AMD Ryzen 2000 (12nm Zen+) expectations

[gOJDO_n]

Guys, If somebody is new on the forum that doesn't mean he is new in life and noob in computers and hardware. Some "teachers" here are asking me off-topic questions about my knowledge and are making assumptions about me. I opened the thread to talk about Ryzen+, not about me. If you really are knowledgeable on the topic and have more accurate vision, please contribute by expressing your own thoughts on the topic, not about me! We are here to share some knowledge and positive thoughts, we are not on an e-penis contest.
Thanks.

Wow, the 1600x "kicks the crap" out of the 7820x by a whopping 0.8%.

If you look at the average FPS, at 4K resolution, the 1600x is 1% faster than 7820X (99.6 / 98.8 = 1.01%) and that difference melts to 0.2% at UHD. At FHD the 7820X outperforms the 1600x by 0.7%, while at HD it outperforms the 1600X in average for 8.2%.
These are the minimal FPS @4K and @FHD:

So, we have 7820X which costs more than twice(let alone the mainboard which also is twice as expensive) and has 200% more RAM bandwidth, 33% more cores and 10% higher boost clocks performing as 1600X in games in average @FHD, @QHD and @4K. In my book, by definition, 1600X kicks the crap out of 7820X(core for core and clock for clock), but if you think that other term should be used, say it.

 

[frozentundra123456]

Guys, If somebody is new on the forum that doesn't mean he is new in life and noob in computers and hardware. Some "teachers" here are asking me off-topic questions about my knowledge and are making assumptions about me. I opened the thread to talk about Ryzen+, not about me. If you really are knowledgeable on the topic and have more accurate vision, please contribute by expressing your own thoughts on the topic, not about me! We are here to share some knowledge and positive thoughts, we are not on an e-penis contest.
Thanks.


If you look at the average FPS, at 4K resolution, the 1600x is 1% faster than 7820X (99.6 / 98.8 = 1.01%) and that difference melts to 0.2% at UHD. At FHD the 7820X outperforms the 1600x by 0.7%, while at HD it outperforms the 1600X in average for 8.2%.
These are the minimal FPS @4K and @FHD:

So, we have 7820X which costs more than twice(let alone the mainboard which also is twice as expensive) and has 200% more RAM bandwidth, 33% more cores and 10% higher boost clocks performing as 1600X in games in average @FHD, @QHD and @4K. In my book, by definition, 1600X kicks the crap out of 7820X(core for core and clock for clock), but if you think that other term should be used, say it.

Obviously the term is "gpu limited".

 

[IRobot23]

why are we/you doing this?
Intel vs AMD, beside TPU benchmarks are nonsense.

Hopefully, we get some new info about ryzen 2000

 

[gOJDO_n]

why are we/you doing this?

Because I'm interested about the subject and I want to share my thoughts with those like me.

 

[CatMerc]

https://www.computerpoweruser.com/article/24920/qa-with-james-prior?sf169542191=1

Thanks to the article we now know the voltage IF runs at. It's 0.7v. Pretty damn low. Could be why memory speeds are limited? IF is unstable at higher clocks?

 

[IRobot23]

Because I'm interested about the subject and I want to share my thoughts with those like me.

Here is subject about Ryzen 2000

 

[CatMerc]

Anyone know the minimum 14nm LPP voltage?

 

[epsilon84]

Guys, If somebody is new on the forum that doesn't mean he is new in life and noob in computers and hardware. Some "teachers" here are asking me off-topic questions about my knowledge and are making assumptions about me. I opened the thread to talk about Ryzen+, not about me. If you really are knowledgeable on the topic and have more accurate vision, please contribute by expressing your own thoughts on the topic, not about me! We are here to share some knowledge and positive thoughts, we are not on an e-penis contest.
Thanks.


If you look at the average FPS, at 4K resolution, the 1600x is 1% faster than 7820X (99.6 / 98.8 = 1.01%) and that difference melts to 0.2% at UHD. At FHD the 7820X outperforms the 1600x by 0.7%, while at HD it outperforms the 1600X in average for 8.2%.
These are the minimal FPS @4K and @FHD:

So, we have 7820X which costs more than twice(let alone the mainboard which also is twice as expensive) and has 200% more RAM bandwidth, 33% more cores and 10% higher boost clocks performing as 1600X in games in average @FHD, @QHD and @4K. In my book, by definition, 1600X kicks the crap out of 7820X(core for core and clock for clock), but if you think that other term should be used, say it.

Skylake X performs poorly in games relative to CFL due to the higher latency mesh bus design. CFL chips with the ringbus design generally run games much better.

When not GPU bound (4K) Coffee Lake based chips are significantly faster than the 1600X (or any Ryzen CPU for that matter), whether it be clock for clock (IPC), or in absolute terms:
https://www.computerbase.de/2017-10/intel-coffee-lake-8700k-8400-8350k-8100-test/5/
http://www.pcgamer.com/intels-8th-gen-processors-deliver-a-huge-jump-in-performance/
https://www.techpowerup.com/reviews/Intel/Core_i7_8700K/18.html

 

[raghu78]

Anyone know the minimum 14nm LPP voltage?

https://www.globalfoundries.com/sites/default/files/product-briefs/product-brief-14lpp.pdf

No idea about minimum. Nominal Voltage of 14LPP is 0.8v.

 

[CatMerc]

https://www.globalfoundries.com/sites/default/files/product-briefs/product-brief-14lpp.pdf

No idea about minimum. Nominal Voltage of 14LPP is 0.8v.

Hmm... Wouldn't be surprised if power efficiency really is why it was clocked as it was.

 

[The Stilt]

https://www.computerpoweruser.com/article/24920/qa-with-james-prior?sf169542191=1

Thanks to the article we now know the voltage IF runs at. It's 0.7v. Pretty damn low. Could be why memory speeds are limited? IF is unstable at higher clocks?

Since there are several domains running at the same speed, it could be any of them (which acts as the actual limit).
If the maximum MEMCLK is something that could have been improved by increasing a voltage, it most likely would have been done by now.
Most of the internal voltage domains run of a dLDO, meaning the voltage can be adjusted through the firmwares. 0.7V is so low that it wouldn't be limited by any of the usual dLDO sources used.

 

[IRobot23]

All they would need to do is fix that DF to 2,5-3Ghz+ for desktop AM4 processors. I wish it would as simple as that.

From this ryzen should run really good with higher Dram speeds and lower clocks.

 

[PhonakV30]

Skylake X performs poorly in games relative to CFL due to the higher latency mesh bus design. CFL chips with the ringbus design generally run games much better.

When not GPU bound (4K) Coffee Lake based chips are significantly faster than the 1600X (or any Ryzen CPU for that matter), whether it be clock for clock (IPC), or in absolute terms:
https://www.computerbase.de/2017-10/intel-coffee-lake-8700k-8400-8350k-8100-test/5/
http://www.pcgamer.com/intels-8th-gen-processors-deliver-a-huge-jump-in-performance/
https://www.techpowerup.com/reviews/Intel/Core_i7_8700K/18.html

He didn't talk about CFL , did he ? He Only said about Ryzen 1600 vs Core i7 7820X .

 

[raghu78]

All they would need to do is fix that DF to 2,5-3Ghz+ for desktop AM4 processors. I wish it would as simple as that.

From this ryzen should run really good with higher Dram speeds and lower clocks.

Yeah DF runs at memory controller speed. So for DDR4-2400 to DDR-3200 that is 1.2 Ghz - 1.6 Ghz. If AMD can improve the ratio of DF/Mem controller speed to 3:2 and support higher DDR4 4000 speeds they will achieve DF speeds upto 3 Ghz which should significantly reduce their memory latency.

 

[Topweasel]

Yeah DF runs at memory controller speed. So for DDR4-2400 to DDR-3200 that is 1.2 Ghz - 1.6 Ghz. If AMD can improve the ratio of DF/Mem controller speed to 3:2 and support higher DDR4 4000 speeds they will achieve DF speeds upto 3 Ghz which should significantly reduce their memory latency.

Not happening. No matter how important everyone thinks that IF speeds will fix what ails Ryzen (if that was really a thing). AMD isn't going to make any changes to it considering it's implementation isn't just limited to cross CCX communication.

It's a pipe dream and I don't know why people are so hooked AMD making this changes on what is primarily a process transition chip.

 

[raghu78]

Not happening. No matter how important everyone thinks that IF speeds will fix what ails Ryzen (if that was really a thing). AMD isn't going to make any changes to it considering it's implementation isn't just limited to cross CCX communication.

It's a pipe dream and I don't know why people are so hooked AMD making this changes on what is primarily a process transition chip.

Pinnacle Ridge is a client only die. AMD are not limited by the constraints of sharing this same die with the server cpus like the first generation Zen die was. We will find out what AMD was able to achieve at Pinnacle Ridge launch. Until then we can speculate freely.

 

[epsilon84]

He didn't talk about CFL , did he ? He Only said about Ryzen 1600 vs Core i7 7820X .

And that is a totally useless comparison because anyone who is actually going to play games will get the cheaper and faster CFL platform. Thats like comparing CFL to TR for gaming and saying TR sucks. It would be a totally useless comparison because you are comparing HEDT to a consumer desktop platform.

 

[itsmydamnation]

And that is a totally useless comparison because anyone who is actually going to play games will get the cheaper and faster CFL platform. Thats like comparing CFL to TR for gaming and saying TR sucks. It would be a totally useless comparison because you are comparing HEDT to a consumer desktop platform.

No, your going to get CFL or Ryzen because intels HEDT is a crap platform with over priced processors. Doesn't change the fact that those processors are in direct competition. Also your TR example is kind of fail because TR has a higher XFR clock then regular ryzen. So you can see plenty of review where TR gaming mode beats 1800x just from the extra clock.

Also your links dont really represent your position very well ~17% more clock speed for CFL vs 1600X~19% more 720P performance........

If AGESA 1.0.0.7 bring further latency improvements ( ether better memory support, faster memory or fabric performance) and PR can get XFR to 4.5 then CFL doesn't look very shiny any more........

 

[IRobot23]

Yeah DF runs at memory controller speed. So for DDR4-2400 to DDR-3200 that is 1.2 Ghz - 1.6 Ghz. If AMD can improve the ratio of DF/Mem controller speed to 3:2 and support higher DDR4 4000 speeds they will achieve DF speeds upto 3 Ghz which should significantly reduce their memory latency.

There is something going on with pinnacle ridge. I believe that is why intel rushed their z370 platform.

IF shoulf be extremely power hungry at those speeds, but nobody would mind extra 20W for insane reduce of memory latency. So AMD what is going to be like?

 

[gOJDO_n]

Guys, regardless of the "GPU Bottleneck" at any resolution, in some games CFL performs much(~10%) better, while in other Ryzen. So, there is still a lot of room for improvements on both architectures before the GPU becomes the only bottleneck. Ryzen performs quite good up to a certain framerate while there is enough bandwidth and the data is accessed "on time". Once the cores start to compete for bandwidth the CPU becomes a bottleneck. By doubling the DF bandwidth and halving the absolute latency, Ryzen will get a great boost in performance in bandwidth starved apps like games. If doubling the IF clock requires, let's say, doubling the operating voltage, that means its power consumption and heat output will be increased for 8 times which is not a. I hope that 2x IF clock can be achieved with voltage lower than 1.0V which means the power consumption of IF should up to 4 times higher.

As for SKL-X, I think the main issue is not the mesh clock, but the cache structure. Increasing the L2 size from 256kB to 1MB L2 is significantly increasing the L2 access latency. The L3 is also changed, which on SKL-X is non-inclusive and with reduced capacity per core.

 

[frozentundra123456]

Guys, regardless of the "GPU Bottleneck" at any resolution, in some games CFL performs much(~10%) better, while in other Ryzen. So, there is still a lot of room for improvements on both architectures before the GPU becomes the only bottleneck. Ryzen performs quite good up to a certain framerate while there is enough bandwidth and the data is accessed "on time". Once the cores start to compete for bandwidth the CPU becomes a bottleneck. By doubling the DF bandwidth and halving the absolute latency, Ryzen will get a great boost in performance in bandwidth starved apps like games. If doubling the IF clock requires, let's say, doubling the operating voltage, that means its power consumption and heat output will be increased for 8 times which is not a. I hope that 2x IF clock can be achieved with voltage lower than 1.0V which means the power consumption of IF should up to 4 times higher.

As for SKL-X, I think the main issue is not the mesh clock, but the cache structure. Increasing the L2 size from 256kB to 1MB L2 is significantly increasing the L2 access latency. The L3 is also changed, which on SKL-X is non-inclusive and with reduced capacity per core.

So you are saying improving the cpu will remove GPU bottlenecks? Just the opposite actually. The *GPU* needs to improve to remove the bottleneck at higher resolutions. And if you are saying in your first sentence that Ryzen performs better than CL in gaming, I would like to see some benchmarks (and not a < 1% difference in a gpu limited scenario, which you call "trouncing"). I think there may be one on two games which for some reason perform better on Ryzen, but overall CL is faster in almost every game.

 

[raghu78]

There is something going on with pinnacle ridge. I believe that is why intel rushed their z370 platform.

IF shoulf be extremely power hungry at those speeds, but nobody would mind extra 20W for insane reduce of memory latency. So AMD what is going to be like?

I think Pinnacle Ridge is going to be AMD's first Zen based CPU built on high performance libraries. I asked for important 12LP information through Daniel Nenni of semiwiki and GF refused to disclose 12LP info such as
1. Highest track height supported at 12LP
2. Largest Contacted Poly pitch at 12LP
3. Minimum Metal Pitch at 12LPP.

Summit Ridge used 14LPP HD libraries (CPP=78nm and MMP=64nm , 9T) because the same die was used for server and client and the priority was for the maximum density and lowest power as server chips were more important to AMD's growth prospects.

https://www.pcper.com/news/Processors/AMD-Details-Zen-ISSCC

14LPP had HP libraries (CPP=84nm, MMP=64nm, 9T) and UHP (CPP=84nm, MMP=64nm, 10.5T)
https://pc.watch.impress.co.jp/img/pcw/docs/733/713/html/9.jpg.html

With PR I think AMD has gone for maximum performance. I think AMD has gone for higher track height and relaxed CPP to achieve clock parity with Intel. We will know soon as AMD is going to start giving out PR information in bits and pieces starting at CES 2018. Going forward AMD will have separate dies for client and server and I expect 7nm Rome to be built at GF 7SoC to maximize density and power efficiency and Zen 2 desktop at 7HPC to maximize clock potential.

 

[Topweasel]

Pinnacle Ridge is a client only die. AMD are not limited by the constraints of sharing this same die with the server cpus like the first generation Zen die was. We will find out what AMD was able to achieve at Pinnacle Ridge launch. Until then we can speculate freely.

It's not just "Ryzen", but it is also possibly TR. Also as AMD has stated RR includes most of the changes for Zen+. On top of that AMD isn't going to invalidate all of their testing of the interlink protocol, for a one off die that is going to be replaced a year later that will be forced to confirm back to the standard they already set.

On top of that it's bad optic's to have that "potentially' big of an upgrade and then regress with Zen 2. Also I think it's a bit silly that people have latched on to this as the the only reason it's a little lacklustre in games. It probably has little to do with it.

IF isn't going to be touched.

 

[IRobot23]

It's not just "Ryzen", but it is also possibly TR. Also as AMD has stated RR includes most of the changes for Zen+. On top of that AMD isn't going to invalidate all of their testing of the interlink protocol, for a one off die that is going to be replaced a year later that will be forced to confirm back to the standard they already set.

On top of that it's bad optic's to have that "potentially' big of an upgrade and then regress with Zen 2. Also I think it's a bit silly that people have latched on to this as the the only reason it's a little lacklustre in games. It probably has little to do with it.

IF isn't going to be touched.

Why you think that IF is not going to be touched? higher core clock or IPC won't help much.
Even, if Ryzen 2 would have 15% higher IPC with 1067MHz DF clock at stock 2133MT/s DRam will be slower then ryzen with 2133MHz DF, DRAM 2133MT/s in games.

 

[raghu78]

It's not just "Ryzen", but it is also possibly TR. Also as AMD has stated RR includes most of the changes for Zen+. On top of that AMD isn't going to invalidate all of their testing of the interlink protocol, for a one off die that is going to be replaced a year later that will be forced to confirm back to the standard they already set.

On top of that it's bad optic's to have that "potentially' big of an upgrade and then regress with Zen 2. Also I think it's a bit silly that people have latched on to this as the the only reason it's a little lacklustre in games. It probably has little to do with it.

IF isn't going to be touched.

Pinnacle Ridge is not a one off die. Going forward server and client will have separate purpose built dies. 7nm Rome will be built at GF 7SoC 6T for maximum density and power efficiency while 7nm Ryzen will be built using 7HPC for maximum clock potential. The changes AMD makes in Pinnacle Ridge are not one off. Rather they are the beginning to a dedicated client die cadence.