New around here? Register your SA Forums Account here!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Locked thread
New Zealand can eat me
Aug 29, 2008

:matters:


defaultluser posted:

Can anybody here help out with trying to recreate the 36 seconds AMD got for an live demo of Blender test? If you have an 8-core, it would be incredibly helpful :D

Kyle from [H] is trying to recreate the 36 seconds both machines took in the live demo, but is failing to do so even with a 10-core (low 40s range). With 8 core Broadwell-E at 3.2 GHz, it's 52 seconds, which smells of bullshit (or the "live feed" was heavily edited)

defaultluser posted:

Just did a quick sanity check, and my Ivy Bridge Core i3 3.2 GHz took 4:14 to run the same test.

I managed 61 (70 @ 200 samples) seconds with 2600k@4.8Ghz. Fury did it in 25.5 once the render kernels were loaded (1:15 without)

36 seconds is the kind of performance I expected to see by 2013 when I built this, in 2011. Now that it's 2016 I can stand to wait a while longer

Adbot
ADBOT LOVES YOU

New Zealand can eat me
Aug 29, 2008

:matters:


A comment on the freesync 2 article had mentioned that this "automatic HDR rendering" and TrueAudio Next are the only two unused Polaris features left.

Will anything happen with TrueAudio Next? It doesn't seem to see a lot of adoption. That said we did just see Valve push out a HRTF sound option for CS:GO.

New Zealand can eat me
Aug 29, 2008

:matters:


PerrineClostermann posted:

The really weird thing with ME is, as it was described at a talk at either Defcon or C3, it's basically a completely self-contained microcontroller that has direct access to the hardware the CPU runs on. Great for management, but also great for :tinfoil: And if you try to cripple it, apparently the CPU will shutdown/freeze/stop functioning after 20 minutes or so.

Actually, some people have figured out how to nuke it and also trick it into thinking it's still there. Confirmed working for Sandy/Ivy Bridge, I haven't checked back since but I'm sure other people have tried it with newer processors.

:tinfoil: all the way down

New Zealand can eat me
Aug 29, 2008

:matters:


My Z68 motherboard died. If I can't source a local replacement, a refurb from HK would get here at around the same time as the Ryzen preorders are supposed to start.

Tempted to double down and sprinkle some tax return love on AMD. I don't regret my Fury yet, I'll play ginuea pig. I really want to see what this will do with a nice AIO like the H110X or w/e

So I guess this but unironically maybe?

New Zealand can eat me
Aug 29, 2008

:matters:


I have figured out how to get my 2x16 DDR4-3000 CL15 to run at 2666 with DOCP enabled. Just enable it then manually set the speed from 3000 to 2666. What I don't understand is why the bios detects 15C but RYZEN MASTER is setting it to 16. I read a random amazon review that said Ryzen supposedly prefers even numbered CL timings, and had luck running their 3200 kit at CL14 instead of CL16. I have not had any success with that and attempting to do so results in me having to reset CMOS and start anew. I'm going to try the 4x8 kit again with this month's bios and see if that's good for anything tomorrow.

For whatever reason, with DOCP enabled, the manual processor clock settings in the BIOS don't seem to always apply. I have to go into RYZEN MASTER and manually increase the processor speed, in addition to copying over all of the memory settings from the read-only current profile, to the new profile I'm making, to prevent a restart (which it would want for changed memory settings).

Also, it's reporting at 1T for 2x16, shouldn't that be 2T? There are still situations when various screens in the bios don't even display consistent information, and I don't know if I can trust RYZEN MASTER or other utilities.

I have until April 15 to return either set of ram (2x16 DDR4-3000 CL15 and 4x8 DDR4-3200 CL16, both Corsair), if anyone has something they'd like me to attempt out of curiosity, let me know. I am anticipating eventually returning both and getting a stupid 3600 or 4000mhz pair. The problem is this essentially forces me to choose between gaming performance, and simply having a shitload of ram for when I'm compiling poo poo and encoding things/running VMs

Even going from 2133 to 2666 has been good for anywhere from 7 to 30!!! % increase in performance, less so in synthetic benchmarks, moreso in games. DiRT Rally really responded well to it. DO4M already ran great, but I'm seeing CPU frame times as low as 2.3ms as opposed to ~3.5 previously. That's actually insane IMO as that's potentially over 400FPS in what is IMO one of the best looking games out there (even though its capped at 200). So I guess AMD's claims of being "200hz VR ready" aren't that farfetched.

ASSUMING Dual GPU DX12 actually works out, I will eventually upgrade to a board with 2x16 PCI-E, I have a feeling that the Fury OC will make an excellent companion card to something like a Vega with 16 Compute units.

I also managed to break 14000 in Firestrike, which was just really satisfying if ultimately meaningless.

But this explains how people are throwing up 30,000 Multicore scores in Geekbench, 1 stick of 8GB ram set to an obscene speed. Assuming the ram is fast enough that it doesn't matter and a majority of the benefit comes from increased infinity fabric speeds??

Is there a guide to manually editing bios firmware somewhere? Like do people bindump and pick it apart, or just raw hex editing or what?

If anyone has had success with faster memory, specifically 2x16 or 4x16 sets >3200mhz with the ASUS PRIME B350 PLUS, do let me know.

Don't waste your breath telling me I'm wasting my time, I specifically bought this to keep myself busy with meaningless bullshit. Also going to ignore anyone who "lol AMD"'s, as this is the loving AMD thread and you should know better. Its not even a funny joke. There's real loving potential here

New Zealand can eat me
Aug 29, 2008

:matters:


AVeryLargeRadish posted:

Go watch this video on Ryzen RAM OCing. Short version: The Ryzen memory controller is forcing 1T command rate, you need RAM using Samsung B die chips if you want to get decent timings on Ryzen, also if you want anything over 2666 your motherboard needs an external clock generator, only some of the high end X370 boards have that so 2666 is as high as you get on the ASUS PRIME B350 PLUS.

That doesn't make any sense because it shows up as 2T sometimes!!! Also they literally list speeds up to 3200mhz in the specifications, and there are at least a few sets of ram out there that will DOCP 3200mhz, iirc one is for sure listed in the QVL.

Maxwell Adams posted:

AMD's infinity fabric runs at half the speed of your ram. Cranking that up as much as possible helps out with the bottleneck in there.

Came here to say this. Potato Salad is wrong, and should feel bad.

New Zealand can eat me
Aug 29, 2008

:matters:


No but I'm talking about the QVL for my specific motherboard. That has confirmed working DOCP3200 profiles.

New Zealand can eat me
Aug 29, 2008

:matters:


AVeryLargeRadish posted:

Does your RAM come with those profiles and have they been specifically tested with your model of motherboard? I'm only seeing two kits, the Galaxy HOF4CALCS3600K17LD162C and the Corsair CMK16GX4M2B3600C18, both DDR4-3600 kits that are approved for your board with a Ryzen CPU at 3200MHz.

Those do, yes, which invalidates everything you said above, which was primarily the point I was making. Regardless, we should see another bios patch before April 9th that will change all of this anyways.

New Zealand can eat me
Aug 29, 2008

:matters:


A mildly useful benchmark for those who have to do something like compile Unreal Engine 4 a lot:



Just under 19 minutes total. 1800X, 2x16 3200 @ 2666 (16-17-17-36), 256gb 600P NVME

Edit: as an aside, doing this turned out to be a better "stress test" or "shakedown" than anything else I've ever used. In the process of trying to build this I discovered that a windows update had disabled HPET (and I wouldn't have noticed had RYZEN MASTER not said anything when I checked it after the BSOD), and then hit another BSOD because I was hitting 500MB/s writes on the NVME and this triggered something that apparently required new firmware, which seemed to make the SSD a tiny bit faster overall.

eames posted:

Keep an eye on the power supply components. Most B350 boards I've seen have 4+2 phases which is ok for a future 4 core APU but barely enough for a 8 core CPU at stock frequency. Good X370 boards have up to 12+2 phases.

This is ultimately why you're not going to go past ~4.1ghz with any B350 board I'm aware of right now. Also might as well get a good external clock to go stupid with the ram, it really does matter a whole loving lot.

I'm gonna return the B350 board and buy a real motherboard now. Still, for $99, the ASUS PRIME is "pretty good value"

Have been loving the poo poo out of this thing in DiRT Rally, even with just a Fury OC. (1080p Ultra, 2xMSAA)



Not that I have to justify that this somehow results in a much better driving experience or anything, but I recently beat one of my best times (28th global on Monte, Group B AWD) with relative ease. It's just so much more responsive, I've never experienced such a high minimum FPS with such fidelity.

You'll notice I was playing on Medium settings in that replay, after a few runs at Ultra I figured if more really was better, I could live with ugly trees for a better driving experience.

New Zealand can eat me fucked around with this message at 06:03 on Mar 28, 2017

New Zealand can eat me
Aug 29, 2008

:matters:


BurritoJustice posted:

Doesn't sound like too big a deal as you're never going to go past 4.1GHz with Ryzen anyway unless you luck out with a 99th+ percentile chip.

I've been seeing more and more posts of 4.2-4.3 on people that had otherwise given up on going past 4.1 on the various enthusiast spots, so I'm not so sure that's true anymore.

E: I can get away with 4175 if I really want to use absurd voltages, but I think if I had way more power states it would definitely be stable past that. I leave it at 4100 because that seems the most stable with this board.

New Zealand can eat me fucked around with this message at 06:11 on Mar 28, 2017

New Zealand can eat me
Aug 29, 2008

:matters:


Regrettable posted:

Yeah, some people are getting slightly better overclocking results now that there have been some bios updates.

Yeah, definitely saw a bit more stability out of the ASUS Prime B350 w/ 0513. It's neat/terrifying that you can just download and then load it right out of the downloads folder from the BIOS.

New Zealand can eat me
Aug 29, 2008

:matters:


We can't tell if it's non-flat lids or 'baby millennials with low grip strength' who are afraid to torque down the 4 screws on the heatsink. Apparently people have had positive results from uninstalling/reinstalling their heatsink and really tightening the screws. :rolleyes: maybe the pins scared them


The 0605 BIOS from 04/01 (there is another, more recent bios that does not include the updated microcode and is just stability/temperature fixes) for the B350 Prime board has been great for me so far. Managed to get my 2x16 DDR4-3000CL15 kit running at 2933 w/ CL18-16-16-34. Refuses to be stable past 4075mhz at any configuration. Single-core scores of 2450+ in CPU-Z feel good but aren't all that useful. After "settling" for 2933 @ 4.05, my minimum frame times are up ~11% over what they were with the 2133/2400 @ 4.2ghz. Spent almost the entire weekend grinding the Quake Champions Beta, was averaging ~140fps, 90-150 according to FRAPS. Ultra everything, 1080p 100% res. Not bad for DX11, but I get the impression the VK experience will be much, much better.

IIRC, Stock Geekbench 4 result: https://browser.primatelabs.com/v4/cpu/2137985 (4264/21198 single/multi core scores)

Where I'm at right now: https://browser.primatelabs.com/v4/cpu/2451303 (4584/24560 single/multi core scores)

In other words, 7.5% single-core and 15.8% multi-core increase between 2133@3850 and 2933@4075! (or 1.8%/8% increase vs 2666@4.125)

So far, weird things have been happening to enthusiasts loving with their bios firmware manually to push past DDR4-3600. Something to do with having to drop down to PCI-E 2.0 and also potential USB 3 flakiness? I want to say they were pushing BLCK past 115? Crazy things. Let them do all the hard work

Definitely don't discount the importance of ram timings, going from 18-18-18-36 to 18-16-16-34 was good for a solid 2% in my loose testing. BIOS lets me input CL17, but just reverts to 18 on boot. Definitely safe to say Ryzen prefers even CL timings. If we can run our ram at the XMP/DOCP speeds and timings that they were made for by the end of May, that'd be pretty neat.

I still think some fuckin stupid fast 4233mhz GSKILL CYBER TRIDENT is gonna be the key to actually besting a 6900K in most "real world" tasks once we get the BIOS/Chipset firmware sorted.

My multi-core score of 24560 is just 4.8% shy of the 25759 "average" of the 6900K according to the Geekbench 4 processor charts. Considering that the 6900K by itself costs $1099, and I bought this 1800X, H110i, B350 Prime, 2x16 DDR4-3000CL15, & 256GB 600p NVME for ~$980, that's pretty fuckin cool. Do I officially think Intel should be worried? Yes, yes I do.

Edit:

Double Edit: If you have a specific benchmark you want me to run (like some niche bullshit use case compiling huge open source projects or various encoding/rendering tasks), just make a gist with the most straightforward instructions possible, and direct links to everything I need to do said thing.

I ran "ffmpeg.exe -i bbb_sunflower_1080p_60fps_normal.mp4 -c:v libx264 -x264-params "nal-hrd=cbr" -b:v 6M -minrate 6M -maxrate 6M -bufsize 12M -preset veryfast -c:a aac -b:a 128k -report -benchmark output.mp4" for a video goon and managed to encode 6k x264 1080p60 at 251 fps.

excuse the phone picture of a computer screen, was the easiest thing to do in a twitter dm conversation

In doing so, I found out that the Intel 600.p NVME is actually NOT FAST ENOUGH to fully utilize the 1800X. I could not get utilization past 78% or so. Was hitting peak sustained write speeds of 540MB/s, which is about all these are good for. In the ~85 Matches of Quake Champions I played, only one person ever loaded into the game faster than I did, and they had the Samsung 960 Evo Pro or whatever.

New Zealand can eat me fucked around with this message at 01:19 on Apr 10, 2017

New Zealand can eat me
Aug 29, 2008

:matters:


Wait until May when AMD releases the official bios fix to all of the motherboard companies so they can release new firmware updates. The difference is significant enough that spending $30-100 more on ram will potentially net you 20%+ perf gains where it counts.

That said, it does seem like they're trying to move along a lot faster than that, probably because they're scrambling to iron this poo poo out before the 1500X

Adbot
ADBOT LOVES YOU

New Zealand can eat me
Aug 29, 2008

:matters:


Palladium posted:

Mind helping me running CS:GO and eyeball the average FPS? 1080p, cs_italy, vsync off, -maxplayers_override 36, max bots.

Sorry, this isn't even a benchmark let alone niche bullshit. At least find me a map/script that collects some sort of performance information and documents it.

  • Locked thread