AMD Briefly Shows Off Zen “Summit Ridge” Silicon

[moproblems99]

Intel continued to thrive under Pentium 4, AMD's fortunes under Bulldozer...not so much. AMD's x86 server market share - thanks to the server orientated Bulldozer/Piledriver is at an all time low of less than 0.7% - the lowest figure since AMD entered the server market with the debut of Opteron in 2003

I should have elaborated that I was referring to usability not financial success.

 

[RejZoR]

Rest my case about you sounding like a 12 year old man.... cause you can afford more than some people makes your opinion better? I'm not having a dig btw just saying the whole "I have this many cores and 64GB RAMS, so I am better than you" thing comes off as douchey man, it's not relevant, same as jeb ends in BMW's who think they can drive like wankers and are better than other road users....

Again I haven't read the whole case in point you two were arguing just you seem a lot better than trying to prove a point because your HW is much better than the other guys... that don't mean jack

Now let's all wait for Zen and speculate and opinionate, no need to get into the whole my dads bigger than your dad, he said she said playground stuff is there

Dude, I don't even know what you're banging on about...

 

[cadaveca]

I'm not so sure about that. AMD has publicly laid blame at Globalfoundries door on more than a few occasions - well deserved given Glofo's rather creative predictions, but AMD haven't been shy in letting the market know why the volumes and top bin parts were M.I.A.

Thanks for making what I said in sarcasm relevant. That was nearly 5 years ago. There are many aspects in which AMD can improve and thereby increase their success; not just in performance.

 

[EarthDog]

Darling, AMD has always been slower in SuperPi, even when it was a leader in performance in the golden era of Athlon XP. How do I know? Because I was overclocking those puppies and benching in SuperPi a lot back then. And they were always slower in seconds while totally raping Intel in everything that actually matters.

Darling?

That was just one example, poor one or not. There are plenty more sweetheart.

 

[the54thvoid]

it's easily over SB IPC if it's true

I should flippin' hope so. When Zen is out SB will be 5yrs old...

 

[Frag_Maniac]

I fail to see the difference between the P4 and Bulldozer debacles. You make it sound as if no one is getting by on Bulldozer and Steamrollers. While they aren't as good by any stretch of the means, you can still accomplish any task perfectly fine with Bulldozers and Steamrollers.

1. P4 was not holistically problematic for gaming, Bulldozer pretty much is, even in a lot of multi threaded games.

2. Bulldozer users aren't really "getting by" on a lot of AAA games that others are enjoying, and even a lot of those that used to swear by them are admitting they made a mistake buying it. I never complained about my P4 Northwood. If anything it gamed far longer than I anticipated it would.

3. I (as are most here) am talking about gaming and the people buying them for that purpose, not multi tasking.

4. (Even though you made this point to someone else), IPC is being repetitively mentioned, because AMD are basing their entire Zen campaign on it, and now just one specific title that is very Dx12 biased. And even at that, they're now saying they're "targeting", vs prior to that assuring a 40% increase.

 

[cdawall]

1. P4 was not holistically problematic, Bulldozer IS.

2. Bulldozer users aren't really "getting by" on a lot of AAA games that others are enjoying, and even a lot of those that used to swear by them are admitting they made a mistake buying it.

3. I (as are most here) am talking about gaming and the people buying them for that purpose, not multi tasking.

4. (Even though you made this point to someone else), IPC is being repetitively mentioned, because AMD are basing their entire Zen campaign on it, and now just one specific title that is very Dx12 biased. And even at that, they're now saying they're "targeting", vs assuring a 40% increase.

What are you talking about P4 users had just as bad of a time trying to run games as bulldozer users. The bulldozer architecture operates great in games that allow for multithreading. The only games with substantially frame rate issues are horribly coded and do not allow more than 1 or 2 cores to be utilized. This issue can be "fixed" by setting the core count by module which shuts off half the "cores"

 

[Frag_Maniac]

What are you talking about P4 users had just as bad of a time trying to run games as bulldozer users. The bulldozer architecture operates great in games that allow for multithreading. The only games with substantially frame rate issues are horribly coded and do not allow more than 1 or 2 cores to be utilized. This issue can be "fixed" by setting the core count by module which shuts off half the "cores"

Apparently you've never played on a P4 Northwood, and they came out well before dual core was popular enough to require one for most games. Hell, I even played original Crysis on my Northwood, and with max textures. I had a 4:3 CRT display that I set to 1200x900 for that game, with Med Shaders and Shadows, and only had to employ a few simple file edits to tweak Object Quality and view distance.

Keep in mind dual core was the first step of multi core CPU gaming, so it took a while for devs to adopt. I think the problem here though, is you're not putting each chip in it's own timeline. P4 Northwood fared WAY better in it's time, than Bulldozed is in it's.

 

[looncraz]

I'm pretty sure AMD is talking only low thread performance improvement too. I'm really doubting multi thread performance will see a 40% gain, especially when they're now saying they're "targeting" that, vs achieving it.

If anything, Zen will see its largest boosts in multi-threaded loads - most especially floating point multi-threaded loads... the reply to this continues below, but I also want to address:

I think that if Skylike is twice as fast as Bulldozer, then Skylake is 100% faster. If Zen is 40% faster than Bulldozer, then Skylake is 60% faster then Zen. No?

A doubling in performance over Bulldozer - even with 'just' a 40% IPC jump over Excavator - is easily believable.

Single threaded performance:
BD: 100%
PD: 109%
SR: 117%
XV: 128%
ZN: 180%

Multi-threaded:
ZN + SMT (15%~20%) = 207%~216%

I ignored all the multi-threaded improvements made in the construction core family as they should have little bearing on Zen's multi-threaded performance. Still, the scaling should be much better than what was seen with the construction family.

Check out benchmarks. Specifically the single threaded benchmarks, which when at the same clocks, gives you an idea of the difference in IPC. Look at bulldozer vs Skylake in Super Pi for example. I can take a nap running 32M on AMD.

Things aren't that way any more. Excavator, from which AMD's 40% boost is applied, performs rather well in SuperPi.

See here: http://excavator.looncraz.net/

I also made a few Zen performance estimates. The amount of math involved is glorious, but all my data is disclosed and linked on the site - even though I post all my images in relative terms.

 

[Frag_Maniac]

If anything, Zen will see its largest boosts in multi-threaded loads ...

Makes no sense really. Clearly Bulldozer's main caveat is poor low thread performance. Multi thread wise it does much better.

 

[the54thvoid]

If anything, Zen will see its largest boosts in multi-threaded loads - most especially floating point multi-threaded loads... the reply to this continues below, but I also want to address:



A doubling in performance over Bulldozer - even with 'just' a 40% IPC jump over Excavator - is easily believable.

Single threaded performance:
BD: 100%
PD: 109%
SR: 117%
XV: 128%
ZN: 180%

Multi-threaded:
ZN + SMT (15%~20%) = 207%~216%

I ignored all the multi-threaded improvements made in the construction core family as they should have little bearing on Zen's multi-threaded performance. Still, the scaling should be much better than what was seen with the construction family.



Things aren't that way any more. Excavator, from which AMD's 40% boost is applied, performs rather well in SuperPi.

See here: http://excavator.looncraz.net/

I also made a few Zen performance estimates. The amount of math involved is glorious, but all my data is disclosed and linked on the site - even though I post all my images in relative terms.

In a scientific sense, someone who pays (as you state in your opening page) to demonstrate a principle or process will be blindsided by their own bias.
Using maths to justify end conclusions is invalid when you don't know enough about Zen to compile accurate data.
Your work is commendable but the enthusiasm behind it also compromises it's neutrality and validity.

Only time will tell how well Zen does. And I do believe only the mad want it to fail. Intel's pricing structure needs a shake up (much like Polaris will hopefully do to Nvidia mid range).

 

[FlanK3r]

I should flippin' hope so. When Zen is out SB will be 5yrs old...

It is, but...compare clock to clock in pure tests SB to Skylake single thread (no test with AVX, AVX2 or XOP, FMA)...Difference? 7-10% max.

 

[EarthDog]

If anything, Zen will see its largest boosts in multi-threaded loads - most especially floating point multi-threaded loads... the reply to this continues below, but I also want to address:



A doubling in performance over Bulldozer - even with 'just' a 40% IPC jump over Excavator - is easily believable.

Single threaded performance:
BD: 100%
PD: 109%
SR: 117%
XV: 128%
ZN: 180%

Multi-threaded:
ZN + SMT (15%~20%) = 207%~216%

I ignored all the multi-threaded improvements made in the construction core family as they should have little bearing on Zen's multi-threaded performance. Still, the scaling should be much better than what was seen with the construction family.



Things aren't that way any more. Excavator, from which AMD's 40% boost is applied, performs rather well in SuperPi.

See here: http://excavator.looncraz.net/

I also made a few Zen performance estimates. The amount of math involved is glorious, but all my data is disclosed and linked on the site - even though I post all my images in relative terms.

I Think you are wearing rose colored glasses if you believe ANY AMD CPU does "rather well" in Super pi or other similar applications. Those in the know wouldn't touch an amd for pi.

 

[looncraz]

Makes no sense really. Clearly Bulldozer's main caveat is poor low thread performance. Multi thread wise it does much better.

Bulldozer's multi-threaded strength is purely a function of its core count, nothing more. Its scaling is HORRENDOUS due to the module penalties.

Zen has all of the following benefits over Bulldozer for multi-threaded:

1. No module penalties (+15% thread scaling, right off the bat)
2. Fast L3 (unknown adder, but an adder nonetheless)
3. Fully dedicated FPUs
4. SMT (+15~20%)

Even if Zen had the same IPC it would demolish Bulldozer in multi-threaded workloads.

In a scientific sense, someone who pays (as you state in your opening page) to demonstrate a principle or process will be blindsided by their own bias.
Using maths to justify end conclusions is invalid when you don't know enough about Zen to compile accurate data.
Your work is commendable but the enthusiasm behind it also compromises it's neutrality and validity.

Only time will tell how well Zen does. And I do believe only the mad want it to fail. Intel's pricing structure needs a shake up (much like Polaris will hopefully do to Nvidia mid range).

There is less conflict in buying the product from which I created the baseline data than if the product had been given to me to create the data to sale that product. My numbers are also entirely accurate as I benchmarked them several times and validated them with third parties to ensure I was within the same ballpark. I performed exactly zero manipulations to the results. The full, raw, results are linked in the article.

BTW, I know a great deal about Zen. I'm probably one of the most knowledgeable people on the subject outside of AMD or GloFo. I broke down the GCC patch into pipelines: http://looncraz.net/ZenAssignments.html

As for objectivity, that is mostly guaranteed by the use of methods deployed. Scientists are often very enthusiastic about what they are doing - the only way they can be objective is by following standards and procedures - which I did religiously.

I Think you are wearing rose colored glasses if you believe ANY AMD CPU does "rather well" in Super pi or other similar applications. Those in the know wouldn't touch an amd for pi.

Excavator does rather well... compared to Steamroller They saw a nearly 18% improvement for that one, genuinely worthless, benchmark.

 

[EarthDog]

See the forest outside of the trees.


You may want to edit your posts instead of triple posting.

 

[looncraz]

You may want to edit your posts instead of triple posting.

Sorry, didn't realize I did that

We can still be friends, yes?

 

[Dent1]

Apparently you've never played on a P4 Northwood, and they came out well before dual core was popular enough to require one for most games. Hell, I even played original Crysis on my Northwood, and with max textures. I had a 4:3 CRT display that I set to 1200x900 for that game, with Med Shaders and Shadows, and only had to employ a few simple file edits to tweak Object Quality and view distance.

Keep in mind dual core was the first step of multi core CPU gaming, so it took a while for devs to adopt. I think the problem here though, is you're not putting each chip in it's own timeline. P4 Northwood fared WAY better in it's time, than Bulldozed is in it's.

You mean the P4 Northwood that was getting handily beat by the lower clocked budget AMD Duron and AMD Sempron?

 

[Frag_Maniac]

You mean the P4 Northwood that was getting handily beat by the lower clocked budget AMD Duron and AMD Sempron?

LOL, where do you guys get this malarkey? Before dual cores were necessary, the P4 Northwood was a quite popular chip for gaming, and it OCed well, and didn't run as hot as a toaster like a lot of AMD product.

 

[Frag_Maniac]

Even if Zen had the same IPC it would demolish Bulldozer in multi-threaded workloads.

It will demolish it regardless, but only because Bulldozer is a crap design. Where it will smash it most is where Bulldozer is weakest though, and that's as everyone knows, in low thread applications.

I'm talking real world performance, not IPC per core, because there's always going to be overhead in octo core CPUs until games catch up. So the biggest noticeable difference will be in low thread applications. You don't seem to grasp that.

 

[Mr.Scott]

You mean the P4 Northwood that was getting handily beat by the lower clocked budget AMD Duron and AMD Sempron?

QFT

 

[looncraz]

LOL, where do you guys get this malarkey? Before dual cores were necessary, the P4 Northwood was a quite popular chip for gaming, and it OCed well, and didn't run as hot as a toaster like a lot of AMD product.

AMD was still had the IPC lead after the Northwood die shrink came to market. The only thing that really changed was that P4 Northwood used less power.

There were many times when Northwood was beat out by an Athlon XP clocked 600Mhz or more lower - especially in multimedia.

Take a look here: http://techreport.com/review/3289/pentium-4-northwood-2-2ghz-vs-athlon-xp-2000

It will demolish it regardless, but only because Bulldozer is a crap design. Where it will smash it most is where Bulldozer is weakest though, and that's as everyone knows, in low thread applications.

I'm talking real world performance, not IPC per core, because there's always going to be overhead in octo core CPUs until games catch up. So the biggest noticeable difference will be in low thread applications. You don't seem to grasp that.

IPC creates real world performance in lieu of other bottlenecks. A 40% IPC improvement over Excavator is nearly an 80% improvement over Bulldozer (albeit not evenly distributed). Multi-threaded workloads aren't uniform, which is what I think you're trying to say (albeit, poorly).

Games don't usually thread well, but many applications scale very well with more cores. However, reviewers focus on applications that tend to scale well - Cinebench, Handbrake, etc... - and, in those applications, Zen should nearly double Bulldozer's performance.

 

[Frag_Maniac]

If anyone has credible benches from a legit site that compares a same time frame, same price point AMD single core CPU significantly outperforming a P4 Northwood, I think it would have been posted by now.

Here is the exact conclusion of looncraz's link.

"Conclusions
Performance-wise, it's a toss-up. I would like to declare one or the other of these processors the clear winner, but that's just not possible. The Athlon XP 2000+ and Pentium 4 2.2GHz are locked in a dead heat for the title of "fastest x86 processor."

So the best you can say is it tied for fastest. And mind you, that was AMD's answer to the Northwood that had just released, so prior to that, the Northwood was king. I mean how many times have we seen this, where AMD can only match vs dethrone an existing dominate chip? Lately they can't even match.

 

[looncraz]

If anyone has credible benches from a legit site that compares a same time frame, same price point AMD single core CPU significantly outperforming a P4 Northwood, I think it would have been posted by now.

Here is the exact conclusion of looncraz's link.

"Conclusions
Performance-wise, it's a toss-up. I would like to declare one or the other of these processors the clear winner, but that's just not possible. The Athlon XP 2000+ and Pentium 4 2.2GHz are locked in a dead heat for the title of "fastest x86 processor."

So the best you can say is it tied for fastest. And mind you, that was AMD's answer to the Northwood that had just released, so prior to that, the Northwood was king. I mean how many times have we seen this, where AMD can only match vs dethrone an existing dominate chip? Lately they can't even match.

They are also comparing an AMD CPU clocked at 1.6GHz to an Intel CPU clocked at 2.2GHz.

Also, I augmented my prior post to respond to another of your posts.

 

[Dent1]

LOL, where do you guys get this malarkey? Before dual cores were necessary, the P4 Northwood was a quite popular chip for gaming, and it OCed well, and didn't run as hot as a toaster like a lot of AMD product.

Malarkey? It's truth. Durons and Semprons outperformed P4 Northwood at lower cost.

P4 Northwood were popular because Intel was more mainstream had a stranglehold. They used to force retailers to push their inferior products and punish manufacturers for using AMD parts.

Anyways I'm talking raw performance here, not heat. Mr Subject Changer.

 

[Frag_Maniac]

Malarkey? It's truth. Durons and Semprons outperformed P4 Northwood at lower cost.

Again, if you're talking same time frame, same price point single cores that are significantly faster at gaming as has been implied, post up benches from legit sites, and I'll eat my hat, but so far you're just tootin your horn.

All I've seen thus far is a link to a bench that admits AMD's answer was at best equal to it.