AMD's Future Chips & SoC's: News, Info & Rumours.

[juanrga]

A ~10% performance improvement means very little when your competitor is already 30--40% ahead in performance...

Pulling numbers from a dark place again, I see.

For all that I'd love for you to go hunt for a bazillion graphs that prove your point, I don't want to go look for the graphs that contradict it. It's just tiring to do.

Now, whatever you say, if they do indeed get a ~10% increase in brute performance due to (base) clock and other improvements that don't really alter Zen's inner shenanigans, it's a great gain given effort put. I'm pretty sure we'll see a decent bump for Zen v2 when it lands, so this refresh will be good to have until then. Like I already said, Intel is already filling up the line up in anticipation, so they're wary.

A 10% update was claimed by AMD. The 30--40% gap is from mainstream reviews.

 

[jdwii]

"30%-40%" performance means very little at what frequency at what task?

Core for core its more about 10-15% lower IPC with 4-4.2ghz top OCs vs intel which is getting 1ghz more clock speed in a lot of cases when it comes to overclocking.

Also Ryzen is far behind Intel in AVX2 operations

 

[juanrga]

"30%-40%" performance means very little at what frequency at what task?

Core for core its more about 10-15% lower IPC with 4-4.2ghz top OCs vs intel which is getting 1ghz more clock speed in a lot of cases when it comes to overclocking.

Also Ryzen is far behind Intel in AVX2 operations

The IPC gap span from 10% in benches more favorable to RyZen muarch to 30% in benches more favorable to CoffeeLake muarch.

The average IPC gap is around 15% in applications. If you multiply this IPC gap by the stock clock gap you obtain a performance deficit of 33.6%. This is the reason why a 8700k is able to match a 1800X in multithreaded applications such as Blender, Handbrake,... despite the Ryzen chip has 33% more cores.

In games the gap can be even higher. I have seen the 8700k beating the 1800X by more than 60% in some DX12 game.

 

[jaymc]

Eh yeah whatever...

I wonder if we factor in how much slower intel chips will be running because of meltdown/spectre patchs this will hamper their performance even further...
It actually seem's like Intel is going backwards instead of forwards in performance at the moment anyone notice that lol...
Plus it's the first engineering sample etc expect speeds to go up from here.

Seem's to me the gap is closing fast.. Hell by the end of the year Intel's performance lead will be gone in the data centre's (Per Core that is), they have already lost it in certain way's (core count/addressable memory/pci express lanes) depending on how you look at it really I guess.
An then Desktop will follow suit wohoo ! Good times eh

 

[jdwii]

Eh yeah whatever...

I wonder if we factor in how much slower intel chips will be running because of meltdown/spectre patchs this will hamper their performance even further...
It actually seem's like Intel is going backwards instead of forwards in performance at the moment anyone notice that lol...
Plus it's the first engineering sample etc expect speeds to go up from here.

Seem's to me the gap is closing fast.. Hell by the end of the year Intel's performance lead will be gone in the data centre's (Per Core that is), they have already lost it in certain way's (core count/addressable memory/pci express lanes) depending on how you look at it really I guess.
An then Desktop will follow suit wohoo ! Good times eh

"Ryzen 7 1800X vs. Core i7 8700K, Meltdown & Spectre Updates Benchmarked"
https://www.youtube.com/watch?v=ovCqcUwpVGc

Watching this now

Edit yeah basically look at the end and see the difference in 1080P medium for gaming 12:30 in

197/161=1.224 or 22.4% less performance on a 1800X based off a 12 game average at 1080P medium.

Then take a look at the all core turbo of an 8700K and 1800X i believe a 1800x is 3.8ghz for an all core turbo and a 8700K all core turbo is 4.3ghz so 4300/3800=1.132 or 13.2% so in this case i will take 22.4-13.2 and that number would equal 9.2%

So i conclude based off those tests Ryzen was 9.2% worse per cycle. I seriously say 10-15% less IPC on average when comparing Ryzen to Intel's latest i even can test this somewhat myself. But yes there is cases where IPC is worse or better then those numbers best to mainly ignore the outliers in IMHO, which is why i hate "up to" statements from any company.

Note that this was done with 3200mhz memory on both Coffeelake and Ryzen

 

[juanrga]

Eh yeah whatever...

I wonder if we factor in how much slower intel chips will be running because of meltdown/spectre patchs this will hamper their performance even further...
It actually seem's like Intel is going backwards instead of forwards in performance at the moment anyone notice that lol...
Plus it's the first engineering sample etc expect speeds to go up from here.

Seem's to me the gap is closing fast.. Hell by the end of the year Intel's performance lead will be gone in the data centre's (Per Core that is), they have already lost it in certain way's (core count/addressable memory/pci express lanes) depending on how you look at it really I guess.
An then Desktop will follow suit wohoo ! Good times eh

We already have a good amount of preliminary benchmarks. Kaby on W10 takes a performance hit of 0--2% in applications and 3--6% in games. It remains to be tested the performance hit of Spectre on AMD.

 

[juanrga]

Third lawsuit for AMD

https://globenewswire.com/news-release/2018/01/19/1296729/0/en/SHAREHOLDER-ALERT-Brower-Piven-Notifies-Investors-of-Class-Action-Lawsuit-And-Encourages-Those-Who-Have-Losses-In-Excess-Of-100-000-From-Investment-In-Advanced-Micro-Devices-Inc-Na.html

 

[jaymc]

Eh yeah whatever...

I wonder if we factor in how much slower intel chips will be running because of meltdown/spectre patchs this will hamper their performance even further...
It actually seem's like Intel is going backwards instead of forwards in performance at the moment anyone notice that lol...
Plus it's the first engineering sample etc expect speeds to go up from here.

Seem's to me the gap is closing fast.. Hell by the end of the year Intel's performance lead will be gone in the data centre's (Per Core that is), they have already lost it in certain way's (core count/addressable memory/pci express lanes) depending on how you look at it really I guess.
An then Desktop will follow suit wohoo ! Good times eh

"Ryzen 7 1800X vs. Core i7 8700K, Meltdown & Spectre Updates Benchmarked"
https://www.youtube.com/watch?v=ovCqcUwpVGc

Watching this now

Edit yeah basically look at the end and see the difference in 1080P medium for gaming 12:30 in

197/161=1.224 or 22.4% less performance on a 1800X based off a 12 game average at 1080P medium.

Then take a look at the all core turbo of an 8700K and 1800X i believe a 1800x is 3.8ghz for an all core turbo and a 8700K all core turbo is 4.3ghz so 4300/3800=1.132 or 13.2% so in this case i will take 22.4-13.2 and that number would equal 9.2%

So i conclude based off those tests Ryzen was 9.2% worse per cycle. I seriously say 10-15% less IPC on average when comparing Ryzen to Intel's latest i even can test this somewhat myself. But yes there is cases where IPC is worse or better then those numbers best to mainly ignore the outliers in IMHO, which is why i hate "up to" statements from any company.

Note that this was done with 3200mhz memory on both Coffeelake and Ryzen

It can't be just in gaming bench's...It stands to reason. An I'd like to see his numbers verified I'm not so sure about him anyway..
Intel should be hit worse as they are affected by all three variants. There was an article by Gordon Mah Ung... He did some testing in various situations an workloads.
Here it is actually: https://www.pcworld.com/article/3247847/computers/heres-how-much-the-meltdown-and-spectre-fix-hurt-my-surface-book-performance.html

Major performance hits in different workloads an especially in NVME performance as far as I recall.
I think the later patches may have reduced these numbers slightly but I haven't seen any new bench's. As I say Intel is affected by all three variants, AMD's is affected mostly by one variant of spectre an the other is an "optional" update.

It stands to reason Intel is gonna be hit harder as their "fixes" are disabling way more stuff in their processors than AMD's are...It would be nice to see some decent testing in various workloads with both CPU's in comparison.

Jay

 

[gamerk316]

Eh yeah whatever...

I wonder if we factor in how much slower intel chips will be running because of meltdown/spectre patchs this will hamper their performance even further...
It actually seem's like Intel is going backwards instead of forwards in performance at the moment anyone notice that lol...
Plus it's the first engineering sample etc expect speeds to go up from here.

Seem's to me the gap is closing fast.. Hell by the end of the year Intel's performance lead will be gone in the data centre's (Per Core that is), they have already lost it in certain way's (core count/addressable memory/pci express lanes) depending on how you look at it really I guess.
An then Desktop will follow suit wohoo ! Good times eh

"Ryzen 7 1800X vs. Core i7 8700K, Meltdown & Spectre Updates Benchmarked"
https://www.youtube.com/watch?v=ovCqcUwpVGc

Watching this now

Edit yeah basically look at the end and see the difference in 1080P medium for gaming 12:30 in

197/161=1.224 or 22.4% less performance on a 1800X based off a 12 game average at 1080P medium.

Then take a look at the all core turbo of an 8700K and 1800X i believe a 1800x is 3.8ghz for an all core turbo and a 8700K all core turbo is 4.3ghz so 4300/3800=1.132 or 13.2% so in this case i will take 22.4-13.2 and that number would equal 9.2%

So i conclude based off those tests Ryzen was 9.2% worse per cycle. I seriously say 10-15% less IPC on average when comparing Ryzen to Intel's latest i even can test this somewhat myself. But yes there is cases where IPC is worse or better then those numbers best to mainly ignore the outliers in IMHO, which is why i hate "up to" statements from any company.

Note that this was done with 3200mhz memory on both Coffeelake and Ryzen

It can't be just in gaming bench's...It stands to reason. An I'd like to see his numbers verified I'm not so sure about him anyway..
Intel should be hit worse as they are affected by all three variants. There was an article by Gordon Mah Ung... He did some testing in various situations an workloads.
Here it is actually: https://www.pcworld.com/article/3247847/computers/heres-how-much-the-meltdown-and-spectre-fix-hurt-my-surface-book-performance.html

Major performance hits in different workloads an especially in NVME performance as far as I recall.
I think the later patches may have reduced these numbers slightly but I haven't seen any new bench's. As I say Intel is affected by all three variants, AMD's is affected mostly by one variant of spectre an the other is an "optional" update.

It stands to reason Intel is gonna be hit harder as their "fixes" are disabling way more stuff in their processors than AMD's are...It would be nice to see some decent testing in various workloads with both CPU's in comparison.

Jay

Not necessarily. If the Specter patch has a significantly higher performance loss, then it may essentially "hide" the performance loss of Meltdown.

 

[jaymc]

Linus Torvalds: (Intel's) patches are COMPLETE AND UTTER GARBAGE:

https://lkml.org/lkml/2018/1/21/192

 

[-Fran-]

Heh, Linus always makes my day when he is angry about something.

In other news: http://www.tomshardware.com/news/amd-raven-ridge-apu-specifications,36375.html

Is there anything in there that we didn't already know?

Cheers!

 

[juanrga]

Eh yeah whatever...

I wonder if we factor in how much slower intel chips will be running because of meltdown/spectre patchs this will hamper their performance even further...
It actually seem's like Intel is going backwards instead of forwards in performance at the moment anyone notice that lol...
Plus it's the first engineering sample etc expect speeds to go up from here.

Seem's to me the gap is closing fast.. Hell by the end of the year Intel's performance lead will be gone in the data centre's (Per Core that is), they have already lost it in certain way's (core count/addressable memory/pci express lanes) depending on how you look at it really I guess.
An then Desktop will follow suit wohoo ! Good times eh

"Ryzen 7 1800X vs. Core i7 8700K, Meltdown & Spectre Updates Benchmarked"
https://www.youtube.com/watch?v=ovCqcUwpVGc

Watching this now

Edit yeah basically look at the end and see the difference in 1080P medium for gaming 12:30 in

197/161=1.224 or 22.4% less performance on a 1800X based off a 12 game average at 1080P medium.

Then take a look at the all core turbo of an 8700K and 1800X i believe a 1800x is 3.8ghz for an all core turbo and a 8700K all core turbo is 4.3ghz so 4300/3800=1.132 or 13.2% so in this case i will take 22.4-13.2 and that number would equal 9.2%

So i conclude based off those tests Ryzen was 9.2% worse per cycle. I seriously say 10-15% less IPC on average when comparing Ryzen to Intel's latest i even can test this somewhat myself. But yes there is cases where IPC is worse or better then those numbers best to mainly ignore the outliers in IMHO, which is why i hate "up to" statements from any company.

Note that this was done with 3200mhz memory on both Coffeelake and Ryzen

It can't be just in gaming bench's...It stands to reason. An I'd like to see his numbers verified I'm not so sure about him anyway..
Intel should be hit worse as they are affected by all three variants. There was an article by Gordon Mah Ung... He did some testing in various situations an workloads.
Here it is actually: https://www.pcworld.com/article/3247847/computers/heres-how-much-the-meltdown-and-spectre-fix-hurt-my-surface-book-performance.html

Major performance hits in different workloads an especially in NVME performance as far as I recall.
I think the later patches may have reduced these numbers slightly but I haven't seen any new bench's. As I say Intel is affected by all three variants, AMD's is affected mostly by one variant of spectre an the other is an "optional" update.

It stands to reason Intel is gonna be hit harder as their "fixes" are disabling way more stuff in their processors than AMD's are...It would be nice to see some decent testing in various workloads with both CPU's in comparison.

Jay

AMD is affected by two Spectre flaws and the vulnerability of AMD CPUs to Meltdown continues being "unclear", although Windows and linux are treating AMD CPUs as no vulnerable by now.

All three patches are optional. For instance giving the command "noibrs noibpb nopti" to the linux boot disables the three patches on Intel hardware.

 

[juanrga]

Eh yeah whatever...

I wonder if we factor in how much slower intel chips will be running because of meltdown/spectre patchs this will hamper their performance even further...
It actually seem's like Intel is going backwards instead of forwards in performance at the moment anyone notice that lol...
Plus it's the first engineering sample etc expect speeds to go up from here.

Seem's to me the gap is closing fast.. Hell by the end of the year Intel's performance lead will be gone in the data centre's (Per Core that is), they have already lost it in certain way's (core count/addressable memory/pci express lanes) depending on how you look at it really I guess.
An then Desktop will follow suit wohoo ! Good times eh

"Ryzen 7 1800X vs. Core i7 8700K, Meltdown & Spectre Updates Benchmarked"
https://www.youtube.com/watch?v=ovCqcUwpVGc

Watching this now

Edit yeah basically look at the end and see the difference in 1080P medium for gaming 12:30 in

197/161=1.224 or 22.4% less performance on a 1800X based off a 12 game average at 1080P medium.

Then take a look at the all core turbo of an 8700K and 1800X i believe a 1800x is 3.8ghz for an all core turbo and a 8700K all core turbo is 4.3ghz so 4300/3800=1.132 or 13.2% so in this case i will take 22.4-13.2 and that number would equal 9.2%

So i conclude based off those tests Ryzen was 9.2% worse per cycle. I seriously say 10-15% less IPC on average when comparing Ryzen to Intel's latest i even can test this somewhat myself. But yes there is cases where IPC is worse or better then those numbers best to mainly ignore the outliers in IMHO, which is why i hate "up to" statements from any company.

Note that this was done with 3200mhz memory on both Coffeelake and Ryzen

It can't be just in gaming bench's...It stands to reason. An I'd like to see his numbers verified I'm not so sure about him anyway..
Intel should be hit worse as they are affected by all three variants. There was an article by Gordon Mah Ung... He did some testing in various situations an workloads.
Here it is actually: https://www.pcworld.com/article/3247847/computers/heres-how-much-the-meltdown-and-spectre-fix-hurt-my-surface-book-performance.html

Major performance hits in different workloads an especially in NVME performance as far as I recall.
I think the later patches may have reduced these numbers slightly but I haven't seen any new bench's. As I say Intel is affected by all three variants, AMD's is affected mostly by one variant of spectre an the other is an "optional" update.

It stands to reason Intel is gonna be hit harder as their "fixes" are disabling way more stuff in their processors than AMD's are...It would be nice to see some decent testing in various workloads with both CPU's in comparison.

Jay

Not necessarily. If the Specter patch has a significantly higher performance loss, then it may essentially "hide" the performance loss of Meltdown.

Correct. Meltdown patch affects SSD performance by 6% the Spectre patch affects SSD by more than 40%.

 

[juanrga]

Linus Torvalds: (Intel's) patches are COMPLETE AND UTTER GARBAGE:

https://lkml.org/lkml/2018/1/21/192

Apart from Linus confounding IBPB and IBRS, I am not sure if he understands that the patches also apply to AMD. The default boot for older AMD processors (e.g. families 10h, 12h and 16h). ) is "ibrs 2 ibpb 1" and for new AMD processors is "ibrs 0 ibpb 2".

A much more serious discussion of those issues can be found is one of the replies from one of the guys has worked in those patches

https://lkml.org/lkml/2018/1/22/598

 

[8350rocks]

"30%-40%" performance means very little at what frequency at what task?

Core for core its more about 10-15% lower IPC with 4-4.2ghz top OCs vs intel which is getting 1ghz more clock speed in a lot of cases when it comes to overclocking.

Also Ryzen is far behind Intel in AVX2 operations

The IPC gap span from 10% in benches more favorable to RyZen muarch to 30% in benches more favorable to CoffeeLake muarch.

The average IPC gap is around 15% in applications. If you multiply this IPC gap by the stock clock gap you obtain a performance deficit of 33.6%. This is the reason why a 8700k is able to match a 1800X in multithreaded applications such as Blender, Handbrake,... despite the Ryzen chip has 33% more cores.

In games the gap can be even higher. I have seen the 8700k beating the 1800X by more than 60% in some DX12 game.

Source? I have not been able to find a single benchmark that shows Ryzen is behind 30% clock-for-clock. Highest I have seen clock-for-clock is 12%.

I am not talking about you taking some theoretical i3 benchmarks and trying to extrapolate some cocktail napkin math IPC gap. I am talking about processor to processor comparisons showing a gap in real world performance.

 

[Gon Freecss]

"30%-40%" performance means very little at what frequency at what task?

Core for core its more about 10-15% lower IPC with 4-4.2ghz top OCs vs intel which is getting 1ghz more clock speed in a lot of cases when it comes to overclocking.

Also Ryzen is far behind Intel in AVX2 operations

The IPC gap span from 10% in benches more favorable to RyZen muarch to 30% in benches more favorable to CoffeeLake muarch.

The average IPC gap is around 15% in applications. If you multiply this IPC gap by the stock clock gap you obtain a performance deficit of 33.6%. This is the reason why a 8700k is able to match a 1800X in multithreaded applications such as Blender, Handbrake,... despite the Ryzen chip has 33% more cores.

In games the gap can be even higher. I have seen the 8700k beating the 1800X by more than 60% in some DX12 game.

Source? I have not been able to find a single benchmark that shows Ryzen is behind 30% clock-for-clock. Highest I have seen clock-for-clock is 12%.

I am not talking about you taking some theoretical i3 benchmarks and trying to extrapolate some cocktail napkin math IPC gap. I am talking about processor to processor comparisons showing a gap in real world performance.

Hello there, bias.

The 6900K is >45% faster than the slightly higher clocked 1700X in WinRAR. The 7960X is also >50% faster than the 1950X despite having a lower base clock, and slightly slower all core turbo speeds.

Even in the tasks where AMD performs the best, they outperform Intel ever so slightly clock for clock.

 

[goldstone77]

Spoiler

"30%-40%" performance means very little at what frequency at what task?

Core for core its more about 10-15% lower IPC with 4-4.2ghz top OCs vs intel which is getting 1ghz more clock speed in a lot of cases when it comes to overclocking.

Also Ryzen is far behind Intel in AVX2 operations

The IPC gap span from 10% in benches more favorable to RyZen muarch to 30% in benches more favorable to CoffeeLake muarch.

The average IPC gap is around 15% in applications. If you multiply this IPC gap by the stock clock gap you obtain a performance deficit of 33.6%. This is the reason why a 8700k is able to match a 1800X in multithreaded applications such as Blender, Handbrake,... despite the Ryzen chip has 33% more cores.

In games the gap can be even higher. I have seen the 8700k beating the 1800X by more than 60% in some DX12 game.

Source? I have not been able to find a single benchmark that shows Ryzen is behind 30% clock-for-clock. Highest I have seen clock-for-clock is 12%.

I am not talking about you taking some theoretical i3 benchmarks and trying to extrapolate some cocktail napkin math IPC gap. I am talking about processor to processor comparisons showing a gap in real world performance.

Hello there, bias.

The 6900K is >45% faster than the slightly higher clocked 1700X in WinRAR. The 7960X is also >50% faster than the 1950X despite having a lower base clock, and slightly slower all core turbo speeds.

Even in the tasks where AMD performs the best, they outperform Intel ever so slightly clock for clock.

So, basically Intel is better tuned for compressing a rar files, vs. AMD better at decompressing a file. Which operation do 99.9% of common users most often run into? And how often do they use that process to begin with? Comparisons like these, valid, are laughable to be honest. It's like having a 1000% better IPC in a calculator program. Testing workloads most often used are what really matters to the majority of people. I've never been asked what processor do I need to zip my files faster. It's a rather ridiculous argument to make.

 

[goldstone77]

BTW the title of this thread:AMD's Future Chips & SoC's: News, Info & Rumours.

Let's stay on topic, which is News, Info & Rumours pertaining to AMD Future Chips & SoC's.

 

[-Fran-]

Linus Torvalds: (Intel's) patches are COMPLETE AND UTTER GARBAGE:

https://lkml.org/lkml/2018/1/21/192

Apart from Linus confounding IBPB and IBRS, I am not sure if he understands that the patches also apply to AMD. The default boot for older AMD processors (e.g. families 10h, 12h and 16h). ) is "ibrs 2 ibpb 1" and for new AMD processors is "ibrs 0 ibpb 2".

A much more serious discussion of those issues can be found is one of the replies from one of the guys has worked in those patches

https://lkml.org/lkml/2018/1/22/598

That's what you take from that?

Welp, doesn't surprise me.

 

[goldstone77]

Ryzen 5 2500U "Raven Ridge" in the test: AMD hunts Intel at 15 watts in the notebook
01/22/2018 14:00 indicator Jan-Frederik Timm et al.
https://www.computerbase.de/2018-01/amd-ryzen-5-2500u-test-raven-ridge/

tl; dr: With the combination of Zen CPU cores and Vega GPU aka "Raven Ridge" AMD put an end to years of inferiority in the notebook. In the test, the Ryzen 5 2500U in the Acer Swift 3 can dup the competition from Intel not only in games, but well chilled in applications beat.

Conclusion
Raven Ridge in the 15-watt TDP class, even in the form of the smaller Ryzen 5 2500U, can thin the air for Intel's Kaby Lake Refresh competitor: the four Zen cores appear to consume slightly more power at full load, but are in no way inferior to the four cores of competition in terms of performance; and the integrated Vega GPU leaves Intel's HD Graphics no chance. The driver for Raven Ridge was also convincing in the meantime, problems did not arise in any of the five games tested.

So it's no wonder that Intel took the step from two to four cores in its notebook in the last summer with its first processors. Kaby Lake with just two cores at 15 watts as opponent would have dropped the top dog far in all measurements.

Raven Ridge for compact notebooks is impressive
Overall, the first impression of AMD's Raven Ridge in the notebook is thus a very positive, although a question mark remains behind the different performance in the Acer Swift 3. Too fast was Ryzen 5 2500U in the best run but in no case, rather slow in the slowest. Tests with other devices will provide information here. This also applies to the power consumption. If it proves to be higher than Intel at full load, it could be an argument against Raven Ridge, at least in the form of the Ryzen 5 2500U, at least in super thin devices.

 

[jaymc]

Dell PowerEdge R7425 (EPYC SERVERS) Documentations Now Online:

http://www.dell.com/support/home/ie/en/iebsdt1/product-support/product/poweredge-r7425/manuals

 

[juanrga]

"30%-40%" performance means very little at what frequency at what task?

Core for core its more about 10-15% lower IPC with 4-4.2ghz top OCs vs intel which is getting 1ghz more clock speed in a lot of cases when it comes to overclocking.

Also Ryzen is far behind Intel in AVX2 operations

The IPC gap span from 10% in benches more favorable to RyZen muarch to 30% in benches more favorable to CoffeeLake muarch.

The average IPC gap is around 15% in applications. If you multiply this IPC gap by the stock clock gap you obtain a performance deficit of 33.6%. This is the reason why a 8700k is able to match a 1800X in multithreaded applications such as Blender, Handbrake,... despite the Ryzen chip has 33% more cores.

In games the gap can be even higher. I have seen the 8700k beating the 1800X by more than 60% in some DX12 game.

Source? I have not been able to find a single benchmark that shows Ryzen is behind 30% clock-for-clock. Highest I have seen clock-for-clock is 12%.

Examples showing about 30% IPC gap (7-zip, WinRAR, Adobe Lightroom,...) can be easily found in both mainstream and personal reviews

 

[juanrga]

Linus Torvalds: (Intel's) patches are COMPLETE AND UTTER GARBAGE:

https://lkml.org/lkml/2018/1/21/192

Apart from Linus confounding IBPB and IBRS, I am not sure if he understands that the patches also apply to AMD. The default boot for older AMD processors (e.g. families 10h, 12h and 16h). ) is "ibrs 2 ibpb 1" and for new AMD processors is "ibrs 0 ibpb 2".

A much more serious discussion of those issues can be found is one of the replies from one of the guys has worked in those patches

https://lkml.org/lkml/2018/1/22/598

That's what you take from that?

Welp, doesn't surprise me.

He explains why Linus is plain wrong. After correcting him about things he is confounding. He mentions AMD (apparently Linus doesn't understand what he criticizes applies to AMD as well) and explains why retpoline is not enough and IBRS is needed in *current* hardware.

 

[juanrga]

Heh, Linus always makes my day when he is angry about something.

In other news: http://www.tomshardware.com/news/amd-raven-ridge-apu-specifications,36375.html

Is there anything in there that we didn't already know?

XFR for the top model seems to be 4.2GHz. Apparently the memory controller can be overclocked up to DDR4-3600 stable.

There is an interesting speculation that says those chips are already using the 14nm+ process.

EDIT: Not just speculation. AMD James confirms

https://www.reddit.com/r/Amd/comments/7s56pd/the_upcoming_ryzen_apus_seem_to_already_be_the/dt2kxpt/

 

[gamerk316]

Spoiler

"30%-40%" performance means very little at what frequency at what task?

Core for core its more about 10-15% lower IPC with 4-4.2ghz top OCs vs intel which is getting 1ghz more clock speed in a lot of cases when it comes to overclocking.

Also Ryzen is far behind Intel in AVX2 operations

The IPC gap span from 10% in benches more favorable to RyZen muarch to 30% in benches more favorable to CoffeeLake muarch.

The average IPC gap is around 15% in applications. If you multiply this IPC gap by the stock clock gap you obtain a performance deficit of 33.6%. This is the reason why a 8700k is able to match a 1800X in multithreaded applications such as Blender, Handbrake,... despite the Ryzen chip has 33% more cores.

In games the gap can be even higher. I have seen the 8700k beating the 1800X by more than 60% in some DX12 game.

Source? I have not been able to find a single benchmark that shows Ryzen is behind 30% clock-for-clock. Highest I have seen clock-for-clock is 12%.

I am not talking about you taking some theoretical i3 benchmarks and trying to extrapolate some cocktail napkin math IPC gap. I am talking about processor to processor comparisons showing a gap in real world performance.

Hello there, bias.

The 6900K is >45% faster than the slightly higher clocked 1700X in WinRAR. The 7960X is also >50% faster than the 1950X despite having a lower base clock, and slightly slower all core turbo speeds.

Even in the tasks where AMD performs the best, they outperform Intel ever so slightly clock for clock.

So, basically Intel is better tuned for compressing a rar files, vs. AMD better at decompressing a file. Which operation do 99.9% of common users most often run into? And how often do they use that process to begin with? Comparisons like these, valid, are laughable to be honest. It's like having a 1000% better IPC in a calculator program. Testing workloads most often used are what really matters to the majority of people. I've never been asked what processor do I need to zip my files faster. It's a rather ridiculous argument to make.

Compression tends to be a more serial based operation, and is a good test for single core performance.

Decompression tends to be a more parallel based operation, and is a good test for total CPU performance.