Separate names with a comma.
Discussion in 'AMD Processors' started by HardOCP News, Dec 13, 2016.
Ran the Benchmark at 4 Ghz
Specs in sig.
It was live and we recorded it. You could see him start the tests. I even blew it up on my 48" 4K display and while it is not clear, it all points to being ~36 seconds. I don't see that being off by a second or more.
The 6950X with 8C/16T only, at 3.7GHz, which is it top single core Turbo Speed, it still does not make sense to me.
honestly, I think the real question is if you can get the AMD numbers on a production chip.
any chance they are lying about this to make their cpus look significantly better? ie. they know many people dont have a 6900k so they can safely lie about the scores without too many people being able to call them on their bs. so when everyone runs their 4-6 core i7's or i5's or amd 8 cores they see a huge improvment over what they have. as you can see in the thread many people are getting above a min and "looking forward to upgrading" from their pretty decent i7's
Anyone running the handbrake comparison? I pretty sure they said that was available as well.
Also, isn't Ryzen more of a SOC and less of a traditional CPU?
I am not in the tinfoil hat camp on this, I am just in the confused camp. While there are not a lot 6900K out there, those are there and folks have those, and quite frankly, the test I have done here today should mirror what they showed us. Same cores, same clocks, same application and version, and they SHARED the exact file as they explained. I will reach out to AMD and see what they have to say, if anything. The guy that is running AMD CPU PR is a long time buddy of mine and a good all around guy that should want to help us reach some sort of conclusion.
Would need that file and all the settings used. Lots of variables there.
I downloaded Blender, but I'm not sure how to run the test. Help, please.
i7 6800k @ 4ghz 57.76s
Those outside edges were taking a long time compared to the AMD demonstration.
Install, open the file from AMD, the RENDER IMAGE or f12.
Sounds like the render settings for the demo is different then the default render settings in the program.
Also if AMD is using a different version, like a newer version it could have different settings or the tenderer itself has changed. So AMD is not really giving to much out on ability to compare. Maybe some of the presets in the panel options can line up the end performance better. Long time since I've looked at Blender myself.
Kyle, thanks, but it isn't working for me. No worries. I'm sure I'm just missing something. No big deal.
can anyone get a close up of the settings they used? i messed with render samples and got to 18 sec
Hehe, I have lots of friends in tech. Some of them still even like me.
Getting back to Ryzen, I admit I'm intrigued. But I know better than to swallow AMD's bullshit and get hyped. I am eager to see what happens when [H] puts it through their testing suite. Might build a new PC with one of these instead of waiting for Skylake-E depending on how the reviews pan out. Last AMD processor I had was an Athlon XP 2000+ .
Well, they did not mention any changes to the default Blender configuration, and given that AMD mentioned that the file was there for download and to compare at home, I would ASSUME AMD had not made any tweaks. But who knows at this point. Should never assume anything, right?
Here is 6950X at 4GHz with 8C/16T. Giving up trying to make sense at this point. Hopefully get some feedback from AMD.
I hope they fixed all the per core temperature reporting which was FUBAR on the last AMD CPU I looked at. We did bring their attention to that.
All this is exciting however, no matter how it pans out. I hope AMD can find a good performance price niche no matter the clocks. If Zen can get to 4GHz, it will hopefully be able to find a good enthusiast market.
Sigh, you guys make me want to get out of bed and start benchmarking. Fine you win.
Also did anyone compile blender from source not just grab a binary? Lots of prepackaged apps on Windows that come from Linux projects are compiled at sort of lowest common denominator settings for compatibility. Compare that to say Gentoo where blender is going to fly being custom compiled with every x64 extension known man enabled. Don't have a Windows compiler handy but it might get you much faster results.
The Star Wars demo was so very painful. After the specs listed at 37 minutes (lol single data card): "...and its running the hottest game out there. The brand new...I'm gonna give Renee a minute to catch up with me." While waiting for the match to start, "She is like...booting up the machine." Follow this segment up with a super awkward countdown.
AMD marketing is certainly pushing multi-core/threaded performance, so I expect them to be competitive (and hopefully innovative) in this department. I am curious to see single thread performance. Thanks for trying Kyle, but please keep your sanity and stop trying to make sense of the bench results.
I am hoping AMD will help give us some insight in the morning, especially since it advertised "try it at home." We should at least be able to dupe results with released hardware.
You'll notice, when they did the live render, it was obviously pre-cached, since it was already on the guys screen.
5820K @ 4.5Ghz Mem: 2133mhz
currently tweaking blkc / mem for faster render times / stability.
Even still I have not seen enough delta in scores to make their ~36 second render time understandable. Again, not a Blender benchmark guy....
So are these CPU's going to be unlocked or locked down?
I'll bite for $499 or whatever if it's unlocked and we can get a good 4ghz+ clock on these things ..... safely
One more data point - tested on my bone stock 6600K at home and obtained this:
A Blender test that cant be reproduced. The one being around the number is a 4.25Ghz 10 core chip or even more cores?
GPU limited test to show its equal.
Ye, so promising. Really.
The problem is also you can make your own Blender version.
Try out the Ryzen Blender demo for yourself.
*You will need to download Blender 2.78a 64bit from: https://www.blender.org/download/
Set render samples to 100 (under "Sampling" tab on the right side). Anandtech forums are buzzing about it. File comes with the setting at 200.
EDIT: Doing so will practically half your time. Anandtech forum's proof comes from a nasty screengrab from PCper's video, and from SWEclockers still images from the event.
https://forums.anandtech.com/thread...zen-benchmarks.2482739/page-116#post-38629977 The post and the following one.
EDIT: just for fun, with samples set to 100, my i7 2600k (stock) nets 1:00.89.
Going further, my GTX 780 Ti (stock) with 400x400 tiles and samples set to 100 is under 9 seconds.
I go from 77 seconds to 39 seconds on a stock 6700K (No MCE, 4.0Ghz). But that would be way too close to a 6900K.
It was a very small scene with sub-optimal settings (800x800). Nothing like Cinebench, a program that usually tests massively complicated scenes with large outputs. I would expect any recent CPU to get similar numbers. We also can surmise from the TDP test, that AVX is not enabled (since the total system powers are very close in the test, and Intel's 140W TDP is only valid with AVX - supposedly, I am way out of date when it comes to Intel's big platform). EDIT: I am wrong. TDP does not change.
Only time will tell guys, when we get a real production CPU and can test in un-controlled conditions will we really know how it performs. In any case, I hope Ryzen is a success and brings good competition to Intel's best.
Also, according to the SWEclockers images, AMD is running 2.77, not 2.78a (the version the New Horizon website currently recommends).
EDIT: I don't recall any major changes to the renderer from 2.77 to 2.78a, but this and the sampling deal doesn't paint a clean image. Whoever managed the details definitely was not paying attention.
Just making sure that we are all clear on what was stated by AMD and what we are trying to make sense of for those coming in late. Again, I am not part of the tinfoil hat crowd here, just a bit confused and wanting to get it right.
As it stands right now, the demo we were shown is worth jack and shit. And I am not happy with that. I did not want to see AMD lead off its Zen marketing with bullshit, especially after inviting us to compare. It it a great great at credibility (which is SO much appreciated), but as of this moment, it has ended up being bullshit.
Come on Lisa Su! Make me believe! Don't lead off with reasons to make me question the results shown. I am not happy, but I am waiting for clarity. Please. Please. Please. I want to leave this situation knowing that Roy Taylor had nothing to do with the marketing.
Render sample set to 100 as suggested in this post....
Nope. minus 10 seconds....
Shit, bad screen shot as the CPU scaled back down.
CPU set at 3.2GHz across all cores.
100 Samples 26.15 seconds
5820K @ 4.5Ghz / 2133 Mem
Well I just ran it with samples at 100 and I got 26s with six cores @ 4350MHz which seems more less right to me. I'm clocked 26% higher and I have 25% fewer cores.
If Zen at 3,4 is 36s then (26% clock increase = ~20% reduction in execution time ) so we're at ~ 28.8s assuming Zen at 4350mhz. Odd. I get less than that, and that's not even accounting for having 2 less cores yet which would entail a 33% to ~38s.
Yep. Still not adding up.
Rerun at 100 sample, 8C/16T at 3.2GHz. Still no match. With correct screen shot.