Official AMD Ryzen Benchmarks, Reviews, Prices, and Discussion

[piesquared]

Looks like gaming performance is already on the way up even before developers have made optimizations for Ryzen. Sounds like they are working very hard at getting those optimizations in place. Over 300 devs with kits now and 1000 by the end of the year. And people cant wait 1 month for AMD to give developers enough time to work with Ryzen. WHICH IS A BRAND NEW ARCHITECTURE. It already performs very well with slightly lower performance in low res gaming, which is going to be improved. Seriously? Thats all the special interest groups can come up with as a negative for Ryzen? Thats excellent news!

Also, WTF is up with this guy in this video?? Has everyone lost their freakin mind? He comes on to give an update of Ryzen, then proceeds to read official email sent from AMD giving recommendations on how to set up this properly, then completely ignores (no, actually dismisses!) it and goes on to benchmark a game that wasnt on the list of optimized games! Isnt this what we are waiting to see? I know i sure am. Gotta be kidding me. Has this culture gone ADHD with the attention span of a sheep?

 

[piesquared]

Yeah, right now it seems like the 1700 is what people should be getting. The extra ~$100 - ~$200 you spend for a 1700X or 1800X don't provide anywhere near that much value since it seems like most won't get past 4.1 GHz.

Do you lose the power saving tech in SenseMI if overclocking? If so that is one reason i would buy a 17 or 1800x over non x. (higher base clock)

 

[Mockingbird]

Do you lose the power saving tech in SenseMI if overclocking? If so that is one reason i would buy a 17 or 1800x over non x.

Yes. No Turbo, no SenseMI, etc.

 

[Mockingbird]

If thats what actually happens we have a problem here, this actually works in a similar way to have a 2 quad cores cpus.

Windows scheduler update probably gona use Ryzen as it would for a dual socket system.

The real solution is for AMD to reduce latency between the two CCX rather than for developers to come up with some workarounds.

 

[Shivansps]

The real solution is for AMD to reduce latency between the two CCX rather than for developers to come up with some workarounds.

The scheduler must be updated anyway... most games today does not even do "manual affinity" anymore, they just spawn threads and let the scheduler handle the mess, that is also cross-platform effective. That thing of "games must be optimised" for Ryzen just not gonna happen because manual afinity is rare today, that is just old.

AMD knew this, there is no way they did not know that the Windows scheduler had to be updated, i think they are worried that people may call them not true octa cores. Not working with Microsoft to get an update ready before launch was a AMD decision, for sure.

 

[Mockingbird]

The scheduler must be updated anyway... most games today does not even do "manual affinity" anymore, they just spawn threads and let the scheduler handle the mess, that is also cross-platform effective. That thing of "games must be optimised" for Ryzen just not gonna happen because manual affinity is rare today, that is just old.

AMD knew this, there is no way they did not know that the Windows scheduler had to be updated, i think they are worried that people may call them not true octa cores. Not working with Microsoft to get an update ready before launh was AMD desicion, for sure.

If games just spawn a lot of threads, how would the Windows scheduler knows which cores on which CCX to choose?

 

[Shivansps]

If games just spawn a lot of threads, how would the Windows scheduler knows which cores on which CCX to choose?

Thats why the scheduler has to be updated for Ryzen, it probably gonna do the same thing it does for dual socket motherboards. Try to keep all threads of the same app inside 1 module, and move all else to the second module, only move threads from he same app to the 2nd module in special cases, like 1st module fully loaded, i dont completely know what the scheduler for dual socket boards.

And yes, that thing of game deciding where its gonna start a thread is old, it was used at the end of the dual core and beginning of the quad core era, but is not used anymore. By that time it was "ok i have 4 cores, i gonna use one for a physics thread, another for the graphics thread, another for the AI and the last for logic", and the game was starting 4 or 5 threads max. But that is just an old concept.

Today games just start all threads they need, and let the operative system choose were to place them. Thats the most efficient way, you have to account for every type of cpu(and future cpu) otherwise, and its not cross-platform friendly.

 

[Mockingbird]

Thats why the scheduler has to be updated for Ryzen, it probably gonna do the same thing it does for dual socket motherboards. Try to keep all threads of the same app inside 1 module, and move all else to the second module, only move threads from he same app to the 2nd module in special cases, like 1st module fully loaded, i dont completely know what the scheduler for dual socket boards.

That seems pretty pointless.

Why have eight-cores when you are just going to use four of them?

AMD needs to reduce the latency between the two L3 caches or combine them.

 

[itsmydamnation]

That seems pretty pointless.

Why have eight-cores when you are just going to use four of them?

AMD needs to reduce the latency between the two L3 caches or combine them.

Why do they? its an eviction cache..........
You dont even know the latency any way.....

Scheduler being correctly aware in more then enough.

 

[Mockingbird]

You dont even know the latency any way.....

It was talked about in another thread that the data fabric operates at the frequency that is half of that of the operating frequency of the memory controller.

 

[IllogicalGlory]

Much better way to visualize Ryzen's performance.

Here's the original review: http://drmola.com/pc_column/141286

Tested on a Radeon Pro Duo, pretty neat. Still, all these titles are 2013 or earlier, and there are only 5 games tested.

 

[itsmydamnation]

It was talked about in another thread that the data fabric operates at the frequency that is half of that of the operating frequency of the memory controller.

yes that is not latency.....

 

[imported_jjj]

Found some better gaming tests with res scaling, although maybe not the ideal battery of games and clocks.
http://lab501.ro/procesoare-chipseturi/review-amd-ryzen-7-1700-amd-ryzen-7-1700x-cpu-gaming/22

 

[AMDisTheBEST]

can anyone tell me what "Neural Net Prediction" and "Smart Prefetch" actually do?

 

[Shmee]

Well, I decided to go ahead and pick up a 1700 and some fast RAM today (2x8GB DDR4 3200 Patriot). Still need to find a decent mATX motherboard, hope they will be more available soon.

 

[sirmo]

can anyone tell me what "Neural Net Prediction" and "Smart Prefetch" actually do?

Just a branch prediction method employed in the Zen core. It uses a concept of perceptron (Neural Net) to learn and predict what future code branches will be. AMD claims 25% of the IPC uplift is due to this tech. An earlier iteration of it was used in Jaguar cores. That's all I know.

 

[piesquared]

http://www.eteknix.com/memory-speed-large-impact-ryzen-performance/

I hope the boards we be resupplied soon. This looks fun.

 

[CatMerc]

I don't see them giving up the low power consumption in favor of higher clocks. The real money is in the server market, which favors that aspect. I foresee we will continue to get trimmed down server cpus.

Here's the original review: http://drmola.com/pc_column/141286

Tested on a Radeon Pro Duo, pretty neat. Still, all these titles are 2013 or earlier, and there are only 5 games tested.

That's actually pretty damn good, and is the only test I've seen comparing using Radeon.

Someone should test that, see if perhaps NVIDIA's driver stack has a play in this.

 

[Mockingbird]

http://www.eteknix.com/memory-speed-large-impact-ryzen-performance/

I hope the boards we be resupplied soon. This looks fun.

The frequency of the data fabric is actually fixed to the frequency of the memory controller with a 1:2 ratio.

That's why faster memory is having such a big impact.

EDIT: That image turns out to be marketing material from MSI.

 

[piesquared]

The frequency of the data fabric is actually fixed to the frequency of the memory controller with a 1:2 ratio.

That's why faster memory is having such a big impact.

Yes so overclocking the memory boosts performance across the chip.

 

[DrMrLordX]

Why not push launch date by a week and avoid all this... It's beyond me.

Ryzen was supposed to launch in December. Then January. Then Feb 28th. Then March 2nd. They were already late.

According to what chew is saying over on XS, Ryzen's IMC/NB speed tracks perfectly with memory frequency. It affects much of Ryzen's internal IO performance. So you want to run the fastest RAM possible.

 

[.vodka]

Ryzen was supposed to launch in December. Then January. Then Feb 28th. Then March 2nd. They were already late.

I know.

Considering the concerning results are now in a completely different ballpark vs the launch reviews, what harm was one more week? We are still in Q1, so they wouldn't have broken any "this is the last time we delay" promises.

Chew is back? I'll go have a look.

 

[CatMerc]

http://www.phoronix.com/scan.php?page=article&item=amd-ryzen-cores&num=2

Interesting results.

 

[DrMrLordX]

I know.

Considering the concerning results are now in a completely different ballpark vs the launch reviews, what harm was one more week? We are still in Q1, so they wouldn't have broken any "this is the last time we delay" promises.

Well, look at it this way: considering how long many of us will be waiting for our motherboards, it looks like all those launch issues will be ironed out before we get the chance to build.

Chew is back? I'll go have a look.

Yup.

http://www.xtremesystems.org/forums/showthread.php?293130-Ryzen-Return-of-the-Jedi

http://www.xtremesystems.org/forums/showthread.php?292937-Ryzen-and-memory

 

[psolord]

http://www.eteknix.com/memory-speed-large-impact-ryzen-performance/

I hope the boards we be resupplied soon. This looks fun.

Haven't we seen this Witcher 3 benchmark before and it turned out to be fake anyway?