Sun Joins Apple in the Intel Camp for x86 Chips

An anonymous reader writes "Don't worry, SPARC isn't being replaced by Itanic. However, Sun will start using Intel Xeon CPU's in their X86 servers. Further evidence that Intel's Core microarchitecture is winning back a lot of the business that AMD won with Opteron." More coverage at CNN Money and the International Herald Tribune.

Comments

Competition benefit / AMD warchest

[cblack]

Core 2 Duo does seem to offer some benefits over the current opteron line and I think it is great that server vendors can so easily switch between them for new models. I believe Sun has a fairly sizable portion of the x86 server market and it was good to see a company have such success with AMD CPUs. Overall I think the competition is a good thing, but I do worry a bit that AMD will have trouble regaining sales even if they have the better next gen technology due to decreased profits as they lose server vendor sales. I look forward to a next gen battle based primarily on merit.

This is a surprise

[Tim C]

I was at JavaUk06 last year, and in his keynote speech (one of) the Marketing VPs spent quite a lot of time extolling the virtues of their new line of SunFire servers, paying particular attention to their power:performance ratio compared to similar Xeon-based servers. Listening to him then, you'd have thought that Opterons were the best thing since sliced bread. Yes, I realise that his job is to push their current and up-coming products and solutions, but the main thrust of his talk was "Opteron-powered SunFire servers use far less power than those crappy, power-hungry Xeon servers".

Re:This is a surprise

[be-fan]

If this was last year, he was likely talking about the older (P4-based) Xeon, not the new (Core 2-based) Xeon.

But the interesting bit is

[$RANDOMLUSER]

...and for Intel to endorse Sun's Solaris operating system...

What's that about??

Re:So, ahhhh...

[TheThiefMaster]

No, they're using Intel chips in their line of servers that previously used AMD chips. For the pro-AMD slashdotters, this is "a very bad thing"(tm).

sun.com

[Daemonstar]

Of all the links posted in the summary, there's no link to the webcast on Sun's site about the story (01/22/07 @ 10:00 PST, Realplayer 10 required). :P

Sun is not abandoning AMD. Sun is adding Intel.

[gp310ad]

Which makes sense. When there are two competitive players whose product features and performance keep passing each other, why not give the customer a choice and at the same time exploit that competition to improve ones own position...

Questionable

[tomstdenis]

The Core 2 Duo has an awesome ALU, and it is definitely low power.

But they still suck for NUMA. Unless Sun is building desktops I don't see the point of the move until Intel starts rolling out CSI [which by that time AMD will be 65nm working on 45nm parts...].

For the desktop, hands down the Core 2 Duo is the winner. These things are just amazing. Even when overclocked the thing is so cold that the CPU fan turns off and the BIOS warns me (annoying... so I turned the warning off). In terms of IPC it matches the AMD offerings fairly well.

Tom

Re:Questionable

[Anonymous Coward]

AMD reached 65nm last quarter. Intel's already moved to 45nm, though, but AMD's trying to catch up. Still, one of Intel's big strengths is its silicon fabbing capability; some have said Intel isn't so much in the business of designing chips as building fabs.

Re:Sun needs this

[nwhitehorn]

Sun is moving ahead with their SPARC servers, and just taped out a successor to the Niagara. If you'd read the article, you'd know they are replacing their (quite excellent) AMD servers with Intel ones, not SPARC with anything. Sun has quite happily been selling both architectures for some time now.

I'm sceptical

[btarval]

I agree that competition is good for the consumer, but I have to wonder what the effect will really be on their AMD servers. In the server biz, there's a LOT more to it than today's CPU-intensive benchmarks. The other big thing is IO bandwidth, and this is where AMD has been far more competitive than Intel is. AMD appears to be able to continue this lead, based on both companies claimed roadmaps, the last time I looked.

One can only shove so much data across a single bus, and AMD seems to have realized this, while I don't see this as easily done from Intel.

One of the cool things about AMD is the Hypertransport bus. This allows one to offload various peripherals easily onto separate busses, while still allowing them to be shared across CPU's. Offloading PCI peripherals (for example) onto different busses allows one to achieve higher IO bandwidth. In contrast, Intel's current approach seems to be to shove more and more CPU's onto the same bus.

It's as if Intel has completely forgotten about how to keep the CPU busy - that's the main name of the game, and has been for years (to say the least). Idle CPUs are useless, and the more idle CPUs there are, the sillier it is, IMHO.

And AMD appears to be capable of outdoing Intel in the bandwidth area, for both memory and bus bandwidth.

So it looks to me like AMD will continue to be ahead of Intel as far as top-end server solutions go.

In short, I find this particular move puzzling. Sun has traditionally backed the best performance design, and I see Intel still lagging here overall. This strikes me as more of a marketing move, not a real engineering one. It will be interesting to see how popular these Intel-based servers remain.

Re:I'm sceptical

[cbreaker]

I agree with you. The thing is, even if RightSaidFred99 over there thinks Intel is just as good at SMP configurations, it's only NOW just starting to become a reality. AMD has been using HyperTransport since the first Opteron, released several years ago. You've been able to use 4-way Opteron boxes and achieve MUCH better overall system performance then you ever could with a Xeon. Think VMware. When a dual-CPU Xeon outperforms a Quad-CPU Xeon, there's something wrong with the bus architecture.

The "core" CPU is finally, after over 7 years, perhaps better then the current generation of AMD CPU's, but again, it's still based on the same old North-bridge configuration. While Intel has managed to bump up the speed on this bus a bit, and they can more easily support new and faster RAM because the CPU doesn't have the memory controller, it's still the same old. If you're doing 4-way or more, with heavy applications like busy ESX servers, you're going to get a LOT more performance out of your Opteron system, including 4-way systems utilizing multi-core CPU's. Just because CPU's are going dual and multi-core, doesn't mean enterprise servers will ship with only one socket.

I say Good for Intel, the Core CPU is a good one. But, if you look at everything Intel has been doing with their CPU line lately, you'll see that they are generally copying AMD in a lot of places, starting with EM64T (aka AMD64.)

Re:I'm sceptical

[afidel]

Glad you bought into the desktop class benchmarks that all the little internet sites like to publish. Meanwhile I'm happily using Opteron servers for serious N-Tier architecture where their performance per watt and lack of I/O bottlenecks is great. I also use some Intel Core based Xeon's for less demanding workloads where they prove to a good match of price/performance and power use per workload. I guess I actually research stuff, test it in my specific situation, and select the best product, unlike the many hacks in IT =)

Re:Oh Gosh! Sun 386 all over again?

[xenocide2]

The news here isn't that they're using x86 architecture. It's that they're using Intel chips. They already have AMD Opterons for sale, and they're adding Xeons. And Solaris 10 does run on x86. Some see it as a concession that their server CPU designs are little more than a niche market. That diagnosis is probably correct, and if Sun wants to ever dig themselves out of the "Sun is dying" meme, they'll have to start taking advantage of the fact that their competitors are engaged in a price war, one that's also cutting into profits. Sun can still pride themselves on quality server hardware, support contracts and integration, even if they aren't the designer behind most their chips.

Meanwhile, their Niagra still has some niche applications, and will grow as software is designed for dual and quad core chips. If Niagra does what they say it will, people will be forced to consider one Niagra unit versus 6 Xeon servers. Xeons may have fallen in price recently, but they're still not cheap, so that's a calculus that Sun might win some day.

Yeah, I think you're right

[Concern]

What I'm seeing now are people who went google-style with blades buying empty rackspace to cope with hosting providers' power per rack ratio.

Meanwhile Sun's sales guys are selling $14k 72 watt, 8-way, 32-thread T2000's that can replace multiple Opteron (or Core :) blades... IF you're within its application domain. Interesting gamble.

Most webapps probably are... not actually a lot of hot floating point, or math code in general, in that space. But you have to be very careful.

So, it's possible that Sun has turned their biggest disadvantage into their biggest advantage: they're in a niche! Yet they can design whole hardware architectures. So it frees them up to find ways to specialize, and it seems that there may be some payoffs there.

Re:Oh Gosh! Sun 386 all over again?

[Daishiman]

I really, really don't get where this "Sun is dying" thing is coming from. Having a bunch of friends at IBM and several telcos and consulting businesses, it is simply amazing the number of Sun Fire 25K machines being bought everywhere. These are 72-processor monsters that will set you back a cool $2 million each, and they're in pretty hot demand.

In the market for very large servers, there's only three choices: HP SuperDomes, IBM p590s and p595, and SunFire 25ks. The Sun machines have by far the largest market share, and with the support contracts they are making a pretty penny with each.

Re:Sun needs this

[BrainInAJar]

"Today you can get the same functionality with several cheap Intel boxes with either Windows or Linux."

Or UNIX for that matter. Solaris is free to use, and a support contract is about half the price of RHEL...

Sun Benefits

[dupup]

FWIW, it seems to me that Sun agreed to launch a Xeon-based server line in exchange for Intel pushing Solaris x86, rather than the other way 'round. In other words, the emphasis, from Sun's point of view, is that Intel will advocate in favor of Solaris, for which Sun offers Intel some bidness.

IAASE (I Am A Sun Employee), BTW.

Oh no you di-in't!

[JeanPaulBob]

Suh-NAP!

Re:Sun needs this

[TheRaven64]

Depends on the workload. For floating point, the T1 is pretty dire (the T2 is better). The thing it does really well is highly concurrent workloads. AMD have a faster interconnect than Intel at the moment, but they are both crippled by one thing: the speed of light (painfully slow). The amount of work that a modern processor could do while waiting for a cache miss is enormous. Intel currently have some very clever logic predicting and pre-fetching cache misses, but they still suffer from them. Sun, however, cheat. Each core on a T1 has 4 contexts (8 on a T2), so when you get a cache miss it just switches over to the next thread instantly and carries on executing. The T1 core will only sit doing nothing if all four cores are in the middle of a cache miss, while an Intel or AMD CPU will do this for a few hundred cycles on every single cache miss.

For desktop workloads, this isn't such a great thing; most current-generation desktop applications do all their work in one thread, so if that thread has a cache miss you still end up doing a load of waiting because the other threads are not using the CPU much. On a server with a few hundred (or thousand) concurrent users, however, there are always threads waiting to do something, so you can get a phenomenal amount of throughput from this. With the growth in web applications, I expect Sun to do very well.