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

[cytg111]

And thus gutting their own APU business? Again Trump won, anything is possible!!

 

[sirmo]

I don't see any official confirmation here.

semantics.. I mean an official outlet confirming the existence of a rumor.

 

[Dresdenboy]

And thus gutting their own APU business? Again Trump won, anything is possible!!

How much do Intel's iGPUs, their EUs, caches, drivers, OpenCL stack resemble Nvidia's GPUs?

 

[ozzy702]

So... about those Zen benchmarks...

 

[sirmo]

So... about those Zen benchmarks...

We might know more..

 

[Phynaz]

I don't see any official confirmation here.

Some people get reinforcement from having the same rumor repeated. It validates their wishes for it to be true.

 

[KTE]

"

I can keep listing software all day but i need to go to work most things that revolve around users are licensed per user, most things that support that are licensed per proc.
Its only really when you get to things like Oracle, MS SQL etc it becomes per Core.
"

As person who spend good amount of time negotiating with such vendors specifically and 'conversions' from Processor to Core and diving very deep not only into enterprise contracts but also all of the licensing collateral and various programs vendors published over time -I see almost nothing licensed per proc anymore. i.e. Microsoft stopped mentioning it anywhere in the documentation post year 2012, and now has to deal with enterprises that own the licenses bought outright, still valid under software assurance

major software companies are not stupid - selling something per 'processor' and then seeing processor going from 4 cores to 20 plus is a major loss of revenue for them. Also add to it virtualization as a layer , what defines a 'core' (real? HT? ) and how it all mixes together -> management consultants have their field date and earn their fees in helping large corporations work through this.

Regarding 'per user', it is also very treacherous - who exactly is the user? how are consultants handed? how are remote services (i.e. Infosys of the word) with off shore in India handled if we give them access to our systems (which is required for them to be of any use). whenever possible we try to avoid per user licenses and if not possible, severely control it through both procurement and deployment phases.

of cause all of this have nothing to do with CPUs, so sorry for off topic.

I work approving all this daily, and my experience agrees with yours. I can't say much except I'm currently doing this on a company-wide scale.

Licensing is a FULL field in itself.

Enterprise -> all equipment is lease now.

Infra... VM licensing is ENTIRELY core based. Legacy archs cost HUGE to keep running because of this.

Software -> most enterprise ones now per core. Even JBoss, IBM, Microsoft, Oracle Enterprise Licensing... Ones that are not, are becoming so. From the older per socket model.

As for the big hosting datacentres... Google, Microsoft, Amazon, Oracle, IBM, Ark, Salesforce, Adobe, SAP, NTT (soon to include HP, Dell, Facebook)... Now how they operate is very different to the archaic in-house models.

Zen needs to be pitched to the above firms for a successful chance in grabbing server marketshare going forward.

I agree that typical workloads demand more mem than core processing however, hence why consolidation is so big.

Sent from HTC 10
(Opinions are own)

 

[sirmo]

Some people get reinforcement from having the same rumor repeated. It validates their wishes for it to be true.

Was this directed at me? Because that was very far removed from the spirit of my post. I distinctly seem to remember writing "take it with a grain of salt" and "Kyle has been wrong in the past" in that very post. So get out with that BS.

Only people wishing for something here.. are those who don't want to see AMD succeed. For whatever reason that will forever escape me. Since this is supposed to be an enthusiast forum where competition is a good thing for all of us.

 

[Arachnotronic]

Huh? According to Yahoo Finance it's currently at 5.3%. At $20, it would reach 11.3%. BTW I think, $20 is rather high for 2017. Current prices are at levels based on future expectations, thus including next year effects with a risk discount.

Right, I was saying that if AMD went to $20 per share, it would not be worth 60% of Intel's market value. We are in solid agreement here

 

[KTE]

Was this directed at me? Because that was very far removed from the spirit of my post. I distinctly seem to remember writing "take it with a grain of salt" and "Kyle has been wrong in the past" in that very post. So get out with that BS.

Only people wishing for something here.. are those who don't want to see AMD succeed. For whatever reason that will forever escape me. Since this is supposed to be an enthusiast forum where competition is a good thing for all of us.

Very few enthusiasts want to see AMD fail. I am about the only guy in this company even naming AMD. The respective board look at me and start chuckling

Let's talk server CPUs with the traditional infrastructure models. Buying firms want competition? Nah, they don't care. They get major non-contractual leeway and discounts. They just want a good client relationship, primarily.

Shifting to something different requires MAJOR case presenting to a few levels of seniors, where one is bound to bar approval. Even if he just can't be asked with the hassle. Shifting is typically done where the cost savings are substantial vs the effort, where major contract breaches have occurred or audits/compliance failed. Why do you think old, expensive dinosaurs like HPUX are still being leased!?! Sometimes for just hosting one small legacy db with a green screen holding historical data. No kidding!

Remember, any change becomes a project, and any project brings a team... And they have to bring specialist contractors on to see through these changes. Corporate change is never cheap.

BTW, Kyle isn't a source that I could unfortunately turn to though.

Sent from HTC 10
(Opinions are own)

 

[Ajay]

Bingo KTE! AMD will need to provided allot of resources to help any vendor that is willing to hock their server platform. They will need to provide significant resources to any major firm to defer costs, IMHO.

Very few enthusiasts want to see AMD fail. I am about the only guy in this company even naming AMD. The respective board look at me and start chuckling

Let's talk server CPUs with the traditional infrastructure models. Buying firms want competition? Nah, they don't care. They get major non-contractual leeway and discounts. They just want a good client relationship, primarily.

Shifting to something different requires MAJOR case presenting to a few levels of seniors, where one is bound to bar approval. Even if he just can't be asked with the hassle. Shifting is typically done where the cost savings are substantial vs the effort, where major contract breaches have occurred or audits/compliance failed. Why do you think old, expensive dinosaurs like HPUX are still being leased!?! Sometimes for just hosting one small legacy db with a green screen holding historical data. No kidding!

Remember, any change becomes a project, and any project brings a team... And they have to bring specialist contractors on to see through these changes. Corporate change is never cheap.
(Opinions are own)

 

[Majcric]

We might know more..

Ahhh yes let the countdown begin. This feels like an early Christmas present. Fingers crossed,

 

[StrangerGuy]

Very few enthusiasts want to see AMD fail. I am about the only guy in this company even naming AMD. The respective board look at me and start chuckling

Let's talk server CPUs with the traditional infrastructure models. Buying firms want competition? Nah, they don't care. They get major non-contractual leeway and discounts. They just want a good client relationship, primarily.

Shifting to something different requires MAJOR case presenting to a few levels of seniors, where one is bound to bar approval. Even if he just can't be asked with the hassle. Shifting is typically done where the cost savings are substantial vs the effort, where major contract breaches have occurred or audits/compliance failed. Why do you think old, expensive dinosaurs like HPUX are still being leased!?! Sometimes for just hosting one small legacy db with a green screen holding historical data. No kidding!

Remember, any change becomes a project, and any project brings a team... And they have to bring specialist contractors on to see through these changes. Corporate change is never cheap.

BTW, Kyle isn't a source that I could unfortunately turn to though.

Sent from HTC 10
(Opinions are own)

Yup, it's amazing how some kiddo thinks just because they can buy a $200-300 consumer CPU suddenly makes them experts for enterprise because the only thing that matters is performance/price amirite and the server people are all idiots for buying Intel.

 

[Ajay]

Ahhh yes let the countdown begin. This feels like an early Christmas present. Fingers crossed,

Wait, what is this countdown for?

 

[.vodka]

It's a countdown until the event in the 13th.

 

[sirmo]

Let's talk server CPUs with the traditional infrastructure models. Buying firms want competition? Nah, they don't care. They get major non-contractual leeway and discounts. They just want a good client relationship, primarily.

Shifting to something different requires MAJOR case presenting to a few levels of seniors, where one is bound to bar approval. Even if he just can't be asked with the hassle. Shifting is typically done where the cost savings are substantial vs the effort, where major contract breaches have occurred or audits/compliance failed. Why do you think old, expensive dinosaurs like HPUX are still being leased!?! Sometimes for just hosting one small legacy db with a green screen holding historical data. No kidding!

Remember, any change becomes a project, and any project brings a team... And they have to bring specialist contractors on to see through these changes. Corporate change is never cheap.

This is a huge market. Not every case is the same. These are $100m+ contracts being negotiated. People have been yearning for a second source if for anything but as a bargaining chip. Shifting from Intel to AMD is a joke compared to shifting from Intel to Power8/9, and IBM has been recording profits in their hardware division and expects a growth to double digit marketshare by 2020. Google uses Power8 for instance.

In the context of Zen.. AMD doesn't need much of the market to declare Zen a success in servers, 5-10% would be huge (they are less than 1% currently), and it's certainly achievable provided they have a good product. Or do we forget that at one point they had over 20% of the market share? I personally worked with racks of Sun Opteron servers back then, and they were great, no code changes needed, stuff just ran better and consumed less power than it did on competing Intel servers. We ran mission critical telco software on these.. with FCC fines if stuff broke, no issue.

If Zen can match perf/watt of Intel in datacenter it will sell. Period. Most businesses will continue to buy Intel as they always do, but that's not what I am talking about.

 

[Arachnotronic]

So they gained a bunch of share when Intel was fielding NetBurst based junk, and then their share started coming down beginning with the Merom-based chips. The next leg down happened around the Nehalem-EP/EX launches and continued through Westmere.

Bulldozer did not launch for servers until Q3 2011, FYI.

The bulk of the damage to AMD's server business happened well before Bulldozer. In fact, on AMD's Q3 2011 earnings call, Rory Read said this:

Now for our server business. The third quarter was the beginning of a move in the right direction, with server revenue up 27% sequentially. Initial production shipments of our new AMD Opteron products gained traction in the high-performance computing space, where with the help of our key partner, Cray, and some of our most notable customers which included the National SuperComputer Centers in Stuttgart, Germany, in the U.K. and in Switzerland, as well as the United States Department of Energy Titan project at the Oak Ridge National Laboratory. This project, in fact, is likely to be one of the world's fastest supercomputers. We are excited about our key partners, including HP and Dell, who are launching new products based on our new Opteron platform this fourth quarter, and we are confident they will do well against competition, particularly in key workload areas like the cloud and virtualization.

History didn't quite play out the way that a lot of enthusiasts remember it.

 

[sirmo]

So they gained a bunch of share when Intel was fielding NetBurst based junk, and then their share started coming down beginning with the Merom-based chips. The next leg down happened around the Nehalem-EP/EX launches and continued through Westmere.

Bulldozer did not launch for servers until Q3 2011, FYI.

The bulk of the damage to AMD's server business happened well before Bulldozer. In fact, on AMD's Q3 2011 earnings call, Rory Read said this:



History didn't quite play out the way that a lot of enthusiasts remember it.

I am not following your point. Before Bulldozer AMD wasn't competitive. Barcelona was a flop and AMD was already struggling with an ageing architecture. 2011 was the year of the Sandy Bridge. Of course Sandy Bridge was going to sweep the market it was faster and more efficient than AMD. Bulldozer was supposed to be the saviour of the server market share loss. Hence the CMT design??

Let's also not forget that AMD achieved close to 25% of the server market share despite the Intel's illegal "subsidies" to big OEMs.

 

[Arachnotronic]

I am not following your point. Before Bulldozer AMD wasn't competitive.

That's not what AMD was telling its investors back then:

Barcelona was a flop and AMD was already struggling with an ageing architecture. 2011 was the year of the Sandy Bridge. Of course Sandy Bridge was going to sweep the market it was faster and more efficient than AMD. Bulldozer was supposed to be the saviour of the server market share loss. Hence the CMT design??

Sandy Bridge-EP did not arrive until 2012. Study the quote from Rory Read carefully, he was very bullish on Bulldozer Opteron being competitive with Intel.

Let's also not forget that AMD achieved close to 25% of the server market share despite the Intel's illegal "subsidies" to big OEMs.

The subsidies to Dell were for their PC products, AFAIK. In the data center, no subsidies are going to make up for bad, high TCO chips.

 

[sirmo]

Question, why am I constantly having to correct misinformation and defend AMD on here?
I don't know, it just seems odd that for people who supposedly want to see Zen compete, there seems to be so much negative misinformation deriding AMD's achievements and accomplishments happening here. Am I the only one who sees this?

That's not what AMD was telling its investors back then:

Those numbers are MCM chips.. we all know that C2D, Nehalem.. onwards Intel had a single thread performance lead. Magny-cours did well but perf/watt wasn't there.. this is all common knowledge. Those chips had some redeeming qualities.. lots of cores and they were cheaper, and that's what those slides show. To be clear Phenom II (once Barcelona bug was fixed) era chips were not terrible, but they were aging and behind Intel in most metrics by that point, which is certainly what the market share graph reflects. And AMD still had some marketshare in the server at that time, not much but certainly more than what they have now.

The graph I linked by the way shows that historically AMD had a double digit server market share, which disproves some statements made earlier in this thread. how do we get here from that point? What is the point you're making? That CEOs make optimistic forward looking statements? Who doesn't know that? It's listed in the risks section of every earnings report.

The subsidies to Dell were for their PC products, AFAIK. In the data center, no subsidies are going to make up for bad, high TCO chips.

First of all Dell is just one company.. Intel blocked AMD sales on multiple continents, US, EU, Japan... etc. Why would they block desktop sales and not block the more profitable (higher margin) server parts? But don't take my word for it:

http://www.forbes.com/2005/06/28/amd-intel-lawsuit-cx_ah_0628amd.html

That's just one of many examples.

 

[Arachnotronic]

Those numbers are MCM chips.. we all know that C2D, Nehalem.. onwards Intel had a single thread performance lead. Magny-cours did well but perf/watt wasn't there.. this is all common knowledge.

Naples is an MCM, made up of 4x 8 core Zeppelin dies. These will be competing with monolithic Intel dies with lots of cores, as well as monolithic dies with lots of cores from the ARMy.

Those chips had some redeeming qualities.. lots of cores and they were cheaper, and that's what those slides show.

From looking at this slide, AMD was trying to tell the world that it had competitive offerings in servers. According to you, AMD did not have competitive offerings.

To be clear Phenom II (once Barcelona bug was fixed) era chips were not terrible, but they were aging and behind Intel in most metrics by that point, which is certainly what the market share graph reflects. And AMD still had some marketshare in the server at that time, not much but certainly more than what they have now.

Phenom II was a nice chip family, but the real gems were those Athlon II X3/X4 chips derived from the same cores. Great value for the money, I built probably a dozen systems with them.

 

[sirmo]

From looking at this slide, AMD was trying to tell the world that it had competitive offerings in servers. According to you, AMD did not have competitive offerings.

As I mentioned in the previous post, they had redeeming qualities. Price per core (or multi threaded performance per SKU which ever way you look at it) most notably. But it was already an aging architecture at that point. Single thread performance and perf/watt was already behind Intel. Which in the datacenter, especially the latter one is king.

 

[swilli89]

This is a huge market. Not every case is the same. These are $100m+ contracts being negotiated. People have been yearning for a second source if for anything but as a bargaining chip. Shifting from Intel to AMD is a joke compared to shifting from Intel to Power8/9, and IBM has been recording profits in their hardware division and expects a growth to double digit marketshare by 2020. Google uses Power8 for instance.

In the context of Zen.. AMD doesn't need much of the market to declare Zen a success in servers, 5-10% would be huge (they are less than 1% currently), and it's certainly achievable provided they have a good product. Or do we forget that at one point they had over 20% of the market share? If Zen can match perf/watt of Intel in datacenter it will sell. Period.

Great post. The points you make are very solid and this is exactly what AMD is aiming to do. As long as they can get similar perf/watt (and while the 14nm process doesn't clock high, it's been proven to work wonderfully at lower frequencies with much more attractive power curves) and offer some cost savings, they will succeed in their near term mission.

 

[kraatus77]

snip

you know the rules pretty well, why you are continuing derail ?

this negativity is totally unnecessary. nobody asking for being positive but if you can't than be neutral at least.

 

[AtenRa]

Even if they will manage to get a 10% of the CPU Server market (~15Billions) in 2018, it could generate half a Billion of profits per year.
And thats not counting any profits from Desktop/Laptop market (~35Billions) where AMD could reach 20% and generate 1B of profits.

That is viable for 2018, but first lets see if ZEN will compete as we are expecting. That is with high Perf/Watt no matter if IPC is higher than Sandy or Hasswell or whatever.