with this message at 18:36 on Mar 14, 2017 |
|
#
?
Mar 14, 2017 18:33
|
|
|
feels like an AMD launch
|
|
#
?
Mar 14, 2017 18:42
|
|
|
FuturePastNow posted:feels like an AMD launch you almost make it sound charming
|
|
#
?
Mar 14, 2017 19:35
|
|
|
HalloKitty posted:you almost make it sound charming At this point we just accept AMD's impeccable ability to snatch defeat from the jaws of victory.
|
|
#
?
Mar 14, 2017 19:39
|
|
|
Ryzen 5 is being released in a couple weeks according to OCUK, fairly reliable source IIRC. https://www.overclock3d.net/news/cpu_mainboard/overclockers_uk_says_that_amd_s_ryzen_5_cpus_will_release_in_a_couple_of_weeks/1 Based on tests going around, the 1500X might be the best buy simply because it is only 1 CCX. Further, apparently there might be some fuckery with benching being done on clean installs, and possibly some issue causing games to recognize Ryzen as 16 physical core CPU, at least on F1 2016. 
|
|
#
?
Mar 15, 2017 01:02
|
|
|
That sort of gain is probably an outlier. I'm pretty optimistic about gains with a driver and patch but that's a bit too finewine for me.
|
|
#
?
Mar 15, 2017 01:15
|
|
|
Risky Bisquick posted:That sort of gain is probably an outlier. I'm pretty optimistic about gains with a driver and patch but that's a bit too finewine for me. I don't expect a gain of 30-35%, definitely an outlier, but 10-15% is much more reasonable after Windows fixes core parking (this is apparently the discrepancy between Win 10 and Win 7 scores) and some optimizations in games AMD is pushing. Probably won't result in anything amazing happening, but does mean the difference between an Intel and AMD offering is closer to 10% clock for clock.
|
|
#
?
Mar 15, 2017 01:27
|
|
|
My 1700 posted @ 3.9 with 1.2v, so there is hope yet for big things in the coming days.
|
|
#
?
Mar 15, 2017 05:51
|
|
|
guru3d accidentally broke NDA on series 5 Ryzen details
|
|
#
?
Mar 15, 2017 08:48
|
|
|
Risky Bisquick posted:That sort of gain is probably an outlier. I'm pretty optimistic about gains with a driver and patch but that's a bit too finewine for me. Depends on the game but I don't think that example is too much of an outlier. Core parking on/off Memory speed scaling in Deus Ex:MD
|
|
#
?
Mar 15, 2017 09:04
|
|
|
3peat posted:guru3d accidentally broke NDA on series 5 Ryzen details Woah, 169$? B350+ R5 1400 is going to be the budget choice after a G4560+Z270. I think it's legitimately possible to get that combo for the price of just a 7600K.
|
|
#
?
Mar 15, 2017 09:13
|
|
|
VCZ, so grain of salt but usually reliable - https://videocardz.com/67362/amd-pinnacle-ridge-to-feature-up-to-8-zen2-cores Pinnacle Ridge seems to be early 2018, Raven Ridge is mobile only for all of 2017, desktop early 2018. Since Pinnacle Ridge is likely some refinement and clock boosts, it should allow similar cored Pinnacle Ridge products to stand out in direct comparison to their lower clocked Raven Ridge cousins. Also Raven Ridge is coming with 11 Vega cores, odd number TBH but I'm not AMD. My bet is about 750 ti or slightly better performance for the APUs.
|
|
#
?
Mar 15, 2017 10:18
|
|
|
3peat posted:guru3d accidentally broke NDA on series 5 Ryzen details "accidentally" It still infuriates me that 6-cores are going to have one core per CCX disabled. Do you realize how great a VM machine a 6-core with two cores out of one CCX disabled would have been?* Two cores to Linux, entire other CCX to the VM, and no problems with IF/NUMA/schedulers/kumquat getting in the say! In any event, drat, I can't wait for APUs. *in theory SwissArmyDruid fucked around with this message at 11:39 on Mar 15, 2017 |
|
#
?
Mar 15, 2017 10:55
|
|
|
I'm most interested in the 1500x, but will the drop from 16mb to 8mb of cache hurt it much? Will the extra $30 be worthwhile for the 1600 if i don't need the extra cores?
|
|
#
?
Mar 15, 2017 11:25
|
|
|
http://fudzilla.com/news/processors/43118-samsung-ramping-up-to-7nm-next-yearquote:Despite having its CEO behind bars, Samsung will be ready for volume production of its 7nm process node in 2018. SwissArmyDruid posted:Yeah, good luck on getting the required transistor density with planar after going FinFET. That is gonna result in another Fiji-like situation with the die bumping up against that 600nm^2 size again. Holy poo poo you guys I was loving kidding. I mean, yeah, vertical GAA makes the most sense for Samsung, they already have massive amounts of experience with vertical transistors already with V-NAND, this is basically a logical extension of the tech. AMD chips on a GAA process could be _mighty interesting_, now that the WSA lets them go to Samsung for some production. And that's assuming that GloFo doesn't license future nodes from Samsung too. SwissArmyDruid fucked around with this message at 12:13 on Mar 15, 2017 |
|
#
?
Mar 15, 2017 11:59
|
|
|
|
| # ? May 29, 2024 18:05 |
|
|
Lungboy posted:I'm most interested in the 1500x, but will the drop from 16mb to 8mb of cache hurt it much? Will the extra $30 be worthwhile for the 1600 if i don't need the extra cores? I don't think Ryzen hexa and octa cores are worth it for gaming until they figure out how to keep critical tasks from migrating between the clusters. http://www.hardware.fr/articles/956-24/retour-sous-systeme-memoire-suite.html
|
|
#
?
Mar 15, 2017 12:10
|
|