TR Forums

I was reading this article in PC Worldhttp://www.pcworld.com/article/3158919/servers/amd-talks-tough-as-it-drums-up-support-for-32-core-zen-server-chip.html
and it said this "Companies won't make a switch to AMD overnight; it could take a year or more to ensure applications work on the new chips."
Is this true? Why wouldn't apps not work with it immediately

That's not what the statement means.  Businesses have hardware refresh cycles.  They don't just buy stuff whenever it comes out, it's all scheduled.  

In the enterprise space, everything gets validated (or at least should) before it goes into production. For applications that run inside user space (i.e. don't do much to the metal), validation should happen quickly. For a few companies on the development side, AMD should already be sampling Naples to ensure a quick transition.

The flip side are things that do deal with the low level. For example, VMware and other hypervisors do have to deal with some real differences that have historically existed between Opterons and Xeons. Then when hypervisors get updated, companies will then do their own testing for things like VM migration between the new chips and old chips to ensure things go right in production.

Certain hardware will also have to be validated to work with the new AMD chips too. I know that a few NIC cards will actually write to cache directly to shave a little bit of latency when handing off to a CPU. That stuff can be very finicky considering how closely the two parts work together.

Overall there shouldn't be much of a difference between bringing up Naples into a production environment than a new Intel platform unless you're doing hypervisor work or low level hardware stuff.

Yeah, the wording of the article is somewhat confusing if you're accustomed to thinking of an "application" as a single program. In this context it really means the entire application stack, for big cloud providers like Google and Amazon. That includes OS, device drivers, firmware on enterprise gear like SAS controllers and 10Gb+ NICs, and (possibly) a virtualization solution (depending on the role of the server).

TBH it's just a lousy article all around. Although they're talking about Zen Opterons, the pic at the top of the article is a marketing slide for AMD's ARM based server chip; and there's some brain-dead paraphrasing of statements from AMD, like "a higher core count matters as servers can do more".

Does anyone actually care about PCWorld anymore?

Of course there's no reason why existing x86 apps won't run on Zen but as a new platform with a completely new CPU design, new chipsets, new drives, new BIOS, new everything, I understand the caution. I wouldn't think twice about such matters in a home PC and trust AMD to have validated it all well but if you're running a big server farm for a big company you'd want to make sure there's gonna be a smooth transition. It's kinda scary to bet the server farm on an unproven (albeit compatible) CPU + platform.

ronch wrote:

Of course there's no reason why existing x86 apps won't run on Zen but as a new platform with a completely new CPU design, new chipsets, new drives, new BIOS, new everything, I understand the caution. I wouldn't think twice about such matters in a home PC and trust AMD to have validated it all well ...

You're more trusting than I am then.

I'll let you be the guinea pig; I have no plans to be an early adopter. Paying list price AND getting the "privilege" of beta testing a new platform just doesn't sound as attractive to me as it did 15+ years ago.

just brew it! wrote:

Does anyone actually care about PCWorld anymore?

Unfortunately, yes: the people higher up the food chain than us gerbils.

Never said I'll be an early adopter. Time and time again if you've been reading my posts in the comments sections you'll know that I am perfectly contented with my 4yo FX-8350 system and there's no reason for me to get anything better soon. Heck, if it weren't for Thief 4 I'd get an A8-7600 if my FX suddenly blew up.

Gotta say though, when I said I wouldn't think twice about getting a new platform right away I was a bit exaggerating. I'm not one to adopt early although when I got my 8350 it was out for only 2 months, but back then AM3+ has been out for a while and Vishera was pretty much just Bulldozer 1.1.

ronch wrote:

Never said I'll be an early adopter. Time and time again if you've been reading my posts in the comments sections you'll know that I am perfectly contented with my 4yo FX-8350 system and there's no reason for me to get anything better soon. Heck, if it weren't for Thief 4 I'd get an A8-7600 if my FX suddenly blew up.

Well, you do have "RYZEN?" in your sig!

Still rockin' an FX-8350 here too. But building an Intel box in the near future, since "someone" decided to give me a free motherboard.

You don't have to force yourself to get Intel though. Just sell that board and get an AM4 board. Not that I think there's any forcing going on.

Ryzen is in the cards, that's for sure, but I'll probably get a later iteration. I just need it to be significantly faster and more efficient than my FX and in case it falls behind Kaby it'll be priced more aggressively. Of course I'd much rather it matches Intel.

HappySOB wrote:

I was reading this article in PC Worldhttp://www.pcworld.com/article/3158919/servers/amd-talks-tough-as-it-drums-up-support-for-32-core-zen-server-chip.html
and it said this "Companies won't make a switch to AMD overnight; it could take a year or more to ensure applications work on the new chips." 
Is this true? Why wouldn't apps not work with it immediately

There's also the perspective of that statement being made in regards to applications running into bugs or issues in the uArch itself. Forget FDIV, modern chips have flaws discovered all the time. 

http://techreport.com/news/26911/errata-prompts-intel-to-disable-tsx-in-haswell-early-broadwell-cpus

http://techreport.com/review/13741/phenom-tlb-patch-benchmarked

http://www.theverge.com/2012/6/18/3092949/security-vulnerability-x86-64-intel-processor

http://danluu.com/cpu-bugs/

How about a better example. Take Skylake... Intel's own errata list is 32 pages in length.