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

[piesquared]

http://blogs.barrons.com/techtrader...uraged-by-renduchintalas-pragmatic-influence/

Intel president, is expecting Zen to disrupt the market until 10nm parts.

This is not surprising. AMD being able to address 80% of the TAM with Opteron/(or perhaps a new brand) and compete head on in desktop as well should give intel cause for worry. I hope for their sake they dont believe the people that claim nobody will buy AMD products given a similar choice between intel and AMD. Those people that say these things are the loud minority, maybe even paid to swing opinion (not unheard of, look at the mainstream media for example, or astroturfing).

Assuming Vega is high end as well and given HBM2 it would appear that is the case, then AMD suddenly has a very compelling platform. Vega Pro would probably be quite well recieved also. There's also Radeon Pro SSG that makes a nice addition to the AM4 platform. Its a mystery how AMD has been able to engineer so many amazing technological creations. I respect a company for taking risks to push boundries.

 

[cdimauro]

This quote is just false. See the previous comments...

 

[beginner99]

Realistically, I doubt ZEN will be much faster than Haswell but will have to compete on price.

Competing on price is irrelevant in the target market (Enterprise / Server). In the total cost of ownership of a server the CPU plays pretty much an irrelevant role. Whether it's 1k or 2k doesn't really matter if you are running software worth several 100k on it. You will but the faster one even if it cost 2k more.

The "moar cores" approach also has issues since a lot of software now is licensed by core. So having less beefier cores is preferable...

Price only matters in the consumer market but if Zen fails in server, Zen is dead and with it AMD.

 

[AtenRa]

Competing on price is irrelevant in the target market (Enterprise / Server). In the total cost of ownership of a server the CPU plays pretty much an irrelevant role. Whether it's 1k or 2k doesn't really matter if you are running software worth several 100k on it. You will but the faster one even if it cost 2k more.

The "moar cores" approach also has issues since a lot of software now is licensed by core. So having less beefier cores is preferable...

Price only matters in the consumer market but if Zen fails in server, Zen is dead and with it AMD.

Not true,

For small business servers (1-2 socket) 2k on savings can give you more HDD storage, or more ram etc etc.
For large corporate servers 1k savings per CPU can lower the initial cost of purchase a lot.
For data centers/cloud the savings are even more (thousands of CPUs)

 

[swilli89]

Competing on price is irrelevant in the target market (Enterprise / Server). In the total cost of ownership of a server the CPU plays pretty much an irrelevant role. Whether it's 1k or 2k doesn't really matter if you are running software worth several 100k on it. You will but the faster one even if it cost 2k more.

The "moar cores" approach also has issues since a lot of software now is licensed by core. So having less beefier cores is preferable...

Price only matters in the consumer market but if Zen fails in server, Zen is dead and with it AMD.

Points for being overly dramatic but you are pretty off-base with your thoughts.
Tons of server apps are actually open source these days and even then it's not about straight up performance in most situations, it's about scalability and power efficiency. There is definitely no movement in enterprise towards "less beefier" cores.

Also AMD is specifically targeting desktop enthusiast and OEM for an entire quarter with Zen. They are ad a company in no way completely dependant upon Zen succeeding in Enterprise. It's important yes but your statement is a little much.

 

[sirmo]

The "moar cores" approach also has issues since a lot of software now is licensed by core. So having less beefier cores is preferable...

Most growth is happening in cloud, and cloud is mostly powered by Linux.. even Microsoft has embraced it with Azure.. etc. So licensing per core isn't really a factor for AMD when you consider that they currently have less than 1% of the datacenter market share.

 

[beginner99]

Points for being overly dramatic but you are pretty off-base with your thoughts.
Tons of server apps are actually open source these days and even then it's not about straight up performance in most situations, it's about scalability and power efficiency. There is definitely no movement in enterprise towards "less beefier" cores.

Also AMD is specifically targeting desktop enthusiast and OEM for an entire quarter with Zen. They are ad a company in no way completely dependent upon Zen succeeding in Enterprise. It's important yes but your statement is a little much.

Open-Source doesn't mean free. This is a common misconception. If you are running Red Hat Linux, you are paying them and so forth. Now assume you are actually running a free Linux on a free web server you still need sys admins (and because no support more competent ones = more expensive). But my main point is the applications actually running on your free OS and web server can cost a lot be it because of the license or because it was developed in-house. Any medium complex software will cost millions do develop in-house. Add to that all the infrastructure needed in a server/data center like redundancy, the racks and so forth and the cost for the CPU is almost irrelevant.

Most growth is happening in cloud, and cloud is mostly powered by Linux.. even Microsoft has embraced it with Azure.. etc. So licensing per core isn't really a factor for AMD when you consider that they currently have less than 1% of the datacenter market share.

If you run a server in the cloud, you still have to pay the license for the software. Running a Windows Server on AWS doesn't mean you can just skip the Windows license or the oracle license. From Oracle:

For the purposes of licensing Oracle programs in an Authorized Cloud Environment, customers are required to count each virtual core as equivalent to a physical core. This policy applies to all Oracle programs available on a processor metric.

Yeah they talk about virtual cores but the concept still holds. The beefier a core is the less cores you need and hence pay less for licenses. But yeah, if you pay for Oracle you a lost soul anyway.

 

[sirmo]

If you run a server in the cloud, you still have to pay the license for the software. Running a Windows Server on AWS doesn't mean you can just skip the Windows license or the oracle license.

Oracle is enterprise. I am talking about AWS and OpenStack.. Hadoop, Cassandra, Docker swarms.. you name it, none of that stuff has licenses.

There are support contracts from various firms, but they don't charge by the core.

 

[itsmydamnation]

Most things have per proc licencing, very few are only per Core. On top of this Zen has more memory channels and thus you will get Better VM density as you are almost never CPU bound on a VM farm in enterprise and that is a big tangible TCO benefit from Zen.

If you have a really big DB that is really important you were never AMD's target as it looks like Zen doesn't scale above 2 proc.

This has been covered a lot in this thread already.

 

[krumme]

Nope, the analyst is referring to this interview (remember, the Barron's quote is quoting the analyst not Murthy himself). I think the analyst just "put two and two together," and in this case basically got 5.

Do you think that "Zen will disrupt the market" and "I haven't seen enough of the Zen to calibrate its exact competitive level, but we're treating it like it's going to be a legitimate competitor and therefore we need to deliver our best to make sure we're able to deliver the business performance that we expect to deliver over the course of the next 12 months" can really be reconciled?

10-nanometer desktop CPUs won't arrive until late 2018 at the very earliest, more likely in 2019. That's outside of the 12 month window that Murthy was talking about.

Yep. Sometime one have to wonder what those analyst do for a living. Is it fake news business or what?
I think a sounder forecast would be that zen+ on tsmc 7nm 2019 or 2020 will be where serious disruptive effect could happen to server side. Zen first gen will be a bit pushed guven amd situation and at that time amd will have ironed out the weakest parts. Improved the most needed and foremost make use of the new wsa agreement to use tsmc to fab the cpu at a 7nm high freq process node that for the first time in history will sort of rival Intel for both density and high freq performance. If zen shows promising ipc 2017, 2019 will be as interesting as 2017 imo.

 

[itsmydamnation]

i really doupt Zen+ is on TMSC, it will be on GF/IBM's 7nm process.

 

[Atari2600]

Yep. Sometime one have to wonder what those analyst do for a living. Is it fake news business or what?
I think a sounder forecast would be that zen+ on tsmc 7nm 2019 or 2020 will be where serious disruptive effect could happen to server side. Zen first gen will be a bit pushed guven amd situation and at that time amd will have ironed out the weakest parts. Improved the most needed and foremost make use of the new wsa agreement to use tsmc to fab the cpu at a 7nm high freq process node that for the first time in history will sort of rival Intel for both density and high freq performance. If zen shows promising ipc 2017, 2019 will be as interesting as 2017 imo.

It'd be great if AMD could commit to Zen+ going into any of the AM4 boards released H1 2017.

Would certainly bring an awful lot of clarity to upgrade pathways. I'm sure I'm not alone in wanting that.

 

[krumme]

i really doupt Zen+ is on TMSC, it will be on GF/IBM's 7nm process.

Ok because of what?
I would tend to think tsmc high perf 7nm process variant is a nice fit for a high margin server market? To offset the wsa demands and let apu and gpu go gf.

 

[majord]

i really doupt Zen+ is on TMSC, it will be on GF/IBM's 7nm process.

Depends if AMD plan to stick to there ~1 year New architecture update schedule, which has happend in the last two major uARCH's , regardless of process node in the case of Vishera.

K10: Dec 2007
K10.5: Jan 2009 (w/die shrink)

BD: Oct 2011
PD: OCt 2012

Zen: Q1 2017
Zen+ Q1/2 2018 ?

.. which would be too early for 7nm.

 

[KTE]

Most things have per proc licencing, very few are only per Core. On top of this Zen has more memory channels and thus you will get Better VM density as you are almost never CPU bound on a VM farm in enterprise and that is a big tangible TCO benefit from Zen.

In the UK, for enterprise, majority of the big $$ licensing is by Core, not Proc. Enterprise just doesn't get this hardware visibility anymore.

Case in point, they lease the infrastructure from datacentres, they lease VMs, and VM licensing is by Platform-> Core+Mem. Storage is peanuts in comparison and usually very flexible.

The datacentres for cloud/vms, yes, they would have this choice and VM consolidation would be their priority. But they generally buy prebuilt servers directly from OEMs. They buy packaged solutions, en-masse. They don't mess with maintenance and support.

Larger mem requests (and HUGE associated costs) are typically for archaic dinosaurs like AIX/Sparc/HP UX that are becoming a smaller and smaller part of enterprise infrastructure. I think health and government are the two segments really holding onto these platforms.

However, some of the enterprise major software costs are huge... hardware pales in comparison. From that perspective, it can be the Proc.

Sent from HTC 10
(Opinions are own)

 

[beginner99]

Oracle is enterprise.

Yeah and that's what companies use, you know the people that actually buy most servers.

 

[sirmo]

Yeah and that's what companies use, you know the people that actually buy most servers.

Listen dude I work in this field. There is a difference between enterprise apps like Oracle RDBMS products, and hyperscale stuff we mean when we say Cloud. When I say cloud I don't mean enterprise using cloud. I mean cloud providers, the exabyte storage, 100k+ node swarms.. etc. Amazon, Google, Azure, AliBaba.. those guys aren't paying licensing fees per core. I should know I work for one of those guys.

 

[swilli89]

Listen dude I work in this field. There is a difference between enterprise apps like Oracle RDBMS products, and hyperscale stuff we mean when we say Cloud. When I say cloud I don't mean enterprise using cloud. I mean cloud providers, the exabyte storage, 100k+ node swarms.. etc. Amazon, Google, Azure, AliBaba.. those guys aren't paying licensing fees per core. I should know I work for one of those guys.

Exactly. I work for a Fortune 100 company with over 80k employees all using our Oracle database and the computing demands are not as much as you would think. My friend manages sourcing/procurement for our primary Datacenter and while it's impressive, only 10% of the workload and traffic he says is due to Oracle/SAP and most of the work is in-house is geological number crunching.

 

[TheELF]

Exactly. I work for a Fortune 100 company with over 80k employees

I work for no one and have to ask,will large companies be willing to switch to a yet unproven technology?

 

[coercitiv]

I work for no one and have to ask,will large companies be willing to switch to a yet unproven technology?

Nah, they'll just use it in a few elevators and check for stability.

 

[sirmo]

I work for no one and have to ask,will large companies be willing to switch to a yet unproven technology?

Absolutely. If Zen can have competitive perf/watt some will switch. Why do people use Juniper when there is Cisco? There are always reasons. Besides AMD used to be a player in this field. Sun Microsystems invested in Opteron x86_64 in quite a big way back in the day.

I can tell you that some folks are just tired of having only one vendor. They want AMD if for anything but to be able to pit them against Intel and negotiate. Negotiations can then take any number of directions.

This is a big market too.. with fat margins. AMD doesn't need much of it to really boost their revenues.

 

[cdimauro]

Absolutely... no. It's known that Intel has developed custom extensions to its CPUs for specific (BIG) partners, which are enabled only for such partners.

So, it's not as easy as it can be to switch to another CPU vendor.

 

[Glo.]

Absolutely... no. It's known that Intel has developed custom extensions to its CPUs for specific (BIG) partners, which are enabled only for such partners.

So, it's not as easy as it can be to switch to another CPU vendor.

https://twitter.com/BitsAndChipsEng/status/805449548009246720

Well, lets just wait and see.

 

[cdimauro]

I know it. It might be Alibaba.

However I've said "it's not as easy as it can be".

 

[coercitiv]

I know it. It might be Alibaba.

However I've said "it's not as easy as it can be".

You also said "absolutely no"