I.nfraR.ed

Here you can find links to many bios versions, link to 1202 is still working: https://docs.google.com/spreadsheets/d/19gqxeeAdUmkwgE0EvH4MlJ7meeDbEHYP-FHYW-PMyOk/edit#gid=522016589

@chispy Here you go, uploaded everything I have for that card, including the GPUTweak: https://drive.google.com/drive/folders/1zjNjpkBbChOASw7luWiiXTS9yA2JH08_?usp=sharing PS: Some of the files are probably for the PowerColor card I had back then, but I think it was just a reference with a different cooler.

Maybe it's a bios problem, I can select 1900 from the oc menu on the Gene and it works.

What does it help for? It's still 2:1 MCLK:UCLK

The whole bot is using my personal page customization, it would be hard for anyone else to top that

I tried hard for 5150, but no luck. Might try again tomorrow.

Thanks. Voltage scaling above 5GHz is very bad.

Excellent score and very good cpu!

I guess so, it seems to be a new die and has maximum of 2 cores, where Raven Ridge complex contains 4 cores and if yours is RavenRidge then it should have 2 cores disabled, but the full die. https://www.facebook.com/NkdHardware/posts/pfbid02QT9BaH9a8xA6YXvGbKpnjGZk1tDQVVHFsdY2A6WXqXLdj3o8fruVVPKu9ZEqff6dl

I thought they are the same and there's no visual way to distinguish the crystals, so delidding would be pointless?

I wonder if there's a real Picasso 3000G out there, you wouldn't know as the CPUID is always the one of Picasso and tools identify it based on the cpuid - family, model and revision. The rest, like the fab process, is displayed based on this information. For 3000G it is a manual "override" as the cpuid masks it as a Picasso. It's shady practice from AMD side anyway. I've recently read it got a refresh of the box package. From what I know Dali is Raven Ridge 2, or a Raven Ridge refresh, basically a new stepping, I think. The same problem might exists for some of the rest of the APUs in your list.

No, this error is from the vc++ redistributable. AMD OCL is slower than the one linked by @Zombie568, so if you want to get the maximum out of the bench you have to change. I use this version (Intel OCL 2.1): https://nextlimitsupport.atlassian.net/wiki/download/attachments/22683316/opencl_runtime_18.1_x64_setup.msi This pack is also useful Visual C++ Redistributable Runtimes All-in-One: https://www.techpowerup.com/download/visual-c-redistributable-runtime-package-all-in-one/

No, same for me. Attached mine, but it was in the "modded" folder, so I might have done something to it which I don't remember. 89fxau52.f7e

M and Q (C3 and C2 steppings) revisions of K10 cpus are not in separate categories, while there's a difference in clock potential. On the other hand, 1600 and 1600AF are in separate categories. So both options are sensible. In "our" case with the 3000G there's no difference in specs, they are basically the same core, "Dali" being a newer stepping B1 and carrying a cpuid of "Picasso". It's messed up. But on the other hand, if there's any difference in performance, I believe it is insignificant. It's still a RavenRidge "revision". Since there's a "Dali" category already, perhaps the easiest one is to create an additional 3000G "RavenRidge" one. If I had to decide, I'd remove "Dali" and use just one category, but I can see the other option similarly valid. PS: 1600 and 1600AF makes more sense as there's a significant difference between Zen and Zen+, while there isn't between those 2 3000G. It would have been the same if "Dali" was indeed "Picasso" as the cpuid dictates. I don't know if AMD did it on purpose, but it's misleading. Do you know if, e.g. GPU clocks are different or anything else?

These are complete mess. I can't really find a 100% reliable information for the CPUIDs of the different variants. Sources seem to be contradictory. https://github.com/InstLatx64/InstLatx64/blob/master/AuthenticAMD/AuthenticAMD0820F01_K17_Dali_CPUID.txt The InstLatx64 dumps seem to have CPUID 820F01 (model 20h) as "Dali", 810F81 (model 18h) as "Picasso", 810F10 (model 11h) as "GreatHornedOwl", so basically the same as Picasso, but first revision (0), which is also the cpuid of "RavenRidge". CPU World lists an additional CPUID for "Picasso" - the same 810F10. 3000G added as "Dali" doesn't really match the other "Dali" models, e.g. 3020e: https://valid.x86.fr/qsv6dz 3200G, for example, has the same cpuid as 3000G: https://hwbot.org/submission/5339568_naganannan_y_cruncher___pi_1b_ryzen_3_3200g_2min_23sec_517ms and is still displayed as Picasso by cpuz. I think "Dali" is used for mobile APUs only (FP5 socket). Wikichip lists 3000G as an exception in the "Dali" subfamily https://en.wikichip.org/wiki/amd/cores/dali PS: I guess it is accepted as an exception as an incorrect CPUID. One could think it is the the same as 3200G and 3400G, but with less cores. It is in fact "the same" as the RavenRidge counterpart. I've changed the detection in my software to reflect what cpuz displays. IMO there should be only one 3000G category, thus "Dali" designation removed as both variants are Zen-based. I'm not 100% sure though, so maybe contact someone more knowleadgeble and who has access to NDA documents, perhaps Martin from hwinfo or Franck from cpuz? Two different categories also make some sense as they have different CPUIDs, but in reality the APUs are not different performance wise, if I'm not wrong. @unityofsaints's APU is basically a shamelessly rebranded 240GE.

XOC mode of cpuz enabled.

Back to some 32M on Zen4. Should be able to improve it further. 5m 36.210s

Ah, so not exactly "normie" sticks

Have you tried unsafe math optimization? Should be able to run it with that OpenCL version.

I've seen this 2-3 times and I think it is related to unstable system and/or some files get locked/corrupted due to a bsod. Usually a subsequent restart fixes it for me. If not, then a clean BM install should help.

Thanks! Though I might even get the top LN2 scores, but needed a bit higher frequency and that's about the max on the chiller.

1.05V vsoc is enough on my cpu for 8000, but it also needs lower than average for 1:1. And vsoc didn't lift my LN2 CB when tested in the past, might be benefical for other CPUs, I don't know.

The chat part was a bit of a joke, because I'm still a little bit "affected" as I was waiting for this board and it seems I would never be able to buy it. Not that there's much sense in doing so anyway, got a Gene instead. There seems to be one LN2 mode menu in the bios. Gene has a LN2 jumper, but also not sure what it does apart from unlocking the LN2 profile in bios. Doesn't seem to automatically set any voltages alone, if you don't load the profile, but I need more time to test. I never understood the different LN2 modes on the B550 AM4 too. High VSOC didn't help me before, no matter what I did, the 7950X I have was still coldbugged at around -110. For me, it seems to mostly help with UCLK. Haven't tested it with the Gene on LN2 yet, so I can't tell if there's any improvement in that area compared to my old board. I will try to improve zentimings, as it doesn't detect the blck on boards with eCLK, but I don't promise anything. PS: I've tried up to 1.55V vSOC which was the maximum on the Tomahawk. The CPU was doing full pot with 2nd CPU disabled, but not with both CCDs enabled.

You need to have LN2 mode enabled and usually at least -40C temperature, but maybe that can be bypassed with the physical LN2 switch, at least I can on the Gene. You can freely go above the 1.3 limit once in windows though, even on the Tomahawk. Problem is booting to windows Zentimings might not show the real vsoc, don't rely too much on it, especially in this edge case where it is "technically" not allowed. On asus it was still displaying 1.3, but in reallity it is higher. Not sure if always the case or it depends where you input the value in bios. As for the Tachyon, perhaps it is better all 7 people that have the board to create a chat? :p It would be easier. I haven't seen a single bencher which has the board and is not sponsored/connected with Gigabyte. I finally bought a Gene and...there's almost no difference with the Tomahawk. GDM option is the most notable one. PS: It was still flooring it down to 1.3 on the tomahawk and ambient temps, despite allowing me to enter and save a higher value in the AMD OC menu with LN2 mode enabled.

7-zip benchers, engage now!