AMD Presents New Horizon

I'm a systems engineer... (and a "gamer") even I want them to win.

My eyes are crossing all over the place. Used to be that we'd see a product announcement from companies that were more or less accurate. Why didn't they talk about instructions per clock cycle on a per core basis? Why don't we know that? Why don't companies talk about it?

Processor performance used to be easier to gauge. I'm not saying MIPS per core is a gold standard. At least it's concise.... But everything is marketing smear now.

All they had to say was "Ryzen does X instructions per clock" and 40% of this thread simply vanishes...
??? That's all they have been talking about since Bulldozer flopped. Mention of IPC is in all of their material and it's mentioned in the video. Hell we have people here who have tested using BD. Those scores ain't even close.
 
At AT forums the version is that default Blender setting is 128. So, will anyone give it a shot? Since 100 samples clearly get close enough to demoed system, i believe that one should get it close to livestream version.
On my fresh install of Blender 2.78a the default samples was set to 200.
 
At AT forums the version is that default Blender setting is 128. So, will anyone give it a shot? Since 100 samples clearly get close enough to demoed system, i believe that one should get it close to livestream version.
with 128 samples I get 00:42.17. So my I7 6700K at 4.6ghz is almost as fast as Zen?
 
Upstream it was commented upon that successive runs needed a restart to flush the cache, otherwise subsequent runs had faster times than the first run. Brent or Dan did that (restarts) and had consistent times within .1 second. Without the restart, the times varied by 10s of seconds.

I suspect that may be cause of the live demo difference to which you're referring. (Unless I've misunderstood your post.)

The live demo was actually 10s SLOWER, so the cache thing doesn't apply since that was speeding it up slightly.
 
Take your 5960x to a fixed 3.4ghz and report back if you can. Good info!

Sure, here you go

aQDpbWf.png
 
This thread appears to have devolved into page after page of people trying to recreate dozens of unknown variables to reproduce a benchmark that is, at best, of limited value.
 
  • Like
Reactions: xorbe
like this
This thread appears to have devolved into page after page of people trying to recreate dozens of unknown variables to reproduce a benchmark that is, at best, of limited value.
I don't know. It think it helps put amd in check through verification or validation. They have a history of cherry picking things while making boisterous claims that later get proven to be cherry picked and wrong. I find it kinda fun and also hope they are telling the truth this time.
 
I don't know. It think it helps put amd in check through verification or validation. They have a history of cherry picking things while making boisterous claims that later get proven to be cherry picked and wrong. I find it kinda fun and also hope they are telling the truth this time.

I agree - replicability is a key component of research - but when you clearly can't replicate a process, then you already know what you need to know. We won't know for certain until Kyle and others have thorough reviews up, but nothing in yesterday's presentation makes me think Zen is a flop.
 
i5-3320M @ 2.6GHz: 200 samples: 4:40 100 samples: 2:46

Pretty sure I heard they are calling it "the Platfrom".

Just for fun could we start referring to it as "the Platfrom". I we all do it, maybe it will stick.
I could get on board with that :)
 
its probably just different setting sin Blender thats causing the variations. Render programs (as a few on here will know i'm sure) have a MILLION settings that can make differences so i'm guessing this is what it will be (as evidenced by the 100 samples setting alone and the difference it makes). I'd say there are other settings that aren't exactly the same OOTB and thats causing the discrepancies.
 
Folks just need to wait it out till some of these eventually leak out or they release. People will be able to run test and real comparison soon enough and CES isn't that far so even more info will come out then. (says the person who ran the same blender benchmark on his chip to see how it stacks up) xD

Regardless, I'm just glad such positivity and excitement are surrounding an AMD CPU even with the skepticism some seem to have (which after Bulldoze is quite understandable). Real competition is always best for us the consumers!!
 
He's not wrong.

Let's imagine Ryzen is legitimately faster in games compared to a 6900k.

How do you think that would go-down, press conference-wise? Wouldn't you think AMD would show off a number of games, at a variety of resolutions, showing that Rysen tops the intel chip every time?

Or would they show a single game at a GPU-bound resolution, and say 'you cant tell the difference'.

I'm going to go out on a limb here, and say that maybe AMD doesn't know what to do when they have better performance anymore. :D
 
I'm going to go out on a limb here, and say that maybe AMD doesn't know what to do when they have better performance anymore. :D

Currently the entire demo have been a replica of the FX unfortunately. Hopefully the product isn't acting like the FX. But we have to see with time.
 
pgaster has linked to the system specs above

https://i.redd.it/r39v4xwzre3y.jpg

AMD should be more than happy to provide the Blender settings use, reply awaited ...then run then the test. Should get the same results, right?

Just so far reading this thread no one has actually done that, same chip, mobo etc Til then shouldn't all conspiracy theories/moral outrage stay canned?
 
I don't know. It think it helps put amd in check through verification or validation. They have a history of cherry picking things while making boisterous claims that later get proven to be cherry picked and wrong. I find it kinda fun and also hope they are telling the truth this time.

I agree that the tests need to be replicable.

Why on earth would AMD post the file and invite people to test on their own, and then not share the settings used? At best this is just pure incompetence, and at worst it is intentionally misleading.

The problem is - however - that we are just grasping at straws and guessing at what settings they might have used. That is never going to result in any kind of adequate outcome.
 
Yep, we have to wait and see what AMD does. Either answers with details or silence.
 
Yep, we have to wait and see what AMD does. Either answers with details or silence.

I am upset that they did not share details for us to replicate the handbrake / blender settings on our own systems -- but i have to assume they would not be so foolish to run the benchmark side by side with a 6900k without using the exact same settings. So what i can do is take my benchmark and compare it with kyles 6950k with two cores disabled test and get a general idea where i sit.... which besides a couple laptops posted is back of the pack :)

This thread has inspired me to get back into overclocking though, which is nice. I may have to swap the cooler on my 8320e and see what i can squeeze out of it.
 
I am going to report on my 6800k with 100 samples when I get home. See where this comes in line with Ryzen. I bet the 6800k @ 4ghz is equal or faster.
 
Folks just need to wait it out till some of these eventually leak out or they release. People will be able to run test and real comparison soon enough and CES isn't that far so even more info will come out then. (says the person who ran the same blender benchmark on his chip to see how it stacks up) xD

Regardless, I'm just glad such positivity and excitement are surrounding an AMD CPU even with the skepticism some seem to have (which after Bulldoze is quite understandable). Real competition is always best for us the consumers!!

And I usually would not even try to replicate results until we were invited to by the CEO of AMD and the given the file to do it.

Thanks for the rest of your incredible insight.
 
Likely issue is one department did not know they changed the base settings on blender. A common issue in a big corporation is one department does not quite get all the info it should. But all in house benchmarks should be taken with a bit of skepticism on how it will relate in real world reviews. They usually give you a decent idea of how close they really are, but I am more surprised by the fact the AMD cpu was using less power than the Intel cpu.
 
Why on earth would AMD post the file and invite people to test on their own, and then not share the settings used? At best this is just pure incompetence, and at worst it is intentionally misleading.

I think we all know that AMD's marketing department is pretty much incompetent, so I don't think they were intentionally misleading anyone. Hopefully they get back to Kyle with the settings they used for the demo so everyone can stop whipping themselves up into a frenzy.
 
New Result with Render 100 Samples - 37.85 (37 seconds, that's a huge difference for me, don't understand what that means or why mine went down so much)

I ran it twice, second run at 100 Samples was 37.48

So I put it back to 200 just to double check, and it's 1:13.98 at 200 Samples

For whatever reason, that samples is making a whopping difference on my system.

So yeah, I am consistently getting 37 seconds at Render Samples 100

Ok, Im' kinda a Blender noob. How do you switch the sample from 200 to 100?
 
??? That's all they have been talking about since Bulldozer flopped. Mention of IPC is in all of their material and it's mentioned in the video. Hell we have people here who have tested using BD. Those scores ain't even close.

Ok.. so how many instructions per clock cycle does bulldozer do?
 
yeah I don't think "platfrom" was intentional either!

edit: lol I just had to "correct" that as the forum spell checker auto-corrected it for me.
 
Last edited:
Halving the time of the Blender tests because they used 100 cycles instead of 200, seems to be a simple thing. I imagine that whoever set up the demo picked a number. In a large organization, I highly doubt that the same person who checked the "100" box for the demo was also involved in linking the Blender download to the community. The default 200 selection versus using 100 in a demo is not some sort of conspiracy. After all, BOTH cpus used the same setting. This is just a friggin' minor communications issue. One guy didn't know that the other guy changed a default.

Select 100 runs, try your rig, then compare to the demo. (Or, just cut all the published times using default values in half.) C'mon. This is taking a molehill and bulldozing it into something it isn't....


The problem is that it still doesn't match up. If you set it to 100, then the i7 in the demo is too slow. It should have done that demo in the 20s secs not the 30s like in the demo.
 
I found this after some searching:

http://www.forbes.com/sites/patrick...onstrations-and-product-details/#6cb657c51411

"These systems were otherwise identical except that one system was running on AMD’s new 8 core Ryzen processor while the other ran on Intel’s i7-6900K processor. Both systems ran their processors on stock air coolers, meaning that there isn’t any unfair cooling advantage for one chip or the other. Both processors ran DDR4 2400 memory, however the Intel system ran in quad-channel while the AMD system only ran in dual-channel since the Intel system is capable of quad-channel, and that is theoretically the fastest configuration of the RAM."



edit:

Now Reddit has a slide with system specs, and it seems the Intel had dual channel, not quad:


https://i.redd.it/r39v4xwzre3y.jpg

Looks like Forbes screwed up...


That just fuckin' shows that AMD doesn't know how to build a quad-channel Intel machine. You actually need 4 sticks of RAM to get quad-channel...
 
I agree - replicability is a key component of research - but when you clearly can't replicate a process, then you already know what you need to know. We won't know for certain until Kyle and others have thorough reviews up, but nothing in yesterday's presentation makes me think Zen is a flop.

I wouldn't call it a flop, just that something is fishy. It's probably a decent processor.
 
so after seeing the 200 vs 100 thing I tried it and got:

32bit at 100

100 32.PNG


64bit at 100

100 64.PNG


and that is almost exactly half of my 200 run scores. still no idea why my 64bit is so slow...
 
hahaha im kinda dyslexic so i didnt catch it lol


Even for the non dyslexic our brains have sort of an "autocorrect" feature to help us make sense of the world, so when reading it is easy to miss typos.

This is why proof reading is such a difficult task. It goes counter to the basics of how our brains work.

This is also why you can never trust your own brain. It is biased six ways to Sunday and works to help us delude ourselves.

The smartest people out there are the ones who know to not trust their own brains and design things around then to work around the limitations of the human brain and bias.
 
  • Like
Reactions: noko
like this
I think it's best to take a wait and see approach to making any decisions about Ryzen. AMD has a long history of being deceptive when it comes to benchmark scores. AMD said that Ryzen has a 40% IPC improvement compared to Excavator. AMD has said similar things in the past comparing previous processor architectures to their predecessors or comparing them to Intel CPUs. These statements have always been untrue outside of very specific circumstances. In the Horizon event video it was said that AMD actually exceeded their 40% IPC goal which would potentially place it closer to Haswell than Ivy Bridge as originally thought. Unfortunately, given AMD's past statements this could mean that the 40%+ improvement is a best case scenario that will only show up in one or two benchmarks. This isn't necessarily the end of the world if Ryzen clocks high enough but exaggerated IPC means that those clocks will need to be even higher to overcome that deficiency.

It seems to me that what we've seen is carefully cherry picked to show Ryzen in the best light possible. This is par for the course with AMD as its something they've done time and time again. They do it with CPUs and GPU's alike. On the surface its easy to conclude that AMD is up to its old tricks and that Ryzen could easily be another Bulldozer. On the other hand, the one thing that's changed is AMD's ability to keep secrets better than they once did. Given that, AMD could have more up their sleeves (example: Ryzen clocking 1GHz+ over stock, hex core models with substantially higher clocks etc.) than we know but that's not a horse I'd want to bet on. Fortunately, anyone with a Sandy Bridge or newer system probably isn't hurting and just wants to upgrade. It doesn't hurt to wait and see what Ryzen offers when reviewers actually get them on the bench and spend time with real CPUs. As long as the damn thing doesn't run as hot as Chernobyl, it could be a winner if the thing is capable of clocking high enough. High temperatures don't necessarily stop enthusiasts anymore than higher power requirements. On that front, a 5960X or 6900K both have a TDP of 140Watts and still clock well enough.

The 5960X only has a base frequency of 3.0GHz (3.5GHz) via turbo. 3.4GHz on Ryzen isn't that far off and it wouldn't be that far behind in terms of IPC assuming AMD's statements about IPC can be taken at face value. Granted the newer Broadwell-E CPUs are somewhat faster but don't usually clock as well as Haswell-E does making the difference almost a wash. We can get 5960X's to 4.3-4.5GHz on modest AIO / water cooling configurations so its not unimaginable that Ryzen could do the same. If you can get that kind of performance for the right price, I'd say its a winner provided the X370 platform is good enough. Unfortunately, what little we've seen on that (if true) isn't terribly encouraging. I think AMD really needs a platform for Ryzen that at least equals X99 if not its successor at a reduced price. That alone could drive some people into the AMD camp. If Ryzen is a winner in some of the applications that justify the extra cores of Intel's HEDT platform, it could be a nice low cost alternative. Again the motherboard platform has to be there for it.

Serious question though, if anyone knows the answer, how much benefit would quad channel provide over dual channel in Handbrake/Blender scenarios?

Actually, nevermind, it turns out not much in CPU heavy tasks: http://www.pcworld.com/article/2982...ing-truth-about-their-performance.html?page=2

I have to ask: If AMD really did do something so stupid, does it really seem impossible or improbable that AMD did other things to hamper the Intel machine to make Ryzen look better? Again AMD has a track record of skewing test results to make themselves look better. I understand they need to generate some hype and some sales but AMD has to be careful as pulling shit like that can backfire on them quickly.
 
Back
Top