*** AMD "Zen" thread (inc AM4/APU discussion) ***

[d_brennen]

I remember Lisa Su saying specifically that there's was lots left on the table with the Zen architecture, probably as they needed a product ASAP and it was already very capable

 

[AndreiD]

There is, improving the memory controller and doing something about the fabric should reap some good IPC improvements. Pinnacle Ridge is still at least 7% behind Coffee Lake IPC according to The Stilt's tests, so it's not far off considering how abysmal Bulldozer and its derivatives were.

 

[AndreiD]

Anyone want 10-15% IPC boost and up to 16 cores for Zen 2?

Think that's extremely optimistic considering Zen+ only got a 1~2% IPC uplift over Zen. I want that to happen, but realistically I think we'll see 5% to maybe 10% at most if they by some miracle improve IF latency & the memory controller a lot. Core wise I don't think we'll see that much of a jump for AM4, power deliver is just lacking on too many motherboards, at most we'll see 12 cores via the CCX increasing to 6 cores each.

 

[humbug]

Think that's extremely optimistic considering Zen+ only got a 1~2% IPC uplift over Zen. I want that to happen, but realistically I think we'll see 5% to maybe 10% at most if they by some miracle improve IF latency & the memory controller a lot. Core wise I don't think we'll see that much of a jump for AM4, power deliver is just lacking on too many motherboards, at most we'll see 12 cores via the CCX increasing to 6 cores each.

3.5% actually just from tweaking the same architecture, Zen 2 is a proper improvement over Zen 1.

But even then they don't need to boost the IPC more than another 5% anyway given its pretty much on par with Coffeelake, what they do need to do is increase the clock speed and with that Global Foundries are citing a 40% uplift in performance, IE clock speed.

As for the core count, Starship is getting 12 core Chiplets, Rome 16 core Chiplets and this is 7nm vs 14nm, half the power consumption, so the 16 core Rome Chip is the same power levels as the current 8 core.

IPC.

Ryzen 3000 is gona be awesome

 

[Zaasina]

Would be quite amusing if these AMD Zen APU's beat Intels offerings, 50% of their profits come from the server market.

http://seekingalpha.com/article/3055346-data-center-group-drives-intels-profitscodes

It would be very strange!

 

[Panos]

All is well, what we need is better motherboards. We expected the 300 series to be "cheap" but this continues to 400 series.
The manufacturers are selling cheap made AM4 boards for the same price as the better engineered Intel ones.

 

[SupraWez]

Would be nice but really need to wait for the release and reviews.

Although I never understand the mindset of people on a tech forum saying they dont want more cores, its not like we are having to pay more for them either, they are cheaper

 

[humbug]

All is well, what we need is better motherboards. We expected the 300 series to be "cheap" but this continues to 400 series.
The manufacturers are selling cheap made AM4 boards for the same price as the better engineered Intel ones.

That isn't true, if you actually look at them you'll see that, if AMD's boards are bad Intel's are shocking, for example....

Intel, at best this is a 3 phase



We know this is a 4 phase

 

[humbug]

Actually the Intel one is also a 4 phase, i missed the one at the top.

The point is tho be careful about saying AMD components are cheap compared to Intel, too many people would love that to stick and Buildzoid needs to move on to looking at Intel's boards because they are no better.

 

[humbug]

This one caught my eye...

B450 TUF Gaming £95 and Z370 TUF Gaming £140, look that the CPU VRMs, they are identical on both boards.

 

[CAT-THE-FIFTH]

It seems rather weird the AMD motherboards are getting far more scrutiny than the Intel ones. It does make me wonder,especially after SKL-X and the fact Intel is now releasing 8C CPUs, which will push VRMs even more than previous generations of SKL uarch derived CPUs.

Edit!!

Look at the scores for the Core i7 8700 in this Hexus review:

https://hexus.net/tech/reviews/systems/120437-msi-infinite-a-8th/?page=3

Its running in a B360 board,which seems to be not upto the task,and holding back the CPU,so it can barely compete with a Ryzen 5 1600X in a better motherboard.

 

[humbug]

It seems rather weird the AMD motherboards are getting far more scrutiny than the Intel ones. It does make me wonder,especially after SKL-X and the fact Intel is now releasing 8C CPUs, which will push VRMs even more than previous generations of SKL uarch derived CPUs.

Edit!!

Look at the scores for the Core i7 8700 in this Hexus review:

https://hexus.net/tech/reviews/systems/120437-msi-infinite-a-8th/?page=3

Its running in a B360 board,which seems to be not upto the task,and holding back the CPU,so it can barely compete with a Ryzen 5 1600X in a better motherboard.

Where does Buildzoid post? what forum? i want to, well not call him out exactly but put some pressure on him to scrutinise Intel's boards, no differently to what he has already spent a year scrutinising AMD's boards.

 

[Panos]

Where does Buildzoid post? what forum? i want to, well not call him out exactly but put some pressure on him to scrutinise Intel's boards, no differently to what he has already spent a year scrutinising AMD's boards.

Actually Buildzoid likes the AMD CPUs, and buys the motheboards on his own, but he is critical of most of them on the VRM side. Check his channel.

https://www.youtube.com/channel/UCrwObTfqv8u1KO7Fgk-FXHQ

What I am interesting is the reason that the Intel boards he reviews are on Gamers Nexus, while in his channel has a lot of AMD work.

Also he is very vocal and critical because the £90 boards from MSI & Asus especially, are the same with the £170 ones.

Or the lies of Asrock and Gigabyte about their boards, which forced both companies to change their websites and state the correct facts. (like the digi PWM IR by Asrock while it was a non IR chip and not what was advertised either).

 

[humbug]

Actually Buildzoid likes the AMD CPUs, and buys the motheboards on his own, but he is critical of most of them on the VRM side. Check his channel.

https://www.youtube.com/channel/UCrwObTfqv8u1KO7Fgk-FXHQ

What I am interesting is the reason that the Intel boards he reviews are on Gamers Nexus, while in his channel has a lot of AMD work.

Also he is very vocal and critical because the £90 boards from MSI & Asus especially, are the same with the £170 ones.

Or the lies of Asrock and Gigabyte about their boards, which forced both companies to change their websites and state the correct facts. (like the digi PWM IR by Asrock while it was a non IR chip and not what was advertised either).

This is precisely my point, he has spent a lot of time recently laying hard into AMD's boards, which is fine, but Intel's boards are exactly the same and have been around a lot longer, why is it that he never bothered to scrutinise sub £150 Intel boards for years but as soon as AMD and back to being competitive he's absolutely scathing about every single one of their sub £150 boards that are exactly the same as Intel's sub £150 boards that he never bothered to look at.

 

[Panos]

This is precisely my point, he has spent a lot of time recently laying hard into AMD's boards, which is fine, but Intel's boards are exactly the same and have been around a lot longer, why is it that he never bothered to scrutinise sub £150 Intel boards for years but as soon as AMD and back to being competitive he's absolutely scathing about every single one of their sub £150 boards that are exactly the same as Intel's sub £150 boards that he never bothered to look at.

Must say I agree. The last board he did for intel was 2 days ago the X299 from EVGA, which is a monster board praising it for being mATX (no word on the X399 Taichi). Previous one are the top of the range Z370s and Z270s. Nothing middle or low range like the H series. -_-

 

[CAT-THE-FIFTH]

This is precisely my point, he has spent a lot of time recently laying hard into AMD's boards, which is fine, but Intel's boards are exactly the same and have been around a lot longer, why is it that he never bothered to scrutinise sub £150 Intel boards for years but as soon as AMD and back to being competitive he's absolutely scathing about every single one of their sub £150 boards that are exactly the same as Intel's sub £150 boards that he never bothered to look at.

Especially since many companies probably recycled designs from the Skylake era right up to now.

 

[AndreiD]

The Intel mobo situation is similar to AMD though, if you want a good VRM you have to jump on the top end, there's 1-2 mid range boards that might look decent VRM-wise, like the ASRock Z370 Extreme4 or the ASUS X370/X470 Prime Pro, but that's about it.
Intel's saving grace is that they only allow overclocking on Z370 boards, but I have no doubt that if you tried to overclock the upcoming 8 cores on the lower end Z370 boards you'd encounter a lot of issues like throttling or thermal shut down. Same with putting a 2700X on a basic B350/B450 motherboard, you can overclock on them, but very limited.

For either it's best to do some research beforehand and hopefully find community lists like these:
https://www.hardwareluxx.de/community/f12/pga-am4-mainboard-vrm-liste-1155146.html
https://www.hardwareluxx.de/community/f12/lga-1151-mainboard-vrm-liste-1175784.html

@humbug The Stilt actually did proper IPC testing and the difference is 1~2%: https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/page-72#post-39391302


That's with spectre patches on the Intel platforms. ER means extremities removed and >=256b are workloads which use wider than 128 bit instructions (which aren't really used in the consumer space, except for video encoding and a few more niche uses).

 

[humbug]

The Intel mobo situation is similar to AMD though, if you want a good VRM you have to jump on the top end, there's 1-2 mid range boards that might look decent VRM-wise, like the ASRock Z370 Extreme4 or the ASUS X370/X470 Prime Pro, but that's about it.
Intel's saving grace is that they only allow overclocking on Z370 boards, but I have no doubt that if you tried to overclock the upcoming 8 cores on the lower end Z370 boards you'd encounter a lot of issues like throttling or thermal shut down. Same with putting a 2700X on a basic B350/B450 motherboard, you can overclock on them, but very limited.

For either it's best to do some research beforehand and hopefully find community lists like these:
https://www.hardwareluxx.de/community/f12/pga-am4-mainboard-vrm-liste-1155146.html
https://www.hardwareluxx.de/community/f12/lga-1151-mainboard-vrm-liste-1175784.html

@humbug The Stilt actually did proper IPC testing and the difference is 1~2%: https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/page-72#post-39391302


That's with spectre patches on the Intel platforms. ER means extremities removed and >=256b are workloads which use wider than 128 bit instructions (which aren't really used in the consumer space, except for video encoding and a few more niche uses).

This is now the third time you posted this, its already been completely debunked as fake work by an Intel shill.
Stop posting it over and over again...

 

[AndreiD]

Read the article before calling it BS, it's a lot more comprehensive than taking one Cinebench slide and equating it to IPC, that just shows you do not understand how to compare IPC between different architectures.
The Mesh is better (higher bandwidth, lower latency) for higher core counts, Ring is better for lower core counts, it doesn't scale as well.
The Stilt didn't use gaming for his tests, which is where Skylake-SP takes a huge hit because of Mesh & it's new cache hierarchy. Skylake-SP has less L3 cache and it's non-inclusive, which is going to be non-ideal for gaming. Skylake-X in my opinion is a terrible buy for anything gaming related.
Though to be fair I think he should have included a few gaming tests in there, it would knock Skylake-SP IPC down a peg, but at the same time it would probably increase Coffee Lake's lead over Summit & Pinnacle Ridge.

 

[humbug]

Read the article before calling it BS, it's a lot more comprehensive than taking one Cinebench slide and equating it to IPC, that just shows you do not understand how to compare IPC between different architectures.
The Mesh is better (higher bandwidth, lower latency) for higher core counts, Ring is better for lower core counts, it doesn't scale as well.
The Stilt didn't use gaming for his tests, which is where Skylake-SP takes a huge hit because of Mesh & it's new cache hierarchy. Skylake-SP has less L3 cache and it's non-inclusive, which is going to be non-ideal for gaming. Skylake-X in my opinion is a terrible buy for anything gaming related.
Though to be fair I think he should have included a few gaming tests in there, it would knock Skylake-SP IPC down a peg, but at the same time it would probably increase Coffee Lake's lead over Summit & Pinnacle Ridge.

You told me this the second time you posted this, i read it the first time you posted it, i didn't need to read it a second time, i don't need to read it now.

Its obvious you're trying to contradict everyother performance review with this, but i don't see how you think that can work, i'll take my information from multiple sources and draw a conclusion from the consistency of that, this is an outlier, it contradicts everyother data set so of course it BS.