IBM Opts for AMD

ExE122 writes "Since the unveiling of the low-cost, low-energy AMD Operton in 2003, Intel has been struggling in the server-grade processor insdustry. Now, IBM has announced their decision to use the AMD Opteron processor in their new line of BladeCenter servers. System x3455, x3655 and x3755 rack-mount servers, two-way Bladecenter LS21, and four-way LS41 blade servers sporting the new AMD processors have already been announced. IBM will continue this transition over the next three months.

From the article:
"IBM's choice is by all means an important victory over rival Intel, which is struggling to sell the remaining deposit of server processors before the general acceptance of Woodcrest X5100 chips. Unfortunately for Intel, at the end of the second quarter, Advanced Micro had 26 per cent of the market for servers built on personal computer chips, more than double its share a year earlier, according to Mercury Research."

Could this be lights out for Intel?"

Sun made the move in April

[SIGALRM]

Earlier this year, Sun announced new Galaxy servers based on the new 3.0GHz Opteron chips (called x56 chips). As part of this announcement, Sun announced 16 new benchmark records. Among the new records, (using Sun Studio Compilers btw):

• New SPEC CPU 2000 FPrates for V40z, beating Dell PowerEdge 6850 based on Xeon on similar benchmark
• New SPEC CPU 2000 FPrates for SunFire X4100, X4200 servers
• Best SPEC CPU 2000 FP numbers on SunFire X2100 servers

Re:That's an easy one.

[kingkade]

Not if AMD's stock price in the recent 6 months is any indication. it's gone from around 42 to 17. AMD has accomplished a lot but Merom, Conroe and Woodcrest seem way too impressive to deny. Intel may finally strike back, especially considering AMD still needs to move to a 65nm process by the end of the year, no less. Please correct me if I'm mistaken.

Re:"Could this be lights out for Intel?"

[MadEE]

Let's see... 100 minus 26... carry the 9... that leaves 74% share left for Intel, right?

Nope, Intel has 72.9% [com.com].

Re:That's an easy one.

[Barny]

Woodcrest (and conroe for that matter) are awsome single cpu solutions, the unified L2 cache and extended proccessing options allow them to bypass most speed problems of traditional code (cache coherancy thrashing, unoptomized code, etc) but all this gain is LOST when useing 2 CPUs (differant sockets) running in 64 bit mode, you will thrash the cache just like usual, the fancy microcode optimizer disables, it all falls apart.

Intel have put a lot of nifty tricks into these new chips, tricks that will only work in single socket desktop systems. If amd add these soputions to their chips (unified cache running as part of memory controler for instance) they will gain not only in single CPU solutions but due to the mesh bus network (hypertransport), it would be able to deal with coherancy much better.

Re:That's an easy one.

[Comatose51]

Your post seems to imply that AMD's stock price went down because of the Conroe and that the market has decided that the Conroe will crush AMD. That's misleading. Intel has recently slashed prices quite dramatically and initiated a new price war with AMD. That's the real cause of the stock price drop as margins in both companies are go down. The effects of new technology on Wall Street's thinking takes a little while to sink in. They're not nearly as quick as Slashdot.

Intel still has no answer to 4P and above Opteron

[Anonymous Coward]

Even with their new Woodcrest CPU finally making it out of vapourware status, Intel has no real answer to Opteron in the 4P and above server space. Itanium is a failed venture that is on life support. It often performs worse than Opteron systems much much cheaper, so no hope there.

Intel might regain a little marketshare in the 1P/2P server space with Woodcrest, but they're still in full retreat in the Enterprise market as more and more companies move to 4P+ servers.

Seeing as AMD are releasing Socket F and Rev G chips soon, much of the performance advantage Woodcrest might have had will vanish quickly.

Re:That's an easy one.

[kingkade]

Your post seems to imply that AMD's stock price went down because of the Conroe

I didn't imply anything or at least wasn't trying to. Ths stock price went down because of the repeated downgrading (first by UBS, if I believe). Since then even the Dell niche had little impact when it rose from around 30 up to 35 and then proceeded to tank toward 17. I agree that those downgrading is because of Intel dropping prices -- because they can afford to. or at least more than AMD can. Couple that with Conroe, and I really don't see what AMD can offer to compete. By the way I'm not against AMD b/c I'm an INTC holder b/c, it's the other way around. I dumped AMD after losing quite a bit :(. They do seem to be picking up some steam FWIW.

The effects of new technology on Wall Street's thinking takes a little while to sink in. They're not nearly as quick as Slashdot.

You're underestimating traders. They research their thousands (like me) to millions of dollars of investment in tech just like they do oil, etc. I bet you they know Otellini's bowel movements a factor that in. Compare that with some intel/amd fanboys drowning out most of the knowledgeable people on this site.

Uhh... this isn't new and it's FUD

[saleenS281]

IBM already has bladecenters with opterons... [ibm.com] why is this news? How is this a defeat for Intel? IBM is basically announcing a refresh of their current lineup... you can bet your ass they'll be doing the same thing when intel rolls out the new Xeon in full force as well.

Remember kids, just because you like to pull for the underdog, doesn't mean it's OK to make false statements about the king.

Spin Away..

[thesupraman]

Wow, AMD Really are spnning evrything they can get their hands on in the last few weeks - could it be that they are trying to divert attention away from something?
Let me think.. what was AMDs last real news?? When is the next major milestone in their processor lineup?

What goes around comes around.. Intel have been busy beavers for a while, relying on their rapidly aging netburst architecure, and hurting for it, while they got their next generation in order - perhaps AMD should have done a bit more work to have an answer ready... oh well.

IBM of course will do anything to divert server attention away from Intel, due to the fact that they have their own large-systems architecture to support - Power, and Intels ia64 is a competitor, whereas AMD have nothing even close to that market - of course IBM want people to like AMD and avoid Intel...

I think the correct technical term is 'Duh!'

Re:Lights out for PPC?

[twiddlingbits]

If the Power6 is 4-5GHz and they don't figure out a way to make it run cooler, then you'll need water cooling on the servers. That's not easy and not cheap. I've seen water cooling on racks (basically heat exchangers) but internal to a server is very different. Intel gave up on the high GHz due to HEAT and I don't recall the Power chips being much cooler than Intel chips. BTW, it's called the "Z-series" and it's all the IBM mainframes (except AS400s) from small to HUGE.

Re:without HyperTransport, AMD would be dead

[tetromino]

Intel mitigates its lack of a modern bus by giving each 771 socket its own FSB to the northbridge (making the northbridge much more complicated as a side effect), upping the frequencey of that FSB to 1333 MHz, and implementing a very clever memory prefetch algorithm to hide resulting northbridge latency. Plus, of course, tons of cache to help limit the amount of data you need to send over the bus. As a result, a dual Woodcrest will still kick a dual Opteron's ass, although Woodcrest on HyperTransport would have been much faster. Of course, all these tricks don't really scale to more than 2 sockets, which is why Intel is pushing so hard for 4-core chips.

Re:End of intel as likely as end of IBM

[rts008]

"But the current direction is much more efficent and has great performance, so I definatly think we'll see Intel regain market share."

I agree 100% for the short-term, but I don't count AMD out of the picture. (short-term: Intel kicks butt, medium-term, AMD kicks butt, long term: lather, rinse, repeat)

Note this: (http://www.eetimes.com/news/semi/showArticle.jhtm l;jsessionid=VCQM2KMTL4VXUQSNDBECKHSCJUMEKJVN?arti cleID=188700612)

(quote from above link) "EE Times: With fab 38, AMD is planning to install a very modern production line. But this fab will launch production only in two years. What are AMDs plans for the time between?

Udo Nothelfer: Presently, the main focus of Fab 36 is an aggressive ramp-up for 90-nm chips on 300 mm. Also we are about to complete the 65-nm technology qualification and will bring them to production in the second half of 2006.

EE Times: Are the technological issues solved, especially regarding lithography and materials?

Nothelfer: With the relevant problems, we are through. We are nicely on track and will reach our goal in the second half of the year. Our next major challenge will be the rapid ramp-up of 65-nm volume production. Our goal is quite aggressive: By mid-2007, we want to have the conversion to 65-nm in fab 36 done. "

He (Nothelfer) then starts talking about the 45 nm hurdles they are working on.

I'm sure Intel is also forging ahead (one hint: get rid of the wole FSB concept, HyperTransport works!), and it will take a long time (if ever) for AMD to break over the 50% market share, but AMD increasing it's market share by any significant amount has an effect on Intel.

BTW, I happen to be partial to AMD cpu's, but I am not a rabid fanboy, I have both AMD cpu's(2), and Intel cpu's (3) on my home network.

Core Duo is a big step up for Intel, and I expect them to take full advantage of this while they can, then AMD will step in with their volley.

Who wins? So far, the customers. Hopefully that will continue, as I for one welcome my Star Trek computer-like overlords!

Re:That's an easy one.

[edmudama]

Not all "blades" are single-socket implementations. Sun's flagship x86 blade is 4 sockets/8 cores using the Opteron 885, with up to 32GB of shared memory for those 4 sockets.

It would make sense for blades to appear in all sorts of configurations, depending on what application is being targeted and the available budget.

Re:That's an easy one.

[Wesley Felter]

all this gain is LOST when useing 2 CPUs (differant sockets) running in 64 bit mode, you will thrash the cache just like usual, the fancy microcode optimizer disables, it all falls apart.

Sorry, but there is no evidence for this. Real-world benchmarks show that 2-socket, 64-bit Woodcrest systems have good performance (usually better than Opteron).

Re:Sun made the move in April

[Wesley Felter]

The x3455 is basically an Opteron version of the x336.

Re:My thoughts exactly

[l3prador]

Aren't they selling off their XScale line [slashdot.org] to Marvell?

Re:Whatever happened to the Cell?

[GauteL]

The Cell processor [wikipedia.org] is an in-order execution chip with strong focus on floating point. For this reason it will probably be really good for scientific calculations and simulation, but not very practical for general server-applications.

The Opteron servers and the Cell servers will most likely live side by side in the product offering from IBM. Apples and Oranges and all that...

Re:That's an easy one.

[Zontar_Thing_From_Ve]

Your post seems to imply that AMD's stock price went down because of the Conroe and that the market has decided that the Conroe will crush AMD. That's misleading. Intel has recently slashed prices quite dramatically and initiated a new price war with AMD. That's the real cause of the stock price drop as margins in both companies are go down. The effects of new technology on Wall Street's thinking takes a little while to sink in. They're not nearly as quick as Slashdot.

Quite right. And please remember that the stock market also believed until a month or two ago that SCO was worth more than $4 a share and they STILL believe that it's worth over $2 a share. I don't think that I have to explain to anyone on Slashdot that the fact that this stock still has any value at all is a good indication that Wall Street really does not understand "this technology thingy" very well.

Re:That's an easy one.

[Anonymous Coward]

I'd like to see one of those benchmarks. Intel 64 bit mode was pig-slow on previous generation of Xeons and I haven't seen any benchmarks for Woodcrest running 64 bit code.

Intel's own tech docs

[Visaris]

Intel's own tech documents state that their famous "Micro-Ops Fusion" does not work in 64-bit mode. This feature is one of the main features that make the Core2 architecture as fast as it is. Losing this knocks performance down around 5 to 6 percent, which puts Opteron just that much closer.

xbitlabs [xbitlabs.com] has some more details.

No one is trying to say that Woodcrest isn't a good 64-bit chip. We are just telling it like it is: woodcrest does not gain much from 32 -> 64-bit code (in some cases it is a bit slower), while the Opteron gains decently almost across the board.

IBM Negates their own advantage.

[Glasswire]

For 4-way MP, one of the reasons that IBM Xeons (like x366) have been excellent performers is the sophisticated memory controller hub chipset ("X architecture", the 'Hurricane' chipset) which has an advanced L4 cache and snoop filter integrated. What IBM now calls the x3850 (basically same as x366) still has this advantage - and combined with the new Intel 'Tulsa' Xeon MP chips with huge L3 caches, IBM should probably outperform their generic design x3655 Opteron (and other Opteron 4-way boxes). The real reason, I suspect, why IBM is doing this is not so much technical (they have felt their Xeon MPs were competative against Opteron 4-ways) but that there is now an AMD market mindshare which had created a disadvantage in some corporate accounts that had adopted the other AMD 4-ways (generally the HP 585) and were happy with it - IBM felt shutout wherever AMD fanboys became dominiant in IT thinking.

In the 2-way DP space the new IBM x3650 with Xeon 5100 series Woodcrest (and any other Woodcrest based DP) will be as good or better than anything AMD throws at the DP space in 2006 -incl the new 2.8GHz F socket stuff.

Re:Lights out for PPC?

[stevesliva]

BTW, it's called the "Z-series" and it's all the IBM mainframes (except AS400s) from small to HUGE.

You missed the re-branding [ibm.com], obviously.

If you really want to debate semantics, I could point out that the AS/400 became the iSeries, which is of course now the System i, which runs on power5.