By: Michael S (already5chosen.delete@this.yahoo.com), January 12, 2011 3:01 am
Room: Moderated Discussions
Brett (ggtgp@yahoo.com) on 1/11/11 wrote:
---------------------------
>David Kanter (dkanter@realworldtech.com) on 1/11/11 wrote:
>---------------------------
>>>My numbers of 65% market share is for two socket servers and above if I remember correct.
>>>Single socket servers numbers are not worth the paper they are printed on.
>>>
>>>For four socket servers and above AMD has ~100% market share.
>>>Intel just introduced a four socket board after having abandoned the market for
>>>the past few years, so maybe Intel has non-zero numbers now.
>>>
>>
>>Sources please : ) Those numbers sound like utter bullshit. I know for a fact
>>that IBM dominates the 8S+ market, and they are pretty much a Xeon only company.
>>That means that AMD could not have had 100% market share.
>>
>>I know that AMD had >50% of the 4S+ market at one point (prior to Tigerton and
>>Nehalem), but I doubt they had >50% of the 2S market. Their marketshare has steeply
>>declined to well under 50% for the 4S market with the introduction of Dunnington
>>and Nehalem-EX though (especially the latter).
>
>Reading the AMD Q2 conference call AMD claims a past historical peak of 26% of
>the total server market. With higher market shares as the socket count goes up.
>
>IBM was the only company selling 4 socket Intel servers, as IBM developed a chip
>for the purpose, a VERY expensive chip.
That's 100% BS. You don't know what you are talking about.
At one point in history Sun didn't sell 4S Intel servers. All other 1-tier and 2-tier x86 server vendors always had 4S Intel-based server machines. You can validate it, for example, by browsing through SAP 2-tier benchmark database.
http://www.sap.com/solutions/benchmark/sd2tier.epx
As to glue vs glueless, as you should know all Intel processors until Nehalem had to access memory through external Northbridge. Is it considered a glue? If it is, then yes, 4S Intel systems were not glueless, but then 1S and 2S Intel systems were also not glueless. Did it prevent vendors of using them?
Intel, and at one point Broadcom, supplied the chipsets and server vendor were free to build their solutions around this chipset unless they wanted to differentiate themselves from the crowd, like IBM and Unisys, by providing either higher performance at 4S or ability to scale to 8+ sockets.
The only real technical difference was that with Opteron you could build 8S server from off-the-shelf parts, However, until Shanghai it made little difference, since (until Shanghai) performance of glueless 8S Opterons was horribly bad.
>Being sold against AMDs glueless (free)
>4 and 8 socket designs. I have never seen market share numbers for this machine
>and assumed it was negligible on a unit basis.
4S is small, but not negligible. Probably, around 3.5% right now and that's the lowest point in 10+ years.
8S+ is indeed negligible.
>On a dollar basis combined with raid
>arrays and software, IBM "server" sales are huge, but only a tiny part of that would
>be CPUs. IBM has bundled 4 socket POWER sales and printed combined numbers, but
>I do not not have even those. And again that would be total system dollar sales
>and impossible to figure out what part was CPUs.
>
>I would love to have real 2 socket and 4 socket unit or dollar share numbers for any years.
>I am even willing to take IDC or Gartner numbers, as flawed as those can be.
>
>A quick google has HP and IBM arguing about who is bigger and who is growing in
>servers, both claim #1 and stealing market share from the other, both quote the
>other as lying and they giving the truth. ;)
>
>I put a tilde on my ~100% AMD market share for 4 socket servers for a reason, the IBM unknown.
>
You have no idea. That's all.
---------------------------
>David Kanter (dkanter@realworldtech.com) on 1/11/11 wrote:
>---------------------------
>>>My numbers of 65% market share is for two socket servers and above if I remember correct.
>>>Single socket servers numbers are not worth the paper they are printed on.
>>>
>>>For four socket servers and above AMD has ~100% market share.
>>>Intel just introduced a four socket board after having abandoned the market for
>>>the past few years, so maybe Intel has non-zero numbers now.
>>>
>>
>>Sources please : ) Those numbers sound like utter bullshit. I know for a fact
>>that IBM dominates the 8S+ market, and they are pretty much a Xeon only company.
>>That means that AMD could not have had 100% market share.
>>
>>I know that AMD had >50% of the 4S+ market at one point (prior to Tigerton and
>>Nehalem), but I doubt they had >50% of the 2S market. Their marketshare has steeply
>>declined to well under 50% for the 4S market with the introduction of Dunnington
>>and Nehalem-EX though (especially the latter).
>
>Reading the AMD Q2 conference call AMD claims a past historical peak of 26% of
>the total server market. With higher market shares as the socket count goes up.
>
>IBM was the only company selling 4 socket Intel servers, as IBM developed a chip
>for the purpose, a VERY expensive chip.
That's 100% BS. You don't know what you are talking about.
At one point in history Sun didn't sell 4S Intel servers. All other 1-tier and 2-tier x86 server vendors always had 4S Intel-based server machines. You can validate it, for example, by browsing through SAP 2-tier benchmark database.
http://www.sap.com/solutions/benchmark/sd2tier.epx
As to glue vs glueless, as you should know all Intel processors until Nehalem had to access memory through external Northbridge. Is it considered a glue? If it is, then yes, 4S Intel systems were not glueless, but then 1S and 2S Intel systems were also not glueless. Did it prevent vendors of using them?
Intel, and at one point Broadcom, supplied the chipsets and server vendor were free to build their solutions around this chipset unless they wanted to differentiate themselves from the crowd, like IBM and Unisys, by providing either higher performance at 4S or ability to scale to 8+ sockets.
The only real technical difference was that with Opteron you could build 8S server from off-the-shelf parts, However, until Shanghai it made little difference, since (until Shanghai) performance of glueless 8S Opterons was horribly bad.
>Being sold against AMDs glueless (free)
>4 and 8 socket designs. I have never seen market share numbers for this machine
>and assumed it was negligible on a unit basis.
4S is small, but not negligible. Probably, around 3.5% right now and that's the lowest point in 10+ years.
8S+ is indeed negligible.
>On a dollar basis combined with raid
>arrays and software, IBM "server" sales are huge, but only a tiny part of that would
>be CPUs. IBM has bundled 4 socket POWER sales and printed combined numbers, but
>I do not not have even those. And again that would be total system dollar sales
>and impossible to figure out what part was CPUs.
>
>I would love to have real 2 socket and 4 socket unit or dollar share numbers for any years.
>I am even willing to take IDC or Gartner numbers, as flawed as those can be.
>
>A quick google has HP and IBM arguing about who is bigger and who is growing in
>servers, both claim #1 and stealing market share from the other, both quote the
>other as lying and they giving the truth. ;)
>
>I put a tilde on my ~100% AMD market share for 4 socket servers for a reason, the IBM unknown.
>
You have no idea. That's all.
