Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[IllogicalGlory]

Because if you tweak your code to mimic the average result of a benchmark suite then your benchmark isn't a benchmark at all, as it just doesn't benchmark anything, so it's useless. If we needed proof about CPU-Z, we have one now :>

Sounds a tad too backroom.

Now i'm extremely tinfoil hat... But that seems dodgy to me

The truth is (practically) never this interesting or exciting. I'm pretty sure CPUID saw an unexpected result, investigated it and discovered a bug.

 

[CHADBOGA]

Still, I do not think people understand the potential impact to AMD's fortunes from Naples and Vega.

What will Vega bring against the competition that previous designs didn't against it's competition at the time?

For so many years now, people keep saying that the next AMD GPU will completely change the landscape and it never does.

 

[tamz_msc]

I think that the reference scores in the new version are kinda off. There is a discrepancy for example between the 6950X and 5960X MT ratio(was ITBM 3.0 operational on the former?) and the 1800X vs the 1700X.

 

[Gikaseixas]

What will Vega bring against the competition that previous designs didn't against it's competition at the time?

For so many years now, people keep saying that the next AMD GPU will completely change the landscape and it never does.

If Vega does a 5870 it may very well change the landscape but it's been a while since AMD produced a GPU that good

 

[DrMrLordX]

To the first part: how do you know? I'm not sure what the old bench tested, but that quote from them explained what the new one does, and it is based off an actual application. If the original CPU-Z was worthless, then the new one should be evaluated on its own merits. They didn't update it to reduce performance, they designed a bench that runs properly across all CPUs, not just Ryzen.

How do I know? Whenever anyone produces any kind of benchmark, the burden of proof is on those who produce the benchmark to establish that the resultant figures mean anything. Without providing source code or any points of reference, the CPU-z bench was basically just . . . numbers that sort of fell in line with how fast we expected CPUs to be relative to one another.

When I run y-cruncher, I can go to the website of the guy who maintains it and get a fairly-detailed explanation of the algorithms represented in the code. Hell you can even narrow down exactly which ISA extensions are used by the code modules. When I run something like Cinebench, I can see what the program is rendering, even if I can't see the underlying code of the render engine. I can still see what it's doing, more or less.

The CPU-z bench just spits out numbers. That's it. Maybe now they've gone to the trouble of describing the bench a little better, and if so, that's great. But it still doesn't explain why the old version of the bench had to be replaced by something that doesn't run as well on Ryzen.

The truth is (practically) never this interesting or exciting. I'm pretty sure CPUID saw an unexpected result, investigated it and discovered a bug.

Without seeing the source, we'll never really know.

 

[Ajay]

This is ridiculous. They haven't even given Ryzen enough time to see how its going to pan out.

http://www.marketwatch.com/story/amds-stock-plunges-toward-biggest-loss-in-over-12-years-2017-05-02

Investors don't care about Ryzen (well they do, but only as a driver for better profits). They care about companies that meet or exceed revenue and profit expectations. AMD fell short. If AMD turns this around, investors will go back in. Also, overnight stock changes of 25% get headlines.

 

[mattiasnyc]

For so many years now, people keep saying that the next AMD GPU will completely change the landscape and it never does.

'For so many years now, people keep saying that the next AMD CPU will completely change the landscape and it never does.'

Know what I mean? While the past provides us with data for assessing probabilities it doesn't necessarily tell us what will happen with 100% certainty.

 

[AtenRa]

I would wait for a few more days and then i would buy AMD stocks at close to $9

 

[ozzy702]

'For so many years now, people keep saying that the next AMD CPU will completely change the landscape and it never does.'

Know what I mean? While the past provides us with data for assessing probabilities it doesn't necessarily tell us what will happen with 100% certainty.

As much as I'm a fan of what AMD has done with Ryzen, it hasn't fundamentally changed the CPU landscape other than potentially bumping up timelines that were already set. We won't really know for another year how things look from a financial standpoint. I hope that Vega and Naples knock it out of the park and that between them and Ryzen AMD gains marketshare because competition is always what's best for the consumer. From an investor standpoint the stock price dump makes perfect sense, investors go where the money is and they'll be back if Vega and Naples success materializes.

 

[Headfoot]

What will Vega bring against the competition that previous designs didn't against it's competition at the time?

For so many years now, people keep saying that the next AMD GPU will completely change the landscape and it never does.

Except for the 5870/5850. Oh yeah, and the 7970. Oh yeah also the 290. But basically never, you're definitely right

 

[Leyawiin]

Except for the 5870/5850. Oh yeah, and the 7970. Oh yeah also the 290. But basically never, you're definitely right

The first two, yes. The third, no. Just... no.

 

[IllogicalGlory]

How do I know? Whenever anyone produces any kind of benchmark, the burden of proof is on those who produce the benchmark to establish that the resultant figures mean anything. Without providing source code or any points of reference, the CPU-z bench was basically just . . . numbers that sort of fell in line with how fast we expected CPUs to be relative to one another.

When I run y-cruncher, I can go to the website of the guy who maintains it and get a fairly-detailed explanation of the algorithms represented in the code. Hell you can even narrow down exactly which ISA extensions are used by the code modules. When I run something like Cinebench, I can see what the program is rendering, even if I can't see the underlying code of the render engine. I can still see what it's doing, more or less.

The CPU-z bench just spits out numbers. That's it. Maybe now they've gone to the trouble of describing the bench a little better, and if so, that's great. But it still doesn't explain why the old version of the bench had to be replaced by something that doesn't run as well on Ryzen.

They explained why the bench had to be replaced. The benchmark had a bug in it that caused every architecture except Ryzen to stall unexpectedly; Ryzen wasn't necessarily excelling, the rest were being held back. It hadn't been discovered until Ryzen exposed this problem, prompting an update. It's a fairly new benchmark, it has been updated many times in the past and this is just the latest one. They didn't intend to have the code stall the way it did and it was hamstringing Intel and BD CPUs, so they tweaked benchmark, explained why they did it and what the benchmark does now.

Without seeing the source, we'll never really know.

Occam's Razor, innocent until proven guilty, etc. Do you have any reason to believe that it was a backroom deal from Intel? Did Intel ask CPUID to change their benchmark to one where the Ryzen 1800X convincingly outscores the 5960X, nips at the heels of the 6950X (it only wins in MT by 12.5%)? To make up an explanation about the unexpected code? Come on.

 

[sm625]

I would wait for a few more days and then i would buy AMD stocks at close to $9

The stock almost touched its 200 day sma. Just a few more cents and it would have. That is always a decent time to buy when a stock makes a move like this.

 

[sm625]

Investors don't care about Ryzen (well they do, but only as a driver for better profits). They care about companies that meet or exceed revenue and profit expectations. AMD fell short. If AMD turns this around, investors will go back in. Also, overnight stock changes of 25% get headlines.

Yep. Investors are extremely shortsighted borderline retarded. Just look at Amazon and Priceline. Investors do very simple equations in their head. Did AMD's new product increase profits for its first quarter? Well no, they arent even making any profit. Wait, why am I holding this turd when I can buy AMZN at 20000 PE? SELL Mortimer SELL. Does it matter that their situation is improving? Nope. Does it matter that no big names have even really bought their new processors yet? Nope.

 

[IllogicalGlory]

The first two, yes. The third, no. Just... no.

It brought $650 performance down to $400 and prompted NV to release the 780 Ti. Its victory and impact was undercut at every turn though. First with the garbage blower cooling giving it a reputation it would never shake off, then the mining craze and then with Maxwell so convincingly beating it.

 

[IntelUser2000]

Yep. Investors are extremely shortsighted borderline retarded. Just look at Amazon and Priceline. Investors do very simple equations in their head. Did AMD's new product increase profits for its first quarter? Well no, they arent even making any profit. Wait, why am I holding this turd when I can buy AMZN at 20000 PE? SELL Mortimer SELL. Does it matter that their situation is improving? Nope. Does it matter that no big names have even really bought their new processors yet? Nope.

Let's look in a less emotionally charged way.

People are very cautious about their investments. Even if the company itself is doing well, outside factors like collapse of the economy and terrorist attacks keep most people afraid to take risks. If you are going to anyway you'd put it in names that you can trust, or at least in terms of financials, stable for the last few years. It was true prior to these threats, it will be even more true nowadays.

If everyone knew 100% about the stock market and what to do, and did as they were supposed to based on the information the result is no one would make any money.

If you are "long" AMD, then all these fluctuations don't matter. If you are long you just have to hope the company doesn't get bankrupt or if you happen to have bought it at multi-year peaks hope that it doesn't get bought out at a price lower than the price you bought the share for.

 

[mattiasnyc]

As much as I'm a fan of what AMD has done with Ryzen, it hasn't fundamentally changed the CPU landscape other than potentially bumping up timelines that were already set.

Sure, but my point was just that the same argument was made ahead of the Ryzen launch and the expectations some people had were very low. But then the chips came out and now the same group is pretty much quiet. "Game changer" or "fundamentally changing the CPU landscape"? No. Clearly exceeding the expectations those who expected Bulldozer 2.0? Yes.

So, with today's AMD we maybe shouldn't just look at the recent past to predict the future.

 

[DrMrLordX]

They explained why the bench had to be replaced. The benchmark had a bug in it that caused every architecture except Ryzen to stall unexpectedly; Ryzen wasn't necessarily excelling, the rest were being held back. It hadn't been discovered until Ryzen exposed this problem, prompting an update.

That makes no sense. Why would the Ryzen score go down if they fixed a bug that caused the benchmark to stall on other microarchitectures? Did they just add the stall to Ryzen to "bring it in line" with the rest of the other chips when they couldn't figure out how to fix the bug on anything else?

Do you have any reason to believe that it was a backroom deal from Intel? Did Intel ask CPUID to change their benchmark to one where the Ryzen 1800X convincingly outscores the 5960X, nips at the heels of the 6950X (it only wins in MT by 12.5%)? To make up an explanation about the unexpected code? Come on.

Do I have any reason? Tell you what, you try answering that yourself before I take a whack at it. If I even bother.

 

[Veradun]

As I previously said, if you change how your benchmark do its thing not because it isn't able to do that, but because it gives you a result you don't want, thus adapting your benchmark to the result you want, then your benchmark is not a benchmark.

 

[Shivansps]

As I previously said, if you change how your benchmark do its thing not because it isn't able to do that, but because it gives you a result you don't want, thus adapting your benchmark to the result you want, then your benchmark is not a benchmark.

Or maybe was just a bug, and they fixed a bug. CPU-Z benchmark fills the L2, the unusual cache structure of Ryzen may cause problems.

Shocking news, new arch, software needs patching.

Seriusly, now many times a benchmark software was updated before because a platform was giving strange results? both too high and too low.

 

[Jan Olšan]

It just means a new version of the benchmark came out and the results are incomparable. Happens all the time, and you can keep using both versions if you want (since it is a synthetic one and not a production sw, where using old version makes less sense).

This said, I heard CPU-Z was also giving atypically high scores to FX cores too, is that right?

In that case it might not be just a case of Ryzen core only not tripping on something specific that Intel doesn't like. It is possible that if the benchmark was going optimally on AMD cores but was limited by some bottleneck on Intels, they felt is is unjust and wanted to make it so that it runs optimally on both. Question of course is, if that bottleneck isn't relevant in the wild too, which would make even the bottlenecked performance indicative of something real.

 

[maddie]

Or maybe was just a bug, and they fixed a bug. CPU-Z benchmark fills the L2, the unusual cache structure of Ryzen may cause problems.

Shocking news, new arch, software needs patching.

It's the opposite of what you propose.

They [CPU-Z] claimed that Ryzen was executing that particular piece of code much more efficiently that other CPUs, and that the code structure was not seen in many, if any, real world applications, so they adjusted the benchmark.

So, not a bug with Ryzen. Makes you wonder if anytime a CPU excells, you just adjust your handicap.

 

[maddie]

It just means a new version of the benchmark came out and the results are incomparable. Happens all the time, and you can keep using both versions if you want (since it is a synthetic one and not a production sw, where using old version makes less sense).

This said, I heard CPU-Z was also giving atypically high scores to FX cores too, is that right?

In that case it might not be just a case of Ryzen core only not tripping on something specific that Intel doesn't like. It is possible that if the benchmark was going optimally on AMD cores but was limited by some bottleneck on Intels, they felt is is unjust and wanted to make it so that it runs optimally on both. Question of course is, if that bottleneck isn't relevant in the wild too, which would make even the bottlenecked performance indicative of something real.

I hope you realize the contradiction in that statement.

 

[Topweasel]

It just means a new version of the benchmark came out and the results are incomparable. Happens all the time, and you can keep using both versions if you want (since it is a synthetic one and not a production sw, where using old version makes less sense).

This said, I heard CPU-Z was also giving atypically high scores to FX cores too, is that right?

In that case it might not be just a case of Ryzen core only not tripping on something specific that Intel doesn't like. It is possible that if the benchmark was going optimally on AMD cores but was limited by some bottleneck on Intels, they felt is is unjust and wanted to make it so that it runs optimally on both. Question of course is, if that bottleneck isn't relevant in the wild too, which would make even the bottlenecked performance indicative of something real.

This can go back and forth because Benchmarks goals are to attempt to create an tool that will accurately measure the performance between competing hardware and software in the fields that the Benchmarks test. So having results that are outside the norm do need to be evaluated and assessed and honestly the goal is getting as close to real world results as possible. But what needs to be evaluated is not the test itself but whether or not something is broken. Like using an intel specific code that just won't run correctly when an a non Intel CPU is generated (this was happening with Benches made using the Intel compiler back in the day). Or if it's trying to do something an the code breaks and the software reports a completely unrealistic result. What you shouldn't do is realize that your code written for testing previous architectures comes out and runs extremely well on it because whatever roadblock you were testing on other devices doesn't apply to this one anymore. That's important information that shouldn't be glossed over because the end result isn't expected.

Lets look at it another way as well. If a device was expected to be lower in Game X. But the results came in had it worse than expected. Now because you were already expecting results to be down. Reviews and Benchers just say the It's particularly bad at Game X. If the edge case when a device is underperforming expectations would not be investigated and fixed then that should be the same for when performance exceeded expectations.

 

[KompuKare]

The first two, yes. The third, no. Just... no.

All that IllogicalGlory already said, plus it was the last time when AMD not only matched Nvidia's perf/area but beat it since it was smaller than GK110 (only had half the DP performance vs un-crippled GK110 (Titan/Titan Black + the pro versions) but the 512-bit GDDR5 interface probably took up more space than GK110's 384-bit one). The BOM wasn't probably wasn't necessarily cheaper because of the 512-bit bus and also because while the skimmed on the cooler the reference board was the typical overspec'ed AMD one.
Execution was awful though with them releasing a part beyond the process's sweet spot (AMD really can't help themselves there, can they?) with a really poor cooler. R9 290 running at 800MHz with lower voltage actually had pretty good perf/watt but wasn't going to beat GTX 780. Nvidia's PR machine of course made sure every site knew how loud it was. Don't recall AMD ever spoiling Nvidia's launches like that even if Nvidia's reference coolers often throttle like crazy (but do so quietly I guess).