[TPU] AMD Zen 2 Offers a 13% IPC Gain over Zen+, 16% over Zen 1 - Page 15 - Overclock.net - An Overclocking Community

Forum Jump: 

[TPU] AMD Zen 2 Offers a 13% IPC Gain over Zen+, 16% over Zen 1

Thread Tools
post #141 of 143 (permalink) Old 10-26-2018, 10:42 AM
Hey I get one of these!
KyadCK's Avatar
Join Date: Aug 2011
Location: Chicago
Posts: 7,297
Rep: 307 (Unique: 217)
Quote: Originally Posted by EniGma1987 View Post
We are talking architecture improvements for generations, not complete arch redesigns that focus on different things. The single anomaly is Bulldozer to Piledriver, and that happened because when BD released it had major node issues as well as a major arch flaw that required a patch to the cache system to workaround. Both the cache flaw was fixed and the node was fixed for the Piledriver release, so we saw a nice speed improvement.
As a side note, AMD did not gain 22% from Thuban to Bulldozer. Thuban was clocking a very consistent average OC of 4.2GHz and Bulldozer was doing 4.5-4.6GHz. Thats a 9% improvement. Since we are talking speed jumps from overclocking here, and the comparisons between BD and PD are overclocks, it is only fair to talk about Thuban's average OC when comparing it to BD as well.

It is possible we could see such a thing again given we have yet another cache issue holding back the speed of Zen arch. Though it isnt an actual "flaw" in the same way as BD had a real flaw and required a full extra copy operation to mirror data so the CPU didnt crash. Zen's cache issue is simply the design doesnt lend itself well to getting extra clocks beyond how it was designed. The cache has been tweaked, but even if AMD can tweak the design enough to allow greater clock headroom in the L2 cache it still wont be massive jumps in headroom.
Yea no, Thubans average was not 4.2, that was a good chip on a good day. If you want to go that route, then in the same category Bulldozer was 4.7-4.8 and PD was 4.9-5.0.

1100T: 3.3-3.7 stock
8150: 3.6-4.2 stock
8350: 4.0-4.2 stock
9590: 4.7-5.0 "stock"

So legitimately no idea where he got 22% from.

That said, BD's design took a MASSIVE change to queue depth and core design (ignoring any and all performance values) to achieve it's clock speed and they are not even remotely comparable to the Stars archs. If you want an example of clocks over time, Phenom II got better over time substantially, going from the 940 at 3Ghz to the 980 at 3.7Ghz scaling perfectly the whole way, though they got steppings, did not have any form of turbo, and certainly no form of boost/XFR to max out the chip from day one.

(18 items)
(7 items)
AMD Threadripper 1950X
Gigabyte X399 Designare
EVGA 1080ti SC2 Hybrid
EVGA 1080ti SC2 Hybrid
32GB G.Skill TridentZ RGB (4x8GB 3200Mhz 14-14-14)
Hard Drive
Intel 900P 480GB
Hard Drive
Samsung 950 Pro 512GB
Power Supply
Corsair AX1200
EK Predator 240
Corsair Graphite 780T
Operating System
Windows 10 Enterprise x64
2x Acer XR341CK
Corsair Vengeance K70 RGB
Corsair Vengeance M65 RGB
Sennheiser HD700
Sound Blaster AE-5
Audio Technica AT4040
30ART Mic Tube Amp
GTX 960m
16GB 1600 9-9-9-27
Hard Drive
Operating System
Windows 10 Pro
4k IPS
▲ hide details ▲
KyadCK is offline  
Sponsored Links
post #142 of 143 (permalink) Old 10-26-2018, 03:03 PM
Avid Memer
chessmyantidrug's Avatar
Join Date: Jun 2008
Location: Wichita Falls, TX
Posts: 5,475
Rep: 230 (Unique: 172)
Quote: Originally Posted by KyadCK View Post
AMD already said that clocks will not be significantly different. They will not get an extra 400mhz with Zen2, and you should not be getting people's hopes up like that.
The only hopes I'm getting up are my own. I'm not nearly as worried about clock speed as I am Infinity Fabric since that's the weakest link at the moment.

El Diablo
(19 items)
Intel Xeon X5670
Asus Sabertooth x58
Mushkin 6x2GB XP3-12800
Hard Drive
Samsung 850 PRO
Hard Drive
Samsung Spinpoint F1
Hard Drive
Samsung 850 EVO
Hard Drive
Toshiba P300
Optical Drive
Samsung SH-S203B
Power Supply
Corsair AX750
Thermalright Ultra 120 Extreme
Fractal Design Define R4
Operating System
Windows 10 Home 64-bit
BenQ RL2455HM
BenQ XL2420Z
Deck Legend - Fire
Razer Lachesis
SteelSeries QcK Gaming Mouse Pad
Astro A50
▲ hide details ▲
chessmyantidrug is offline  
post #143 of 143 (permalink) Old 10-26-2018, 04:57 PM
DRAM Obsessor
CJMitsuki's Avatar
Join Date: Mar 2018
Posts: 1,019
Rep: 7 (Unique: 6)
Quote: Originally Posted by chessmyantidrug View Post
The only hopes I'm getting up are my own. I'm not nearly as worried about clock speed as I am Infinity Fabric since that's the weakest link at the moment.

I agree that the clocks will not be 10%. My 2700x gets 4.558ghz balls to the wall max for ambient AIO cooling then the voltages just become unusable without sub-ambient cooling. With that being said, a 10% increase in clocks would put a golden chip (assuming mine is such a thing) over 5ghz. While I would love for this to be true its highly likely it will not. Real expectations to me are going to be 4.8ghz on golden chips and 4.5ghz normal maximum OC. Even with numbers similar to these at a 10-15% IPC increase would be massive. At 4.558ghz I max out in CB15 at 2102cb so if IPC gains scaled linearly with CB15 that would put that score at roughly 2415cb assuming 15% gain in IPC with theoretical linear scaling (unlikely) on ambient temps at the high end of the spectrum. Thats not taking into account the obvious latency drop for memory that should come with 7nm as the die is much smaller meaning less travel distance within the infinity fabric. This is also not taking into account better memory compatibility and just overall improvements to the IMC/SOC. Im going to guess 3933c14 will be a good memory overclock and should be attainable with fairly decent timings as Ryzen+ can already support 3800c14 but not at optimal timings like 3733c14 has been proven to run with. Its likely that if 3933c14 is possible with decent subtimings then 3800c14 with memory cooling and timings optimized to the maximum potential will be the way to go. Factoring in all of these theoretical gains could be quite amazing but its likely that you wouldnt see this until a bit later in the maturity of the new bios after several revisions are released.

There is cause for excitement even at clock speeds maxing lower than what im guessing as the real gains are going to be in the IPC improvements, memory latency decreasing (im guessing around 40-45ns) and memory compatibility supporting higher frequencies. Memory latency should be a bigger drop than what happened from first gen to Ryzen+ which from my own experiences was a drop of 8-10ns from 65ns on first gen to 55-57ns on 2700x. Die shrink plus overall improvements to the IMC and infinity fabric that is very likely to have occurred makes me lean toward the low 40ns area but I would be happy with 45ns since better gains are seen with latency decreasing after the performance from bandwidth plateaus a bit around 3466mhz area (56,000Gb/s) using max timings.

At the very least any improvement over the 2700x is great not only for AMD fans but Intel fans as well as it forces Intel to stop bending everyone over and scramble to push technology in an area that they had grown comfortable drip feeding consumers technology for profit. They got caught with their pants down which should serve as a reminder to Nvidia as well. Never know who could roll out the next high end gpu that competes with Nvidia at a lower price point. Then Nvidia will be where Intel is now, clawing their way out of the hole their complacency and greed has dug them into. Its like getting caught off guard with a punch to the breadbox, by the time you get done sucking air and stand up you already lost the match and have to work your way back up to the top ranks again. Intel is still sucking air and until they release a new architecture they arent going to be able to do much more with regards to pushing their performance to the next level. Intel is at the end of an architecture while AMD is barely scratching the surface of theirs. It isnt hard to see who is at a major disadvantage in this scenario as not only does Intel have to finish designing a new architecture, they also have to account for a die shrink and quite possibly not the 10nm they have been planning on. If they had to go to 7nm it wouldnt be a horrible thing but it seems like it would be a major setback and the CPU would have to be redesigned for the smaller die. Also, will Intel still be able to run their monolithic die with 7nm or 10nm? If they ended up having to adopt using technology similar to Ryzen their memory latency would suffer and bring them more inline. Ok, im getting carried away. Enough rambling

CPU- AMD Ryzen [email protected] *
GPU- EVGA GTX 1080ti SC2 w/ XOC [email protected] (Morpheus Cooler)
RAM- G.SKILL TridentZ 32gb (4 x 8gb) @ 3800mhz (14-15-15-24-38-1T)
COOLING- CORSAIR Hydro Series H115i Pro w/ Noctua NF-A14 (x7)
CASE- Silverstone RL06 Pro (Modified)

CJMitsuki is offline  

Quick Reply

Register Now

In order to be able to post messages on the Overclock.net - An Overclocking Community forums, you must first register.
Please enter your desired user name, your email address and other required details in the form below.
User Name:
If you do not want to register, fill this field only and the name will be used as user name for your post.
Please enter a password for your user account. Note that passwords are case-sensitive.
Confirm Password:
Email Address
Please enter a valid email address for yourself.
Email Address:


Currently Active Users Viewing This Thread: 1 (0 members and 1 guests)
Thread Tools
Show Printable Version Show Printable Version
Email this Page Email this Page

Forum Jump: 

Posting Rules  
You may post new threads
You may post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off
Trackbacks are Off
Pingbacks are Off
Refbacks are Off