AMD Ryzen (Summit Ridge) Benchmarks Thread (use new thread)

[Dresdenboy]

What about the 40% improvement "on top of all the micro architectural improvements"?

I think he referred to remaining power efficiency improvements, because a part of the process' improvement likely already got used up in uarch and cycle time.

 

[Azuma Hazuki]

At this point, half the reason I want Zen to do well is to make the haters shut their useless mouths for just a few seconds and show some respect. The passive-aggressive "nyah nyah, I'm not actually trolling enough for the mods to spank me!" attitude from some of them is just bewildering.


This type of comment is not allowed in the tech forms
Markfw900

 

[StrangerGuy]

At this point, half the reason I want Zen to do well is to make the haters shut their useless mouths for just a few seconds and show some respect. The passive-aggressive "nyah nyah, I'm not actually trolling enough for the mods to spank me!" attitude from some of them is just bewildering.

Hurrrrr, "respect". You want respect? Show us the actual goods or just take your useless whining about whining elsewhere.

 

[Azuma Hazuki]

Is it so much to ask that people wait for actual, repeatable benchmarks before they spout off? Or, for that matter, to know the history of the brands in question here?

I realize this will probably be wasted on you, but let's try a little thought experiment here: suppose AMD never developed x86_64 ("amd64") or the K8/*hammer uarches. What do you think the CPU landscape would look like now?

 

[StrangerGuy]

Is it so much to ask that people wait for actual, repeatable benchmarks before they spout off? Or, for that matter, to know the history of the brands in question here?

I realize this will probably be wasted on you, but let's try a little thought experiment here: suppose AMD never developed x86_64 ("amd64") or the K8/*hammer uarches. What do you think the CPU landscape would look like now?

Supposed Intel never made x86, what do you think AMD would look like now?

Oh, of course we do know the history of the brands here, especially regarding the track record for the past 10 years.

 

[Azuma Hazuki]

They wouldn't, since the point of AMD's early existence was to second-source x86-uarch CPUs for a company you may know called IBM...remember them?

Answer the question; I'd love to see what shakes out of that

 

[StrangerGuy]

They wouldn't, since the point of AMD's early existence was to second-source x86-uarch CPUs for a company you may know called IBM...remember them?

Answer the question; I'd love to see what shakes out of that

Take your inane original sin crap elsewhere. Ironic threadcrapper is ironic.

 

[Azuma Hazuki]

What the hell are you talking about, "original sin?" Stay on topic.


Profanity is not allowed in the tech forums
Markfw900

 

[deasd]

WCCFTech, so I vote fake. That would be a pretty good result assuming the turbo is working though.

That result is not posted by WCCF but from a comment below that article, which was removed immediately.
I tend to believe it's real data that ever exist in database but got deleted, if it's fake then he have to fake that screenshot with photoshop.

 

[inf64]

That result is not posted by WCCF but from a comment below that article, which was removed immediately.
I tend to believe it's real data that ever exist in database but got deleted, if it's fake then he have to fake that screenshot with photoshop.

I also think the data is real but I guess some will have to see the real Summit Ridge benchmarked before they believe it.

 

[bjt2]

I think he referred to remaining power efficiency improvements, because a part of the process' improvement likely already got used up in uarch and cycle time.

Are you implying that the cycle time of Zen may be lower than BD's and thus clock may be higher?

 

[bjt2]

Not all ALU ports are shared with the FPU. Please, take a look at Haswell schema for this.

Haswell is worse, because the 4 int ports has attached 2 256 bit FPU plus 3 vector ALU, so it's 4 INT/FP/Vec with maximum 2 FP and 3 vec int.
For 128 bit code without FMAC it's half than Zen throughput... I was talking of skylake, that hopefully it's a better true 4 shared ports (i can't find a diagram on google)

Sure, it's an advantage. But here we are talking about the execution ports, and not the unique scheduler. To be more clear, a unique scheduler isn't a bottleneck by definition: it's a shared port which MIGHT be.

But we see that AMD for two threads can do 4 INT PLUS 4 FP, while Haswell 4 INT or 3 INT and 1 FP etc, up to 2 FP or 3 vec... This is what i meant... Obviously unified queue (from queue theory) is better, but not quite with half queues...

You've forgot that an Intel core has different kinds of uops, and that its decoders can decode up to 4 instructions to 6uops, and every uop can be split in 2 simpler uops before being sent to an execution port. The 4 decoders can also merge up to two different instructions in one uop, but with at most 2 merges per cycle.

So, Intel's Micro-op cache can deliver up to 6 uops/cycle, but every uop can carry:
- 1 uop;
- 1 uop -> 2 merged instructions;
- 1 uop -> 2 (simpler) uops.

And this might be one of the reasons why Intel's microarchitectures have good performances even with 8 ports with partially shared ALUs/FPUs.

AMD MOPS and uops are very powerful too... And there is also uop fusion in AMD architectures...

 

[ShintaiDK]

I also think the data is real but I guess some will have to see the real Summit Ridge benchmarked before they believe it.

It does seem so. It took the release of BD to show people otherwise. Steamroller didn't bring 20% IPC and so on

 

[cytg111]

If you dont require evidence for your facts then your beliefs are based on faith.
I for one take my facts with science, white coats and reproducible experiments.
Of course I never were the religious type which may be why I am such a poor fit in the cpu and overclocking section here. Just kidding. .

 

[jpiniero]

That result is not posted by WCCF but from a comment below that article, which was removed immediately.
I tend to believe it's real data that ever exist in database but got deleted, if it's fake then he have to fake that screenshot with photoshop.

Which would be easy to do. Without the individual scores it'd be hard to determine the legitimacy.

 

[Arachnotronic]

Is it so much to ask that people wait for actual, repeatable benchmarks before they spout off? Or, for that matter, to know the history of the brands in question here?

The history of the brands in question here is not favorable to AMD. With K7 AMD was competitive, though Intel was able to extract enough frequency from the second iteration of the P4 (Northwood) to make them viable choices in the marketplace. With K8, AMD opened up a lead on Intel thanks to some core improvements as well as an integrated memory controller.

With Conroe, Intel took a decisive leadership position, though it still hadn't integrated the memory controller -- which required Intel to bolt on boat-loads of L2$ to hide the inherent memory latency of the FSB.

People then expected AMD's upcoming Barcelona CPU (the "Zen" of its day) to come in and clean up, putting AMD back in a leadership position. AMD execs were telling people that Barcelona would outperform AMD's "Clovertown" by 40%, which helped to build hype for AMD's upcoming processors (at the time, Intel was regaining significant share in PCs and servers, which hurt AMD financially -- they had to significantly lower CPU prices to stay in the game).

Barcelona was disappointing, due to low clocks and a ST perf/MHz deficiency relative to Conroe/Kentsfield. AMD tried to market its parts as "true quad cores" (Intel was using an MCM), but the performance numbers spoke for themselves.

AMD did a Barcelona v2 in the form of the Phenom II line. These cleaned up the original Phenom's cache architecture and could clock significantly better. They could not match the Nehalem parts in performance, and Intel's product stack was aggressive enough to keep AMD at bay, but I do remember buying a LOT of Athlon II X3/X4 chips back in the day for builds I did for people because AMD was able to price aggressively enough to stay relevant.

Then we got Bulldozer, what a piece of work that was. It was very hyped up, viewed as the "second coming" by many forum goers across the web. Basically, the "common wisdom" was that AMD would be able to offer up many more cores than Intel and still offer competitive ST performance, which would allow AMD to deliver a K.O. punch in PCs and servers (especially servers actually -- Anand himself said that he thought Bulldozer would do well in the server market).

The reality was that Bulldozer -- in an AMD's execs own words -- was an "unmitigated disaster." A quad core Sandy Bridge (I remember people were told to "wait for Bulldozer" rather than buy SNB, and I believe I actually waited to see what AMD had coming up back in the day) wrecked the FX-8150 in most consumer workloads, and even in heavily threaded workloads that 2600K could pack a punch.

At the end of the day, most enthusiasts seem to have gone with the 2500K/2600K. I loved my 2600K.

AMD only iterated on Bulldozer in the server/enthusiast desktop space once with Piledriver, and since then pretty much left the enthusiast/server market probably because they needed to divert much of their limited R&D resources (which began to fall rapidly as revenue dried up) to Zen. AMD continued to build laptop SoCs with the newer Construction cores though because they couldn't let their PC revenue go to zero and consumers seem to be more forgiving of BD & derivatives shortcomings (data center customers have to worry about the TCO of a given solution, so they can't buy inefficient products because they want to support AMD; consumers who buy laptops at Best Buy could very well buy an AMD powered device and be happy with it, even if the Intel alternative performs better).

In that time, Intel put out Ivy Bridge (+5% IPC over SNB), Haswell (+10% IPC over IVB), Broadwell (+3-5% IPC over HSW), and Skylake (+12% or so over BDW). Aside from the Broadwell issue, Intel managed to keep maximum OC potential roughly constant to around 4.7-4.8GHz even though IPC went up. Intel also integrated more stuff into these chips, dramatically improved iGPU, etc. Kaby Lake in early 2017 won't move IPC forward but it will probably allow for 5GHz-ish overclocks, so performance moves up.

As a consumer, I view one company as having delivered new, worthwhile products for me to buy approximately every year or so. The other abandoned the enthusiast desktop market for a while, but says it's going to be back. We'll see how good Zen ultimately is, but if you look at the 10 year track records of both companies (and the financial reality that Intel has dramatically increased its R&D spending over the last decade while AMD hasn't, and AMD spends about 1/12th of what Intel does), it is not favorable to AMD.

 

[coercitiv]

The history of the brands in question here is not favorable to AMD.

We'll see how good Zen ultimately is, but if you look at the 10 year track records of both companies (and the financial reality that Intel has dramatically increased its R&D spending over the last decade while AMD hasn't, and AMD spends about 1/12th of what Intel does), it is not favorable to AMD.

On one side Intel delivered with their high performance CPU line, on the other side the same giant of a company, both talent and budget wise, was brought down to it's knees by the ARM gnome and friends. I keep seeing the history and budget arguments over and over in the high performance area, but there's an awkward silence when the same logic cannot be used in the low power market: Intel funneled billions and billions of dollars in there and eventually came back home with little to show for.

Even you have abandoned Intel's banner for the more favorable strong winds (pun intended) of Apple's new ARM based desktop class CPUs, meaning you chose to ignore history: Intel showed us before that when they get challenged in CPU performance they will stop at literally nothing to neuter their competition. Based on history trends Apple CPU business will get blindsided and left for dead in a ditch somewhere.

Using past behavior as best predictor for future behavior works only if actors have not been made aware of the consequences of their actions. Ponder on that.

 

[coercitiv]

meh, double post

 

[Arachnotronic]

On one side Intel delivered with their high performance CPU line, on the other side the same giant of a company, both talent and budget wise, was brought down to it's knees by the ARM gnome and friends.

Sure, Intel's Atom efforts didn't work out in smartphones and tablets for a number of reasons.

I keep seeing the history and budget arguments over and over in the high performance area, but there's an awkward silence when the same logic cannot be used in the low power market: Intel funneled billions and billions of dollars in there and eventually came back home with little to show for.

I don't know who said this first, but I have seen it used by ShintaiDK and others: In R&D, you don't always get what you pay for, but you won't get what you don't pay for.

Intel's mobile failure is an example of the latter. As for the "ARM gnome and friends," you do realize that the ARM companies that have been successful (Apple, Samsung, Qualcomm) are not operating on AMD-level R&D budgets.

Even you have abandoned Intel's banner for the more favorable strong winds (pun intended) of Apple's new ARM based desktop class CPUs, meaning you chose to ignore history: Intel showed us before that when they get challenged in CPU performance they will stop at literally nothing to neuter their competition. Based on history trends Apple CPU business will get blindsided and left for dead in a ditch somewhere.

"Neuter their competition"? Seems to me that when Intel was challenged back in the P4 days, it ramped up its investments in the Core product line and came back roaring with an excellent architecture and methodically built off of that success in the following years.

AMD, on the other hand, was forced to dramatically slash R&D spending over the last five years, as a declining PC market (and substantial share loss, particularly in budget systems at the hands of Silvermont and its derivatives), the evaporation of its server share, and other forces took their toll.

Were it not for the game console wins, AMD's revenue today would be at about $2 billion annually at, perhaps, 40% gross profit margins. Given that their operating expenses are now at ~$1.4 billion/year, AMD would have already gone bankrupt under those conditions.

I have no love or hate for AMD -- they are just another tech company in a sea of tech companies. What I find absolutely puzzling is this view that AMD is this holy savior, the only thing standing between us the "noble" consumer and the "evil" Intel trying to vacuum the money from our pockets.

To be quite blunt, I can't understand why people think we'd be paying $500 for Celeron chips if AMD weren't around. AMD hasn't fielded a competitive enthusiast part since...well, maybe 2006?

The Core i7 6700K -- arguably the best gaming/enthusiast CPU out there -- sells for just $339 -- or about $283.75 in 2006 dollars.Core 2 Duo E6700 sold for $530 back in 2006.

Anyway, the point is that Intel can't go crazy with prices or intentionally hold back its products even without AMD around because if they do, people won't be interested in buying their stuff and will simply choose not to buy new technology. That would kill Intel far more quickly than even the best Zen processor that people are imagining.

 

[Hossain84]

slower than Skylake

 

[DrMrLordX]

It does seem so. It took the release of BD to show people otherwise. Steamroller didn't bring 20% IPC and so on

The Stilt already posted that slide in this thread, did he not? 20% was conditional on the benchmark. The stated average was 10%. Please stop with the misinformation.

 

[ShintaiDK]

The Stilt already posted that slide in this thread, did he not? 20% was conditional on the benchmark. The stated average was 10%. Please stop with the misinformation.

Did he? I must have missed it. The conditional was up to if there was any. Just as with Zen. And its far from the first time. Intel could just as well claim up to 80% IPC increase with Haswell for example. But using a fairy tale for IPC increase is misleading. AMD haven't told the truth about performance since when, K8?

 

[lolfail9001]

Did he? I must have missed it. The conditional was up to if there was any. Just as with Zen.

Yeah, but he posted a slide that mentioned average 10% IPC improvement. Something observable from what i know.

 

[Dresdenboy]

Did he? I must have missed it. The conditional was up to if there was any. Just as with Zen. And its far from the first time. Intel could just as well claim up to 80% IPC increase with Haswell for example. But using a fairy tale for IPC increase is misleading. AMD haven't told the truth about performance since when, K8?

Is Powerpoint "smart art" an exact reference? And even your "20%" number includes the "up to" statement. So I even don't need to ask for the footnote. The Stilt posted this:

Source: https://forums.anandtech.com/threads/first-summit-ridge-zen-benchmarks.2482739/page-47#post-38499905

 

[ShintaiDK]

So in other words, the up to 40% with Zen for example can mean anything. Even 10%

Thanks for confirming it.