erek
[H]F Junkie
- Joined
- Dec 19, 2005
- Messages
- 10,894
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Zen 2 IPC, Clocks & Gaming Performance Improvements
- Thread starter erek
- Start date
Master_shake_
Fully [H]
- Joined
- Apr 9, 2012
- Messages
- 17,794
the comments on there are like cancer.
DedEmbryonicCe11
[H]ard|Gawd
- Joined
- Jun 6, 2006
- Messages
- 1,601
Just watch the original video it's not like the article adds anything important.
I think Don W. (Global Marketing Manager - Desktop processors) was being a bit generous citing that 4.1-4.2Ghz clock (when 24/7 Ryzen overclocks are typically in the 3.8-4Ghz). He is the marketing manager after all, so not a surprise he stretches the truth on some things. Still not an excuse for that though.
I'd switch to Ryzen instantly if the refresh assuredly resulted in a solid 500-700Mhz boost for Ryzen.
But without any real confirmation on that front, Ryzen is still going to have to address two stones for me - IPC *and* clockspeed.
Worst case scenario (Don's words) or not - I sincerely hope they can do something about clockspeed first.
Update - I remember Don's name coming up, and it was from an April 2017 statement regarding Vega:
http://www.tomshardware.com/reviews/amd-ryzen-ama,5018-11.html
So it's definitely not the first time he's positively spun things - sigh. Not a good look Don.
I'd switch to Ryzen instantly if the refresh assuredly resulted in a solid 500-700Mhz boost for Ryzen.
But without any real confirmation on that front, Ryzen is still going to have to address two stones for me - IPC *and* clockspeed.
Worst case scenario (Don's words) or not - I sincerely hope they can do something about clockspeed first.
Update - I remember Don's name coming up, and it was from an April 2017 statement regarding Vega:
http://www.tomshardware.com/reviews/amd-ryzen-ama,5018-11.html
So it's definitely not the first time he's positively spun things - sigh. Not a good look Don.
Last edited:
sirmonkey1985
[H]ard|DCer of the Month - July 2010
- Joined
- Sep 13, 2008
- Messages
- 22,414
if they move away from the ultra safe mobile 14nm process they used with ryzen higher clocks could definitely be a good possibility.. i think AMD ultimately did the right thing though going with the process they did just to make sure they had a high yield base line for the first introduction of ryzen. have absolutely nothing to complain about with my r5 1600, things a multi-tasking monster.
as far as the vega comment it all depends on the context it was said in.. since vega frontier which was the first vega release did pretty well against the ti/xp.. since his comment about vega wasn't really specific given what eventually was released it's to easy to spin it which ever way a reader wants to think it meant.
Last edited:
N4CR
Supreme [H]ardness
- Joined
- Oct 17, 2011
- Messages
- 4,947
From what 3rd parties have said about the 7nm process and to investors, up to 5GHz seems to be the jam..
So I would say your estimations would be well within reality, especially with a less low power (they are more efficient than Intel) process.
OrangeKhrush
[H]ard|Gawd
- Joined
- Dec 15, 2016
- Messages
- 1,673
I am sure more detailed articles will come out detailing more changes, however I do agree that Ryzen was a range finder and has done well. The detractors will spin up whatever they can but Ryzen on a performance assessment is very solid and a major step in the right direction for AMD's primary business
DuronBurgerMan
[H]ard|Gawd
- Joined
- Mar 13, 2017
- Messages
- 1,340
I wouldn't trust AMD's Global Marketing Manager if he told me water was wet. I'd have to check first.
But that being said, given the process improvements for Zen 2, I've no doubt we'll see *at least* a modest clockspeed boost. And AMD was telling the truth on their IPC improvements vs. Bulldozer with Zen, so it wouldn't surprise me if Zen 2 gets a modest IPC boost as well.
And TBH, that's all Zen really needs to keep AMD's competitive place in the market. Modest IPC and modest clockspeed improvements. If Zen 2 tops out at 4.3-4.5 instead of 3.8-4.2, and gets a modest 5% IPC boost, then it's in Skylake-X territory on clockspeed and IPC, and obviously will still carry AMD's tremendous pricing advantage on a per core basis. All to the good, IMHO.
But that being said, given the process improvements for Zen 2, I've no doubt we'll see *at least* a modest clockspeed boost. And AMD was telling the truth on their IPC improvements vs. Bulldozer with Zen, so it wouldn't surprise me if Zen 2 gets a modest IPC boost as well.
And TBH, that's all Zen really needs to keep AMD's competitive place in the market. Modest IPC and modest clockspeed improvements. If Zen 2 tops out at 4.3-4.5 instead of 3.8-4.2, and gets a modest 5% IPC boost, then it's in Skylake-X territory on clockspeed and IPC, and obviously will still carry AMD's tremendous pricing advantage on a per core basis. All to the good, IMHO.
CombatChrisNC
[H]ard|Gawd
- Joined
- Apr 3, 2013
- Messages
- 1,140
How about being able to run faster RAM, which also ties into less latency between modules? How about PCIe 4.0?
Just to mention that Zen2 is what was formely known as Zen+.
IPC increase over Zen has been known for a long time. It is not like if the article gives hot news...
It is not the first time that Glofo marketing dept. makes overoptimistic claims. Buldozer was also expected to bring into 5GHz thanks to PDSOI process...
And we know that Glofo already lies and overhypes the 7LP node.
drescherjm
[H]F Junkie
- Joined
- Nov 19, 2008
- Messages
- 14,941
I am with you on everything but "AMD's tremendous pricing advantage on a per core basis". I say currently there is not that big of a pricing advantage. I would put Intel's 10 core equivalent to about 14 AMD Zen cores in highly threaded loads overall. So there is an advantage but I would not call that tremendous. To me the biggest advantage is the platform. Although do you have a good point if you compare the prices of lga2666 workstation processor prices (so you can get ECC).
Last edited:
Presbytier
[H]ard|Gawd
- Joined
- Jun 21, 2016
- Messages
- 1,058
I personally am about ready to upgrade my PC, but I am thinking I might wait till next year to see if Zen 2 is comparable to Intel's 8th Gen releasing in a few weeks. A 6 core i7 is really tempting, but if Zen 2 performance is a good step above Zen then an 8 core Ryzen 7 zen 2 or whatever they call it might be a good upgrade. The big thing is them fixing the latency between infinity fabric.
There is no Zen2 next year. Next year is Pinnacle Ridge, which uses the same Zen cores than Summit Ridge.
DuronBurgerMan
[H]ard|Gawd
- Joined
- Mar 13, 2017
- Messages
- 1,340
You know, my man... I love this shit. I really do.
Of course we already knew about it - even if it hadn't been on the roadmaps since at least last year, it's pretty damned obvious to anyone with a couple of brain cells to put together. The article merely triggered folks talking and speculating about it again.
N4CR
Supreme [H]ardness
- Joined
- Oct 17, 2011
- Messages
- 4,947
Plenty of people got BD to 5GHz under OC.
If Zen2 is 4.5-4.8 which is about what I'm expecting, 5GHz will be no different to 4GHz with Ryzen now.
Intel will have some serious competition across the product stack if that is the case. 12 core Ryzen, 24 core Threadripper... 48 core Epyc all in 4-4.8GHz should cause your stocks some issues.
And yes they have claimed 2018 for Zen2 7nm for a long time (to investors which opens them up for lawsuits), but you and shintel seem to think otherwise for some strange reason.
Salting the sales channel won't work this time, guys.
Presbytier
[H]ard|Gawd
- Joined
- Jun 21, 2016
- Messages
- 1,058
I am still worried even with better clocks and IPC they could still run into issues with latency. Much of RYZEN's current issues in some, but not all, games stems from latency issues.
N4CR
Supreme [H]ardness
- Joined
- Oct 17, 2011
- Messages
- 4,947
Perhaps at DDR 2400 speeds yes, the inter-CCX latency is higher but with 3000-3200 which a majority of properly paired Ryzen rigs can do, it's within 10% of the 7900X... that's not going to significantly impact much. Most of this is compiler/optimisation tweaking which will become more and more commonplace. Same thing happened with the Opteron/Athlon dual cores, they required patches and once this was done the playing field was levelled again.
The intra-CCX latency is far lower than everything from intel, especially on a core count basis, bar fractionally slightly edged out by the 7700k... doubt you'll notice that at all.
2400
3200
Gideon
2[H]4U
- Joined
- Apr 13, 2006
- Messages
- 3,557
They are hardly worth replying to as they dont know anything. They have no clue or just dont want to admit that Intel will lose the process node advantage they have enjoyed forever. 10nm has been a nightmare for Intel and it seems many others had much better success with 7nm so Intel might have just messed up on this one. Zen2 will be here late 2018 on 7nm unless they have huge issues with EULV machines and getting them up to speed they also plan to have Navi as well, so it will be a big product launch if it all goes well. Should be interesting what they tweak with Zen2 and what it clocks to.
Presbytier
[H]ard|Gawd
- Joined
- Jun 21, 2016
- Messages
- 1,058
I play games at high refresh rates, but am beginning to feel choked on a 4 core processor for non gaming workloads. I'm wanting to see how well this 6/12 i7 from intel does.
OrangeKhrush
[H]ard|Gawd
- Joined
- Dec 15, 2016
- Messages
- 1,673
He knew about Zen2 10 years ago, this is just a postdiction.
Last week I was told that Zen+ was what will become Pinnacle Ridge and not Zen2 by same person. If you twist the truth enough you can get cross eyed.
Apparently Pinnacle Ridge will also be the same performance as Summit Ridge because same uArch, much like Sandy - Kabylake is the same uArch but there are gains.
Ole to more postdictions.
Last edited:
5GHz was the original target for stock settings. and only several years latter the target was achieved with the 220W Piledriver.
No one has made that claim. And the slides for investors carefully avoid writing 2018.
Not even close! Everyone said you that next year comes Zen on 14nm+. but for some reason you continue confounding Zen+ with 14nm+. It has been explained to you a dozen of times that next year is Pinnacle Ridge, which uses the same Zen core than Summit Ridge but uses a more mature process named 14nm+.
Zen2 is a 2019 product on 7nm.
Apparently you cannot read. Pinnacle Ridge is a refresh of Summit Ridge with higher clocks. The performance is not the same.
Pinnacle Ridge uses Zen cores. The muarch is the same.
Sandy and Kabylake are two different muarchs.
OrangeKhrush
[H]ard|Gawd
- Joined
- Dec 15, 2016
- Messages
- 1,673
Well then Zen+ on the slide may be for PR then and not Zen2 as so inferred. It is hard to make concessions on performance of PR without any data to this point.
If one assumes the on the fence approach PR may just be an AMD version of Tock so some low level tweaks here and there to reflect X% gains.
JustReason
razor1 is my Lover
- Joined
- Oct 31, 2015
- Messages
- 2,483
Wrong. Wait that wasn't right.
WRONG!!!
The original targets for BD was 6-8Ghz and that was because GloFo was touting that clockspeed possibility. When you look at the architecture it is easy to see that is what their target was, especially when looking at the memory and cache and goes a long ways to explain the preliminary expectation to route Intels offering of the time. 5Ghz is easy on most construction core CPUs with adequate cooling and we see how they fair there, no where near what AMD was reporting before release.
I thought about saying something yesterday when you stated this but figured... NAH. But here you go again. You know for someone who claims to know it all you know nothing on BD apparently.
N4CR
Supreme [H]ardness
- Joined
- Oct 17, 2011
- Messages
- 4,947
LisaSu_CEO@AMD said:
https://hothardware.com/news/amd-confirms-7nm-tape-out-2h-2017-navi-zen-2
This is a tock, practically a die shrink with some optimisations. Anyone who has been in the game for a while knows tape out this year, means Q4 2018 there will be something more than operational by then, if not beginning to enter retail channels.
Considering IBM has had its' fun and moved on to 5nm GF process, it should be smoother sailing than Zen 1.
Once again... Zen+ was renamed to Zen2.
I was talking about stock clocks; so it doesn't matter if one can get 5GHz with adequate cooling.
I was also talking about real targets, and 6--8GHz falls in the fantasy realm.
In the first place, she didn't mention Navi or Zen2, she simply mentioned tape out of products.
In the second place, RyZen tape out was in 2015 and anyone that predicted 2016 launch was wrong.
Last edited:
JustReason
razor1 is my Lover
- Joined
- Oct 31, 2015
- Messages
- 2,483
No you again are making a foregone conclusion ie: after the fact. The original target was 6-8Ghz, and again all the architecture proves this. It obviously failed to hit that target. The 5Ghz 9590 was to get publicity but changed absolutely nothing on the performance front. 9590 was never the TARGET just a last push for relevance.
DeathFromBelow
Supreme [H]ardness
- Joined
- Jul 15, 2005
- Messages
- 7,315
They delivered honest and substantial IPC/clockspeed improvements with each generation of Bulldozer-derived designs. If they can do the same thing here Intel is going to lose the server market.
KazeoHin
[H]F Junkie
- Joined
- Sep 7, 2011
- Messages
- 9,003
A good target for Z2 would be fixing whatever is holding the clockspeed at 4.0 GHz, allowing for higher turbo clocks, and having a separate clock devider for the fabric speed. With those two things alone, you would see a HUGE improvement stock Ryzen VS stock Z2. Overclocked with decent 32000 RAM, it would be much closer. If AMD can boost the OC headroom to the upper 4s, I would be happy.
Gideon
2[H]4U
- Joined
- Apr 13, 2006
- Messages
- 3,557
You and Shintai know nothing about what Zen+ is and yet you act like you do. Zen 2 has always been the 7nm release coming late 2018. Zen+ has always been continuing improvements on the 14nm node just like we ended up with on Bulldozer with higher clockspeeds or that is their internal code for a APU, either way AMD has not been clear on it's meaning and I just refer to previous practices of the company and it's not another release of Zen like you and Shintai thought. You think the fact that the 14nm and 14nm+ were listed in the same year arrow would be a clue it's not a Intel style Tic Toc thing.
gigaxtreme1
2[H]4U
- Joined
- Oct 1, 2002
- Messages
- 3,577
I think that 4 Ghz is a limitation of 14nmLPP, a low power process. 14nm+ or FX 14nm, whatever they are calling it will be a high power process. It has been demonstrated on HBM2 in conjuction with Rambus at Global Foundries in a recent press release. Whether this is the process to be used for Zen+ remains to be seen. Currently there is no 14nm HPP Process on the open Mosis schedule for Global Foundries. Maybe non published fab schedule has this ramping, but we don't know.
Fabric speed is CCX, Die interconnect and MCM interconnect speed. Supposedly it is so far limited to under 2 Ghz, like 1800. Somebody had RAM at 3600, so double IF speed. Look at the most recent press release from Glofo on their site. I'm posting from my phone, so it's hard to link right now.
Fabric speed is CCX, Die interconnect and MCM interconnect speed. Supposedly it is so far limited to under 2 Ghz, like 1800. Somebody had RAM at 3600, so double IF speed. Look at the most recent press release from Glofo on their site. I'm posting from my phone, so it's hard to link right now.
N4CR
Supreme [H]ardness
- Joined
- Oct 17, 2011
- Messages
- 4,947
Real targets eh? Do you know anything about Zen? Obviously not and you're here for FUD again as usual. Just like BD you got assblasted that so many do 5GHz easily. No doubt you'll just quote this part and shift goalposts more as usual.
Stock clocks of 1800 are 4GHz boost, OC all core to 4GHz is not uncommon as it's the silicon limitation and beyond 2nd critical, nice goalpost shifting bro.
Practically same will apply with Zen2 if they are saying 4-5GHz especially if it's not such a lower power and more efficient (than Intel) uarch again like Zen is. If you seriously think they'll have a regression with a die shrink, plus a less LPP process and not a huge jump in transistor count, you're barking up the wrong tree and your shilling is pretty counter productive here. You can get away with that BS on reddit, but this is [H].
Ryzen was a totally different and new uarch. Massively improved over BD, this 7nm shrink is practically a die shrink and some tweaks yet you think they're going to take 2 years again? Come on now.
Do you really think they're going to tape out and launch Navi q4 2018/Q1 2019 after just launching Vega? Or an 7nm APU without Navi? Again, that's just idiotic to anyone with experience and it looks like you shifted that goalpost to the swamp and got pretty muddy with that juan.
You have to FUD shit on everything AMD don't you, how many sheckels per post? Or they just paying you in shares? Where is your 79xx prediction blog? You don't seem to care at all about Intel for some strange reason, yet focus a petty amount of time on shifting goalposts and FUD'ing anything AMD, why is that?
Must be a stressful job with Zen around lol.[/quote][/QUOTE]
OrangeKhrush
[H]ard|Gawd
- Joined
- Dec 15, 2016
- Messages
- 1,673
I built a small system for my brother in law to join me online and to help with his electrical engineering degree.
It was a nothing fancy Ryzen 3 1200/wraith, GTX 1050ti setup on a Mortar B350 board that was on sale for the price of the Asus Prime A320 board so it was a steal at the price.
Anyway I didn't know what to expect performance wise to begin but quickly saw that it is a pretty solid system and plays everything well into the limitation of the GTX 1050ti without breaking a sweat.
I don't really understand the Ryzen doesn't game well, I found no issues with it or the performance.
My 4790K system will need updating soon, I will wait out PR information and decide whether i go CFL or PR.
It was a nothing fancy Ryzen 3 1200/wraith, GTX 1050ti setup on a Mortar B350 board that was on sale for the price of the Asus Prime A320 board so it was a steal at the price.
Anyway I didn't know what to expect performance wise to begin but quickly saw that it is a pretty solid system and plays everything well into the limitation of the GTX 1050ti without breaking a sweat.
I don't really understand the Ryzen doesn't game well, I found no issues with it or the performance.
My 4790K system will need updating soon, I will wait out PR information and decide whether i go CFL or PR.
pendragon1
Extremely [H]
- Joined
- Oct 7, 2000
- Messages
- 52,196
its just not quite as good as intel. but it still works just fine if you look past needing super high fps. when youre playing on a 60hz monitor who cares if it get 75fps instead of 85fps.
did you try OCing at all? if that cpu came with the mid-sized wraith spire you can push it a fair bit on just that.
No. The target was ~5GHz.
Zen+ and Zen++ were the sucesssors to Zen, and they were renamed to Zen2 and Zen3.
Next year comes Zen in 14nm+. Zen2 is a 2019 product. The new roadmaps have been given to you, and they demonstrate that there is no Zen+ anymore, but you continue in denial.
I know enough about Zen; in fact the 95% of what I predicted about Zen turned to be correct.
Once again. I am talking about stock clocks, not about overclocks.
I know, I predicted the 4GHz barrier before launch. And then people attacked me and pretended that the turbo would be 5GHz.
His claim was that Zen2 successor to 1800X will hit 5.7GHz stock. And I say this is BS.
Yes, Zen2 is a 2019 product, and that is why Pinnacle Ridge comes in 2018.
The article about RyZen predictions has a simple explanation. And I gave them to mods when I requested permission to add it to my signature. You couldn't be wrong.
Yes, 4GHz is a limit on the 14LPP, not in the Zen pipeline.
14nm+ is, so far as I know, just a tweaked, more mature, 14LPP. This 14nm+ node is used next year on Pinnacle Ridge, which use the same Zen cores than Summit Ridge.
There is no Zen+ anymore. This was renamed to Zen2.
There is an 14HP process at Glofo, but it is for IBM.
JustReason
razor1 is my Lover
- Joined
- Oct 31, 2015
- Messages
- 2,483
No the target was 6-8Ghz. Maybe your English is bad and this is a communication issue. Target is the end result when DESIGNING the architecture. 5Ghz would not have done enough against Intel at the time and based on the cache and memory architecture 6-8Ghz is indeed the original target.
At any rate your ignorance has absolutely no bounds and your attempt at using BD as proof of something proves you know absolutely nothing.
The target for Bulldozer was ~5GHz, which was the expected range of clocks that Glofo would get from the node. Your 6--8GHz is BS.
The target clock for Bulldozer is mentioned in the AT review of the FX-8150 (color added by me):
That is base clock, if we use the turbo clock we obtain 4.81GHz, which is the ~5GHz target I have mentioned.
The 30% target frequency is also mentioned in the report by Colin Donahue and Jason Lowden ("Clock speed" slide)
The reduction of more than 20% on the number of F04 inverter delays/cycle is mentioned in the Fischer et al. talk at the ISSC conference.
And no, I am not using Bulldozer "as proof" for anything. I know that 5.7GHz (stock) for Zen2 is BS.
Stop posting BS and stop insulting people that doesn't accept your BS.
Last edited: