Discussion AMD Ryzen MegaThread! FAQ and Resources

[juanrga]

Ryzen has already almost and in some made up it's 1080p gaming deficit and will only get better. Ryzen and gaming are all about fast memory . The way Ryzen works a couple of other clocks are directly tied to the memory speed one of them being the HT link/south bridge and has a direct effect on gaming. Most review sites either didn't have the support or bother taking the time to tune memory and just ran it between 2133 and 2400. That has a big negative effect cause it slows other clocks down also. Anyway look at the link 3-4 different sites saying going from 2400 to 3200 mem speed with Ryzen gets you 10-15% in 1080p gaming.Gskill has a Ryzen certified 3488 kit coming soon bet that thing will make Ryzen sing.
http://www.overclock.net/t/1625187/the-ryzen-gaming-performance-gap-is-mostly-gone

I checked the overclock link and they gave as 'proof' the performance scaling of RyZen on GB3 and ARMA3. What they don't say is that non-RyZen chips also scale in those benches. Skylake gets a 25% increase in framerates when going from 2133 to 3000 mem

The GB3 single core only increased by 9% on Ryzen when going from 2133 to 3466. GB single thread score changes by 3% on non-Ryzen chips with a 3000 --> 3600 overclock

 

[-Fran-]

Talking about memory.

I didn't see any memory scaling and arrangement in Ryzen reviews. Only Toms made a mention to the arrangements of banks and speeds, but no specific measurements (that I can remember).

I just want to know what would be the best memory config for Ryzen that minimizes latency for non-APUs and maximizes bandwidth for the upcoming APUs.

EDIT: Found two interesting videos in the AT forums that explain a bit how the CCX blocks arrangement can be a pain sometimes:

https://www.youtube.com/watch?v=JbryPYcnscA

https://www.youtube.com/watch?v=BORHnYLLgyY

It's interesting to see what the patches planned could do (in terms of improvements) for lower then 8 hard thread applications.

Cheers!

 

[NickatNight8320]

Day 1, Day 1 is what I am hearing.
I'm going to go back to Skylake and Broadwell Day 1 testing and compare day 1 numbers between the 3 archies.

 

[Redneck5439]

Won't the Ryzen chips with lower cores overclock more?

Conventional wisdom says that with 2 to 4 lesser cores and 4 to 8 lesser threads the R5 6 and 4 core processors should overclock better. This is also dependent on what silicon AMD uses for those processors. If they use their best silicon for the flagship 6 core 12 thread and 4 core 8 thread processors then they should overclock better- I see no reason why a 4 core R5 with the same silicon used for the R7s wouldn't be able to overclock to at least 4.5Ghz and probably better than that. With AMD trying to go toe to toe with i7 Skylake processors I see no reason why they wouldn't use their best silicon for their flagship 4 and 6 core R5s and try to close the gap with Skylake through more overhead speed. Keep in mind though that Skylake processors can overclock toward the 5Ghz range and I don't see Ryzen 6 or 4 core components getting past 4.8Ghz best case, realistically 4.6Ghz would be the most I would expect to see from those components if AMD uses their best binned silicon.

 

[dgothi]

I just build my new PC with Ryzen 7.
My PC Specs:
Ryzen 7 1700 (no OCing for now), Corsair Hydro Series H60 cool for Ryzen 7 CPU temp at average 36c, Asus x370 Pro motherboard, Corsair Vengeance 2X8GB DDR4-3000 with XMP enable, seasonic power supply 750 and EVGA Geforce 1070 SC. I did test benchmark two games - for honor and The Division.

For honor @ 1440p Benchmark resulted: https://goo.gl/photos/j4EF4qHJVcHo1mC57
The Division @ 1440p Benchmark resulted: https://goo.gl/photos/aE7tLKkHFF8RbEV19

I think this performancece is very well with Ryzen 7. also, I notice The Division's benchmark result show CPU average is only 38% and GPU average is 96%. GPU is loading a lot more than CPU. I assume no bottleneck for CPU because of multi-threading.

For my opinion, I think either Intel and AMD's new Ryzen both are very well. Nothing wrong either one of those you'd buy.

I am trying to find benchmark tools for some games. Do you know any good benchmark tools software? I'd test with TitanFall 2.

-DG

 

[Rookie_MIB]

I just bought a new core set as well. Going to outfit it with my GTX1060 6GB, move my GTX 960 to my home built mini box and sell the other one. Bought the R7-1700 with the Wraith LED cooler (I like it), the Asus B350M-A/CSM board, and the Corsair Vengeance LPX DDR4-3000 RAM. Should be here in a few days.

 

[truegenius]

( was it posted before? )
http://www.zolkorn.com/en/amd-ryzen-7-1800x-vs-intel-core-i7-7700k-mhz-by-mhz-core-by-core-en/view-all/
in it, they tested r7 after disabling its 4 cores and overclocking it to 4 ghz and downlocking i7 to 4ghz ( as r7 wasn't able to go above 4ghz ( even after disabling 4 cores ? ))

 

[dgothi]

I just bought a new core set as well. Going to outfit it with my GTX1060 6GB, move my GTX 960 to my home built mini box and sell the other one. Bought the R7-1700 with the Wraith LED cooler (I like it), the Asus B350M-A/CSM board, and the Corsair Vengeance LPX DDR4-3000 RAM. Should be here in a few days.

Let me know how is it after you get it. Hope you are happy with Ryzen.

-DG

 

[Redneck5439]

( was it posted before? )
http://www.zolkorn.com/en/amd-ryzen-7-1800x-vs-intel-core-i7-7700k-mhz-by-mhz-core-by-core-en/view-all/
in it, they tested r7 after disabling its 4 cores and overclocking it to 4 ghz and downlocking i7 to 4ghz ( as r7 wasn't able to go above 4ghz ( even after disabling 4 cores ? ))

I know that AMD and Ryzen "isn't there" just yet. They are still behind Intel, however it isn't by a truly significant margin anymore. I am just astounded how far AMD has come in one generation. The leap in performance over Piledriver is just amazing.

As far as overclocking, especially at this stage, goes Ryzen has a lot to work out before we know if it can be overclocked better than it appears now. There are bios updates, Windows updates, driver updates, all inbound that can have significant performance boosts for Ryzen. We may see no further overclock potential than we do now, but we may see at least some improvement with bios updates.

 

[Atreyo Bhattacharjee]

MERGED QUESTION
Question from Atreyo Bhattacharjee : "RYZEN Single Thread"

On PassMark it says the Ryzen 1700 gets a 1731 in single thread performance. This seems a bit unreasonable, is PassMark wrong or does the Ryzen 1700 just suck for single thread?

 

[8350rocks]

Take a look at this:
https://www.pcper.com/reviews/Processors/AMD-Ryzen-and-Windows-10-Scheduler-No-Silver-Bullet

Juan if that pans out right, we should be seeing a price drop soon enough (If people actually stop being fanboys).

Edit: I don't understand it that much, and I don't even know if the benchmarks are correct.

Yes, that PcPer link was given above and discussed in this same page.

@juan, AMD themselves have announced 1600x specifications of 3.6ghz base, 4ghz turbo. That is imo going to be the stand out gaming part for ryzen- given the same clocks it should match the r7s in games as 6 core / 12 threads is ample, but at a substantially lower price.

If you check the twitter discussion I gave the specs for the 1600X and the 1500X. If you check his response he said me there is no trace of those models and he speculates the timeline has changed.

I am not fully convinced that the R5 1600X will match the R7 1800X on games. Hardware.fr found that the six-core Broadwell was 4% behind the eight-core Broadwell despite having 6% higher clocks, and the 6C Ryzen has same clocks than the 8C Ryzen. If those percentages apply to Ryzen then the 1600X could be 10% behind the 1800X, but in the other hand the 1600X seems to have more L3 cache per core and that could reduce the gap. It is difficult to estimate accurate performance in this case.

That's a fair point- it's possible a 6c Ryzen might drop a bit of performance. Still, we're not talking much if it does drop a bit. The key (as always) is price, my take away from all this is whilst Ryzen doesn't really compete head to head with Skylake (which let's be honest few of us expected it to), it's close enough that priced well it makes a lot of sense. I do agree with many here in the fact that the main issue isn't the performance is terrible, but rather AMD shouldn't have pitched it as a Skylake competitor. To compare it in current titles against the 7700k was a very poor decision (and if they were going to do so- they should have had a few patched titles ready to back up their claims- I do think things like Ashes will potentially fly with a bit of work as it's the sort of title that can leverage a high core count cpu).

Reviewers compared it to the 7700K, because YT reviewers...

AMD compared it squarely to the 6900K, which is a pretty valid comparison.

AMD compared Ryzen to 7700k both in slides and demos...

Source?

No slide I can recall had a 7700K. They did the streaming demo showing a 6700K choking on a twitch stream...aside from that, I cannot recall a comparison to a non-E series intel processor.

 

[8350rocks]

MERGED QUESTION
Question from Atreyo Bhattacharjee : "RYZEN Single Thread"

On PassMark it says the Ryzen 1700 gets a 1731 in single thread performance. This seems a bit unreasonable, is PassMark wrong or does the Ryzen 1700 just suck for single thread?

The 1700 at stock clocks has a very low frequency. Overclocked it would even out with the 1700X/1800X depending upon what frequency you can get stable OCs.

 

[8350rocks]

( was it posted before? )
http://www.zolkorn.com/en/amd-ryzen-7-1800x-vs-intel-core-i7-7700k-mhz-by-mhz-core-by-core-en/view-all/
in it, they tested r7 after disabling its 4 cores and overclocking it to 4 ghz and downlocking i7 to 4ghz ( as r7 wasn't able to go above 4ghz ( even after disabling 4 cores ? ))

I know that AMD and Ryzen "isn't there" just yet. They are still behind Intel, however it isn't by a truly significant margin anymore. I am just astounded how far AMD has come in one generation. The leap in performance over Piledriver is just amazing.

As far as overclocking, especially at this stage, goes Ryzen has a lot to work out before we know if it can be overclocked better than it appears now. There are bios updates, Windows updates, driver updates, all inbound that can have significant performance boosts for Ryzen. We may see no further overclock potential than we do now, but we may see at least some improvement with bios updates.

Process maturity as well.

I would not be surprised to see processors from 6 months from now having a bit more headroom than the launch chips.

 

[AlexTheRoot]

Could owner of 1700/1700X/1800X test CPU using Indigo Bench? Here is link:
https://www.indigorenderer.com/indigobench
Unfortunately uploading isn't working right now, please post result here.

 

[gamerk316]

Well then:

https://www.pcper.com/reviews/Processors/AMD-Ryzen-and-Windows-10-Scheduler-No-Silver-Bullet

What began as a simple internal discussion about the validity of claims that Windows 10 scheduling might be to blame for some of Ryzen's performance oddities, and that an update from Microsoft and AMD might magically save us all, has turned into a full day with many people chipping in to help put together a great story. The team at PC Perspective believes strongly that the Windows 10 scheduler is not improperly assigning workloads to Ryzen processors because of a lack of architecture knowledge on the structure of the CPU.

In fact, though we are waiting for official comments we can attribute from AMD on the matter, I have been told from high knowledge individuals inside the company that even AMD does not believe the Windows 10 scheduler has anything at all to do with the problems they are investigating on gaming performance.

In the process, we did find a new source of information in our latency testing tool that clearly shows differentiation between Intel's architecture and AMD's Zen architecture for core to core communications. In this way at least, the CCX design of 8-core Ryzen CPUs appears to more closely emulate a 2-socket system. With that, it is possible for Windows to logically split the CCX modules via the Non-Uniform Memory Access (NUMA), but that would force everything not specifically coded to span NUMA nodes (all games, some media encoders, etc) to use only half of Ryzen. How does this new information affect our expectation of something like Naples that will depend on Infinity Fabric even more directly for AMD's enterprise play?

 

[jdwii]

( was it posted before? )
http://www.zolkorn.com/en/amd-ryzen-7-1800x-vs-intel-core-i7-7700k-mhz-by-mhz-core-by-core-en/view-all/
in it, they tested r7 after disabling its 4 cores and overclocking it to 4 ghz and downlocking i7 to 4ghz ( as r7 wasn't able to go above 4ghz ( even after disabling 4 cores ? ))

That's a nice article i would like to say that i believe the 4 and core Ryzen will have less L3 cache and these results meaning the 4 core part will look a little worse.

 

[-Fran-]

Well then:

https://www.pcper.com/reviews/Processors/AMD-Ryzen-and-Windows-10-Scheduler-No-Silver-Bullet

What began as a simple internal discussion about the validity of claims that Windows 10 scheduling might be to blame for some of Ryzen's performance oddities, and that an update from Microsoft and AMD might magically save us all, has turned into a full day with many people chipping in to help put together a great story. The team at PC Perspective believes strongly that the Windows 10 scheduler is not improperly assigning workloads to Ryzen processors because of a lack of architecture knowledge on the structure of the CPU.

In fact, though we are waiting for official comments we can attribute from AMD on the matter, I have been told from high knowledge individuals inside the company that even AMD does not believe the Windows 10 scheduler has anything at all to do with the problems they are investigating on gaming performance.

In the process, we did find a new source of information in our latency testing tool that clearly shows differentiation between Intel's architecture and AMD's Zen architecture for core to core communications. In this way at least, the CCX design of 8-core Ryzen CPUs appears to more closely emulate a 2-socket system. With that, it is possible for Windows to logically split the CCX modules via the Non-Uniform Memory Access (NUMA), but that would force everything not specifically coded to span NUMA nodes (all games, some media encoders, etc) to use only half of Ryzen. How does this new information affect our expectation of something like Naples that will depend on Infinity Fabric even more directly for AMD's enterprise play?

There's a reason AMD didn't make Ryzen 7 NUMA aware. We just need to wait and let AMD give us an answer for that.

I bet it has to do with the Fabric in one way or another, so making it NUMA would make the use of Fabric useless? Well, I think making the OS Scheduler NUMA aware for Ryzen is not the way to go.

Still, I think the CCX partitioning is interesting and it's well backed as an idea. I hope they have the quirks figured out.

( was it posted before? )
http://www.zolkorn.com/en/amd-ryzen-7-1800x-vs-intel-core-i7-7700k-mhz-by-mhz-core-by-core-en/view-all/
in it, they tested r7 after disabling its 4 cores and overclocking it to 4 ghz and downlocking i7 to 4ghz ( as r7 wasn't able to go above 4ghz ( even after disabling 4 cores ? ))

That's a nice article i would like to say that i believe the 4 and core Ryzen will have less L3 cache and these results meaning the 4 core part will look a little worse.

I would imagine Ryzen 7 derivates into Ryzen 5 will have the same or very close to the same L3 per CCX, but I wonder if they'll do the harvesting from the APUs... They should come with either less L3 or no L3 at all.

I'm getting kinda excited about the new APUs, I am surprised no one else is, haha.

Cheers!

EDIT: Typos.

 

[salgado18]

Spoiler

Well then:

https://www.pcper.com/reviews/Processors/AMD-Ryzen-and-Windows-10-Scheduler-No-Silver-Bullet

What began as a simple internal discussion about the validity of claims that Windows 10 scheduling might be to blame for some of Ryzen's performance oddities, and that an update from Microsoft and AMD might magically save us all, has turned into a full day with many people chipping in to help put together a great story. The team at PC Perspective believes strongly that the Windows 10 scheduler is not improperly assigning workloads to Ryzen processors because of a lack of architecture knowledge on the structure of the CPU.

In fact, though we are waiting for official comments we can attribute from AMD on the matter, I have been told from high knowledge individuals inside the company that even AMD does not believe the Windows 10 scheduler has anything at all to do with the problems they are investigating on gaming performance.

In the process, we did find a new source of information in our latency testing tool that clearly shows differentiation between Intel's architecture and AMD's Zen architecture for core to core communications. In this way at least, the CCX design of 8-core Ryzen CPUs appears to more closely emulate a 2-socket system. With that, it is possible for Windows to logically split the CCX modules via the Non-Uniform Memory Access (NUMA), but that would force everything not specifically coded to span NUMA nodes (all games, some media encoders, etc) to use only half of Ryzen. How does this new information affect our expectation of something like Naples that will depend on Infinity Fabric even more directly for AMD's enterprise play?

There's a reason AMD didn't make Ryzen 7 NUMA aware. We just need to wait and let AMD give us an answer for that.

I bet it has to do with the Fabric in one way or another, so making it NUMA would make the use of Fabric useless? Well, I think making the OS Scheduler NUMA aware for Ryzen is not the way to go.

Still, I think the CCX partitioning is interesting and it's well backed as an idea. I hope they have the quirks figured out.

Spoiler

( was it posted before? )
http://www.zolkorn.com/en/amd-ryzen-7-1800x-vs-intel-core-i7-7700k-mhz-by-mhz-core-by-core-en/view-all/
in it, they tested r7 after disabling its 4 cores and overclocking it to 4 ghz and downlocking i7 to 4ghz ( as r7 wasn't able to go above 4ghz ( even after disabling 4 cores ? ))

That's a nice article i would like to say that i believe the 4 and core Ryzen will have less L3 cache and these results meaning the 4 core part will look a little worse.

I would imagine Ryzen 7 derivates into Ryzen 5 will have the same or very close to the same L3 per CCX, but I wonder if they'll do the harvesting from the APUs... They should come with either less L3 or no L3 at all.

I'm getting kinda excited about the new APUs, I am surprised no one else is, haha.

Cheers!

EDIT: Typos.

Maybe because people here are enthusiasts, and look for the top performing parts?

Wait, I'm a bit lost here: NUMA is a OS/software thing, or does the CPU need to be able to handle it? The way I see it, the software must be smart and create threads with independence, then the OS handles the separation between cores. Am I missing something?

Also, question: I've seen on a reddit discussion that the connection between modules has a 22 GB/s bandwidth, but Xeon has a 20 GB/s connection too. Is it that hard to improve this, especially when the Xeon is a true dual-socket, but the Ryzen has two CPUs glued together very closely, and don't rely on the MB to talk to each other?

 

[-Fran-]

Maybe because people here are enthusiasts, and look for the top performing parts?

Wait, I'm a bit lost here: NUMA is a OS/software thing, or does the CPU need to be able to handle it? The way I see it, the software must be smart and create threads with independence, then the OS handles the separation between cores. Am I missing something?

Also, question: I've seen on a reddit discussion that the connection between modules has a 22 GB/s bandwidth, but Xeon has a 20 GB/s connection too. Is it that hard to improve this, especially when the Xeon is a true dual-socket, but the Ryzen has two CPUs glued together very closely, and don't rely on the MB to talk to each other?

Yes, NUMA is implemented in the OS (fully if you want), but the CPU must have a few bits and pieces so latency doesn't make the performance go to hell. In this case, AMD (AFAIK) didn't use regular "NUMA" implementations for Ryzen, so what is in the kernels currently is not optimal for Ryzen's CCX; that is why I mean. So, in short, if you enable the current implementations you have in the OS'es kernels, they won't improve Ryzen performance; I'd even say they'll put a penalty since they won't be using what Ryzen has built in to make effective management of each CCX communication through the Fabric... But that is a big fat assumption.

As for the chache, in my simple mind when you want to move stuff around, latency is king; good speeds are just a consequence of a good implementation usually. This is to say, Intel and AMD have different cache implementations (like, very different AFAIK), so the GB/s you see might have different latency underneath with different mechanisms to off side them.

Cheers!

 

[juanrga]

Won't the Ryzen chips with lower cores overclock more?

Conventional wisdom says that with 2 to 4 lesser cores and 4 to 8 lesser threads the R5 6 and 4 core processors should overclock better. This is also dependent on what silicon AMD uses for those processors. If they use their best silicon for the flagship 6 core 12 thread and 4 core 8 thread processors then they should overclock better- I see no reason why a 4 core R5 with the same silicon used for the R7s wouldn't be able to overclock to at least 4.5Ghz and probably better than that. With AMD trying to go toe to toe with i7 Skylake processors I see no reason why they wouldn't use their best silicon for their flagship 4 and 6 core R5s and try to close the gap with Skylake through more overhead speed. Keep in mind though that Skylake processors can overclock toward the 5Ghz range and I don't see Ryzen 6 or 4 core components getting past 4.8Ghz best case, realistically 4.6Ghz would be the most I would expect to see from those components if AMD uses their best binned silicon.

That "conventional wisdom" is based in the physical laws of silicon and standard binning. Intel chips are a good example

Chip ....................................... Thermal headroom ............... base clock
Broadwell 10C ..................... 14W/core ................................. 3GHz
Broadwell 8C ........................ 17.5W/core .............................. 3.2GHz
Broadwell 6C ........................ 23.3W/core .............................. 3.6GHz

The higher the thermal headroom the higher the stock clocks because each core can dissipate more. This is not happening with RyZen.

Chip ....................................... Thermal headroom ............... base clock
RyZen 8C ............................. 11.9W/core .............................. 3.6GHz
RyZen 6C ............................. 15.8W/core .............................. 3.6GHz
RyZen 4C ............................. 16.2W/core .............................. 3.5GHz

The higher thermal headroom doesn't allow higher stock clocks because cores cannot dissipate more. The reason for this odd behavior is that the top 1800X model doesn't represent average silicon, but it is a cherry picked golden silicon sample. Due to problems with 14LPP, which is not a node optimized for high frequencies. AMD was forced to pre-bin the best silicon for the low-volume sales 8C chips, in an attempt to achieve higher clocks; leaving the worse silicon for the rest of chips. This is the reason why the models with lower cores have smaller clocks. Consider the top quad-core 65W Ryzen. The base clocks would be 4.0--4.2GHz. In the worst case it would get the same 3.6GHz than the 1800X model; however, base clocks are limited to 3.5GHz and the turbo is limited even more: 3.7GHz instead 4.0GHz.

The lower core chips have lower clocks and will surely overclock worse than the 1800X.

Moreover, it seems that those quad-core chips with 3.5GHz base are already cherry picked models, because CanardPC has confirmed that the models that they got their hands on have base clocks of 3.2GHz and lower and lack all-core turbos.

 

[Atreyo Bhattacharjee]

MERGED QUESTION
Question from Atreyo Bhattacharjee : "RYZEN Single Thread"

On PassMark it says the Ryzen 1700 gets a 1731 in single thread performance. This seems a bit unreasonable, is PassMark wrong or does the Ryzen 1700 just suck for single thread?

The 1700 at stock clocks has a very low frequency. Overclocked it would even out with the 1700X/1800X depending upon what frequency you can get stable OCs.

If i was to OC to lets say 3.8-4.0 ghz, would I be on par with the i7 7700k?

 

[juanrga]

Source?

No slide I can recall had a 7700K. They did the streaming demo showing a 6700K choking on a twitch stream...aside from that, I cannot recall a comparison to a non-E series intel processor.

AMD gave us plenty slides comparing R7 to 7700k on multiple workloads including games

And some demos were leaked weeks before launch, which generated a lot of noise in hype sites as WCCFTECH which published 'rumors' about how RyZen was destroying the i7-7700k on games.

 

[dgothi]

Could owner of 1700/1700X/1800X test CPU using Indigo Bench? Here is link:
https://www.indigorenderer.com/indigobench
Unfortunately uploading isn't working right now, please post result here.

I can do that today. I have AMD Ryzen 7 1700 Will run it then let you know.

 

[-Fran-]

Source?

No slide I can recall had a 7700K. They did the streaming demo showing a 6700K choking on a twitch stream...aside from that, I cannot recall a comparison to a non-E series intel processor.

AMD gave us plenty slides comparing R7 to 7700k on multiple workloads including games

And some demos were leaked weeks before launch, which generated a lot of noise in hype sites as WCCFTECH which published 'rumors' about how RyZen was destroying the i7-7700k on games.

Curious they didn't use the 1800X in any of them!

JEEZ, I WONDER WHY.

Cheers!

 

[mertovski1907]

Hi everyone, my spec; ryzen 1700 @3.9gHz,asus prime x370, crucial 8x2G 2400mhz cl15 ram, and my cooler is xigmatek sd1264b. I am at @3.9ghz with 1.33750V and these are my temps;
Idle: between 35-42, average is 36-37
Aida64: 1 hour test, max is 76.1, average is 74-75
Battlefield 1: betwen 56-63,64 average is 60-61
I couldnt decide if my temps are good, i am someone new to overclock. I am waiting for your comments, should i keep my oc settings or not. Thank you all!!