AMD Presents New Horizon

Dullard said:
This is something I brought up when I first tried a Blender benchmark, right after AMD showed Zen edging out a 6900 in the Zen logo blender runs. I was using a 5960X @ 4.6 and was getting pummeled by 3770Ks. So I started digging around in the Blender forums, and there are so many tweaks that you can apply.

From the old thread:
Click to expand...

Having a lot of options doesn't mean that it cannot be a useful benchmark. Just look at how many video quality options there are in a game like GTA V and people use that as a benchmark all the time.

The key to being a useful benchmark in my opinoin is twofold: 1) does it imply the real world performance of a product in any way; and 2) is it reproducible. As long as everyone runs the same settings, then comparing benchmark results is a legitimate endeavor.

The problem I see with Blender is that I have no way of knowing if performance in this "benchmark" is in any way analogous to performance in any other application (there may also still be a question about which instructions are used and/or if one architecture or the other is better with a particular instruction set). Just because an RX 480 is great for Doom with Vulkan enabled, doesn't mean that the RX 480 is a great video card, but it can be great for that particular use.
 
Last edited:
I think they didn't give more specifics about instructions per cycle, because Lisa said (multiple times) that they were continuing to tweak the performance. There's another thing that I *think* they are going to try and implement, but I don't want to sound like I'm crazy. ;)
 
MrCaffeineX said:
The problem I see with Blender is that I have no way of knowing if performance in this "benchmark" is in any way analogous to performance in any other application (there may also still be a question about which instructions are used and/or if one architecture or the other is better with a particular instruction set). Just because an RX 480 is great for Doom with Vulcan enabled, doesn't mean that the RX 480 is a great video card, but it can be great for that particular use.
Click to expand...

Agreed. I have the same issue with 3D Mark as a benchmark. Its results are not necessarily influenced by the same variables that impact game performance, nor do its results translate to game performance in anyway.
 
MrCaffeineX said:
Having a lot of options doesn't mean that it cannot be a useful benchmark.... As long as everyone runs the same settings, then comparing benchmark results is a legitimate endeavor.

.
Click to expand...

When computer benchmarks are on the line, everybody using the same settings is a pretty big ask.

Fwiw, it took me 00:37.42 to run Ryzen 5960X @ 4.6 I'd post a screen grab, but with a 4K screen I doubt yoiu could see the little timestamp. And that's without touching anything but "Render".

Ran it again, 00:37.47, so pretty repeatable here.
 
Dan_D said:
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.
Click to expand...
It would be a black eye on their reputation if they did it. But at the end of the day.. we'll se benchmarks before you can order the CPU anyways. Or at least you should let Kyle review it before you pull the trigger.
 
I think the general thought about AMD is they gave themselves too many black eyes. People were skeptical about the results before they even saw them and that's not due them wanting to see AMD fail.

I wonder if this box will have "premium VR" tacked onto it too? Lol
 
Trimlock said:
I think the general thought about AMD is they gave themselves too many black eyes. People were skeptical about the results before they even saw them and that's not due them wanting to see AMD fail.

I wonder if this box will have "premium VR" tacked onto it too? Lol
Click to expand...

Why would they fail if they have haswell/broadwell level singlethread and can deliver 8c/16t in 95w package ?

It's very easy to impress people with that. Run cinebench with a power reading, that's a standard benchmark you can't fiddle with.
 
I would be wary about the TDP AMD and Intel calculate them differently, and the looks of it, Zen locked frequency gets 95watts, so unlocked, it might be going more of course performance varies too, but that as the benchmark can change quite a bit based on settings, its all up in the air.
 
I, for one, am looking forward to the release.

I find the displayed results encouraging, allthough I will reserve judgement until the CPU's are in the hands of review sites such as HardOCP and Techreport.

Lets see how things go, some competition in the CPU buisness would be _very_ welcome.

BTW, my _old_ FX8150 running stock speeds with 8GB DDR3 1333 took 4 mins 27 seconds (267 seconds) for the Blender test. Not a very impressive result :)

Edit:
A second run, where I let the computer render without any programs in the background gave 249 seconds.
 
Last edited:
  • Like
Reactions: N4CR
like this
Ieldra said:
Why would they fail if they have haswell/broadwell level singlethread and can deliver 8c/16t in 95w package ?

It's very easy to impress people with that. Run cinebench with a power reading, that's a standard benchmark you can't fiddle with.
Click to expand...
I didn't say they would fail.

The problem with this "sneak peak" is AMD again skewing results against a super high end CPU only to get shock value.

If they had showed this against a 4790k and impressed with the results (all the while not showing the settings) I think people would be a bit easier with the showing. With them using a 6900k, I hope they don't plan to also use its pricing.
 
razor1 said:
I would be wary about the TDP AMD and Intel calculate them differently, and the looks of it, Zen locked frequency gets 95watts, so unlocked, it might be going more of course performance varies too, but that as the benchmark can change quite a bit based on settings, its all up in the air.
Click to expand...

Blender is a very light load. So it also raises questions why the power delta is as it is. Perhaps someone have been creative with a compiler.

Its a long time till we can get independent numbers. No 3rd party leaks have been positive as such. GB, AOTS etc.
 
Trimlock said:
I think the general thought about AMD is they gave themselves too many black eyes. People were skeptical about the results before they even saw them and that's not due them wanting to see AMD fail.

I wonder if this box will have "premium VR" tacked onto it too? Lol
Click to expand...

Reminds me of the Beastie Boys line "...with two black eyes, your girl ain't that pretty..."
 
Shintai said:
Blender is a very light load. So it also raises questions why the power delta is as it is. Perhaps someone have been creative with a compiler.

Its a long time till we can get independent numbers. No 3rd party leaks have been positive as such. GB, AOTS etc.
Click to expand...

well that file ya it is an extremely light load, they could have modified the compiler or optimized it for it. I am not a render expert with blender, but I do know, certain features activated on other renderers, like Max or Maya scan line, can affect different CPU's based on what extensions are available and capabilities (what I stated before, thread amounts, CPU cores, CPU cycles) differently.
 
Shintai said:
Like a review? Couple of months or so.
Click to expand...

Launch is supposed to be January 17th so I can't imagine we are more than juat over a a month or so out.

That being said, who knows if your favorite review site will receive samples or will have to go out and buy them themselves, further delaying things.
 
Ieldra said:
Run cinebench with a power reading, that's a standard benchmark you can't fiddle with.
Click to expand...

But there's something about Cinebench that AMD users don't like - I don't recall what the issue is, just that Cinebench isn't favorable for AMD CPUs.
 
Zarathustra[H] said:
Launch is supposed to be January 17th so I can't imagine we are more than juat over a a month or so out.

That being said, who knows if your favorite review site will receive samples or will have to go out and buy them themselves, further delaying things.
Click to expand...
These days I care more about indepth motherboard reviews than CPU reviews.

With CPU's I just care about raw speed lol.
 
Zarathustra[H] said:
Launch is supposed to be January 17th so I can't imagine we are more than juat over a a month or so out.

That being said, who knows if your favorite review site will receive samples or will have to go out and buy them themselves, further delaying things.
Click to expand...

Got a link to the official January 17th?
 
I think we are missing the forest for the trees by narrowing in on this one benchmark. Certainly it would be nice to be able to replicate the results, clearly AMD expected people to be able to otherwise they would not have asked people to do that very thing. Now we saw the device in blender, another rendering app, two separate games, and a 3D modeling tool. Now like everyone else i want hard numbers, but what we did see looks promising for now.
 
Ieldra said:
Why would they fail if they have haswell/broadwell level singlethread and can deliver 8c/16t in 95w package ?

It's very easy to impress people with that. Run cinebench with a power reading, that's a standard benchmark you can't fiddle with.
Click to expand...

I've never said they would fail if they can deliver all of that. What I'm afraid of is that they have exaggerated the performance of Ryzen, showcasing instances where it matches Haswell / Broadwell when in reality such instances are quite rare. There are cases where Bulldozer derived CPUs did very well against Intel. Unfortunately, that was a price/performance ratio alone. Typically you got somewhere between the mainstream quad core and HEDT segments in performance with prices that were competitive or lower than Intel's mainstream segment. However, in no scenario did Bulldozer or its variants fare well against Intel's six and eight core offerings. There is arguably a market for that but AMD made things worse by letting the platform stagnate and prices weren't slashed enough to make them appealing to all but the most cost conscious buyers alone. AMD needs to do better with Ryzen than that. Unfortunately, the few benchmarks we've seen aren't representative of what we can truly expect from Ryzen and even worse, some of those benchmarks are questionable at best. At least in the absence of having more specific testing data.

razor1 said:
I would be wary about the TDP AMD and Intel calculate them differently, and the looks of it, Zen locked frequency gets 95watts, so unlocked, it might be going more of course performance varies too, but that as the benchmark can change quite a bit based on settings, its all up in the air.
Click to expand...

Unfortunately, we don't know how true this is. Final motherboard platforms aren't available. Shipping silicon of Ryzen hasn't been reviewed either. We also don't know how bad the CPU will get while overclocked. Power draw isn't necessarily linear with clock speed increases. In other words, power usage at stock speeds may very well be around 95watts while overclocked processors pull considerably more power than their Intel counterparts. We also don't know how much power the motherboards will pull either. You can't look at total power draw numbers at the wall of a Ryzen based system and know how much of that draw is the CPU.
 
Presbytier said:
I think we are missing the forest for the trees by narrowing in on this one benchmark. Certainly it would be nice to be able to replicate the results, clearly AMD expected people to be able to otherwise they would not have asked people to do that very thing. Now we saw the device in blender, another rendering app, two separate games, and a 3D modeling tool. Now like everyone else i want hard numbers, but what we did see looks promising for now.
Click to expand...

BF1 could have been done on a much slower CPU. And the Dota 2 was so rigged it was painful to watch.

b1_proz_11.png


4K wouldn't be much different.
 
Presbytier said:
I think we are missing the forest for the trees by narrowing in on this one benchmark. Certainly it would be nice to be able to replicate the results, clearly AMD expected people to be able to otherwise they would not have asked people to do that very thing. Now we saw the device in blender, another rendering app, two separate games, and a 3D modeling tool. Now like everyone else i want hard numbers, but what we did see looks promising for now.
Click to expand...

Even if we can replicate the results, we don't know how the CPU will fare in other tests. We don't know if AMD cherry picked these tests to generate hype. For all we know these benchmarks we've seen show us Haswell level performance when in reality, it's barely faster than Sandy Bridge in anything else. I agree that what we've seen looks somewhat promising. Again given AMD's history of skewing benchmark results, I think at least cautious optimism or even some skepticism isn't unwarranted.
 
Dan_D said:
I've never said they would fail if they can deliver all of that. What I'm afraid of is that they have exaggerated the performance of Ryzen, showcasing instances where it matches Haswell / Broadwell when in reality such instances are quite rare. There are cases where Bulldozer derived CPUs did very well against Intel. Unfortunately, that was a price/performance ratio alone. Typically you got somewhere between the mainstream quad core and HEDT segments in performance with prices that were competitive or lower than Intel's mainstream segment. However, in no scenario did Bulldozer or its variants fare well against Intel's six and eight core offerings. There is arguably a market for that but AMD made things worse by letting the platform stagnate and prices weren't slashed enough to make them appealing to all but the most cost conscious buyers alone. AMD needs to do better with Ryzen than that. Unfortunately, the few benchmarks we've seen aren't representative of what we can truly expect from Ryzen and even worse, some of those benchmarks are questionable at best. At least in the absence of having more specific testing data.



Unfortunately, we don't know how true this is. Final motherboard platforms aren't available. Shipping silicon of Ryzen hasn't been reviewed either. We also don't know how bad the CPU will get while overclocked. Power draw isn't necessarily linear with clock speed increases. In other words, power usage at stock speeds may very well be around 95watts while overclocked processors pull considerably more power than their Intel counterparts. We also don't know how much power the motherboards will pull either. You can't look at total power draw numbers at the wall of a Ryzen based system and know how much of that draw is the CPU.
Click to expand...


Yeah, this is what I am worried about too.

Some of this is starting to remind me of the Bulldozer launch. Showcasing the few benchmarks where it does well when compared to Intel, when in the vast majority it may not.

I learned my lesson with Bulldozer (though I should have known it already then, I was already a grown-ass adult :p ) I'll only make decisions about anything once real.independent reviewers have put them to the test.
 
Gasaraki_ said:
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.
Click to expand...

Ahh... Gotcha.

Thanks.
 
AMD really needs to find another person to present this work. Every time I see her, I just cringe a little. She is not a good face for her product.
 
KarVi71 said:
I find the displayed results encouraging, allthough I will reserve judgement until the CPU's are in the hands of review sites such as HardOCP and Techreport.
Click to expand...
Keep in mind that Techreport is now owned by an AMD employee.
 
well that would go along with the way he talks, trying to hide all the ugliness, just not good enough though ;)
 
whateverer said:
You know, your fears of the site becoming an AMD-haven have yet to come true.
Click to expand...
You are reading a bit much into my statement. It is simply a conflict of interest that needs to be known. Nothing more than that.
 
whateverer said:
You know, your fears of the site becoming an AMD-haven have yet to come true.

They gave the GTX 1060 an EDITORS CHOICE AWARD even in the wake of the RX 480. Jeff took his time to do it right, and tested EIGHT GAMES in multiple APIs.

http://techreport.com/review/30812/nvidia-geforce-gtx-1060-graphics-card-reviewed/16

Even though you gave a "mixed" review for the GTX 1060, despite using half the games you normally do, so you could rush it out the door back in July.

http://www.hardocp.com/article/2016..._1060_founders_edition_review/10#.WFHJljFvE3c

Tell me, who's become less objective over the last year? You or the Tech Report? Stop reading into things, the original owner is no-longer a writer, and the site is still doing things to the depth they always have.

I can't say the same of [H]. Whatever happened to "best playable settings" reviews with 7-8 games? I miss the [H] reviews of the last seven years :(

Sometimes, it's okay to take your time, you know?
Click to expand...
Brent authored that article, I think you'd get more answers from him.
 
Trimlock said:
Brent authored that article, I think you'd get more answers from him.
Click to expand...

But Kyle still runs the place, and makes the call on how rushed or thorough reviews are. I'm not discounting the review results, Brent does good work, I'm just saying it's half a review.

Get 'em out by friday!
 
whateverer said:
But Kyle still runs the place, and makes the call on how rushed or thorough reviews are. I'm not discounting the review results, Brent does good work, I'm just saying it's half a review.

Get 'em out by friday!
Click to expand...
Valid points, also is wish they could go back to best payable settings. But these days they aren't given a whole lot of time.
 
You must log in or register to reply here.
Back
Top