Recently, Dell announced that they will be using AMD's Opteron procesors for their server PCs (even limited at that). While I am overjoyed by this news - being an AMD fan - I can't but cast suspicions on this deal. I know AMD has welcomed this but is this just a strategy by Intel to weaken AMD's case against it or is Dell genuinly interested? In the case of the former, Dell can go back to Intel very easily. Anyone following the server level processor benchmarks knows that the Opteron kicks the living daylights out of the Xeon.
I started a thread on FFA Forums about it and have recieved some interesting responses though it did get a bit off topic at the end over the Apple vs PC debate.
A member 'Harold' (id: hga) validly pointed out some good points:
It was noted on Slashdot that Dell has been selling plain AMD processors in a box for a while, even though they don't have any machines that will accept them. The theory was much like the above, although it seems unlikely it would be convincing in court.
In this particular case, it could be a combination of that plus the simple fact that from everything I've heard Intel just can't compete with AMD when you do SMP with more than two processors---and this isn't going to change with Intel's new stuff, which only addresses processor cores, not interconnection.
It probably has a lot to do with HP slowly getting its sales act together. One of the biggest problem AMD has had is issues with "tier 1" computer manufacturers:
IBM has made a huge investment in chipsets for Intel, and only recently officially added AMD servers to it's lineup.
(Even Supermicro, one of the premier "white box" motherboard etc. manufacturers only sells AMD servers on a separate website.)
Sun has outsourced most of their sales to "VARs" and resellers for more than a decade, and simply doesn't care if they do a good job. If you want to buy anything less than a E25K or equivalent, Sun won't give you the time of day. They are supposed to have the best (AMD based) x86 servers, but it doesn't matter how good their stuff is if you can't buy it, and from what I've read most people find they can't. (This is one of several reasons people expect Sun to die.)
HP has been playing the AMD game for a long time, but their sales organization is terribly messed up, and this is a matter of public record. E.g. in The Wall Street Journal their new top dog has commented that fixing it is his highest priority, and in one year his major accomplishment there has been to change the bureaucracy such that salesmen can now spend 40% of their time actually selling, up from 30%.
That leaves Dell, which maintained its ability to sell product, but has been Intel's ... poodle forever, and is now paying for that quite severely in the bottom line. It remains to be seen how much they'll embrace AMD; e.g. if they only offer 4-way or more servers then it's clearly just an attempt to staunch one place they're bleeding, and when Intel get's its interconnect act together Dell could well drop AMD.
About AMD having Intel beat, Harold had some other informative stuff to share:
A lot of this has to do with simple good decision making and execution: for the last few years Intel and AMD have swapped their traditional positions because of of this.
Intel made a number of critical bad decisions, like going with Rambus RDRAM, which was an obviously insane engineering decision, with terrible latency and requiring extremely high frequency PC board lines. That was perhaps the biggest one, requiring two separate one million part (!!!) recalls. The Pentium 4 Netburst architecture could not be pushed as fast as they expected (which delayed the return to the Pentium Pro approach everyone is using), and they put too much effort into Itaniac server chips instead of x86 ones.
Meanwhile AMD got its design and manufacturing act together, bought NexGen and then much of the Digital ALPHA design team which gave them great designs, very critically defined the x86 64-bit future, and came up with the HyperTransport (HT) interconnect as a standard.
HT is critical because Intel simply doesn't have any good stories in this area now or in the forseable future, and related, Intel's line of CPUs therefore require drastically more versions of chipsets (this really hurt them recently). HT also has the juicy potential of auxiliary CPUs: you can take any multi-chip Opteron board and e.g. plug one company's top of the line FPGA on an adaptor into the board.
This is attractive for special uses: the above example runs only at 500 MHz, but has 512 (!) FPUs you can use in parallel along with an embedded PowerPC and 10Mb of memory and plenty of reprogramable gates to connect everything. Number crunchers like the oil and gas industry are (in theory) salivating at this idea. Standardization through HT will hopefully result in all sorts of neat things in addition to higher volumes for chip sets, something we haven't seen work in the marketplace since FPUs were integrated in CPUs.
To finish this riff, HT will soon be the only advantage AMD has, since Intel has finally gotten its act together and is about to start shipping some very good cores designed in Israel. And AMD will probably not beat Intel in manufacturing, especially in the long run.... Basically Intel has always been very strong in manufacturing (whereas if you read e.g. the Wikipedia AMD entry you'll note that several times AMD wasn't able to get high enough yields to take advantage of a superior CPU), and there's the ugly fact that in revenues Intel is ten times bigger than AMD.
Of course AMD is laser focused on just the CPU market (they bailed out of the flash market within the last year), but it's hard to fight that sort of scale disadvantage.
It will be interesting to see how AMD will reply to this new challenge; it will be rough for a while, since they clearly have nothing significant in the works for this year.
Harold's comments are much appreciated as a contribution to the discussion. Thank you Harold!
I will end this update by saying that while I encourage this turn of events, I am not too chuffed until Dell pushes some more incentives and possibly the desktop market into the deal. Also, AMD must remain wary of Intel using this as a tactic to weaken AMD's case against Intel.
23/05: Dell and AMD: A date or a proposal?
Recently, Dell announced that they will be using AMD's Opteron procesors for their server PCs (even limited at that). While I am overjoyed by this news - being an AMD fan - I can't but cast suspicions on this deal. I know AMD has welcomed this but is this just a strategy by Intel to weaken AMD's case against it or is Dell genuinly interested? In the case of the former, Dell can go back to Intel very easily. Anyone following the server level processor benchmarks knows that the Opteron kicks the living daylights out of the Xeon.I started a thread on FFA Forums about it and have recieved some interesting responses though it did get a bit off topic at the end over the Apple vs PC debate.
It was noted on Slashdot that Dell has been selling plain AMD processors in a box for a while, even though they don't have any machines that will accept them. The theory was much like the above, although it seems unlikely it would be convincing in court.
In this particular case, it could be a combination of that plus the simple fact that from everything I've heard Intel just can't compete with AMD when you do SMP with more than two processors---and this isn't going to change with Intel's new stuff, which only addresses processor cores, not interconnection.
It probably has a lot to do with HP slowly getting its sales act together. One of the biggest problem AMD has had is issues with "tier 1" computer manufacturers:
IBM has made a huge investment in chipsets for Intel, and only recently officially added AMD servers to it's lineup.
(Even Supermicro, one of the premier "white box" motherboard etc. manufacturers only sells AMD servers on a separate website.)
Sun has outsourced most of their sales to "VARs" and resellers for more than a decade, and simply doesn't care if they do a good job. If you want to buy anything less than a E25K or equivalent, Sun won't give you the time of day. They are supposed to have the best (AMD based) x86 servers, but it doesn't matter how good their stuff is if you can't buy it, and from what I've read most people find they can't. (This is one of several reasons people expect Sun to die.)
HP has been playing the AMD game for a long time, but their sales organization is terribly messed up, and this is a matter of public record. E.g. in The Wall Street Journal their new top dog has commented that fixing it is his highest priority, and in one year his major accomplishment there has been to change the bureaucracy such that salesmen can now spend 40% of their time actually selling, up from 30%.
That leaves Dell, which maintained its ability to sell product, but has been Intel's ... poodle forever, and is now paying for that quite severely in the bottom line. It remains to be seen how much they'll embrace AMD; e.g. if they only offer 4-way or more servers then it's clearly just an attempt to staunch one place they're bleeding, and when Intel get's its interconnect act together Dell could well drop AMD.
About AMD having Intel beat, Harold had some other informative stuff to share:
A lot of this has to do with simple good decision making and execution: for the last few years Intel and AMD have swapped their traditional positions because of of this.
Intel made a number of critical bad decisions, like going with Rambus RDRAM, which was an obviously insane engineering decision, with terrible latency and requiring extremely high frequency PC board lines. That was perhaps the biggest one, requiring two separate one million part (!!!) recalls. The Pentium 4 Netburst architecture could not be pushed as fast as they expected (which delayed the return to the Pentium Pro approach everyone is using), and they put too much effort into Itaniac server chips instead of x86 ones.
Meanwhile AMD got its design and manufacturing act together, bought NexGen and then much of the Digital ALPHA design team which gave them great designs, very critically defined the x86 64-bit future, and came up with the HyperTransport (HT) interconnect as a standard.
HT is critical because Intel simply doesn't have any good stories in this area now or in the forseable future, and related, Intel's line of CPUs therefore require drastically more versions of chipsets (this really hurt them recently). HT also has the juicy potential of auxiliary CPUs: you can take any multi-chip Opteron board and e.g. plug one company's top of the line FPGA on an adaptor into the board.
This is attractive for special uses: the above example runs only at 500 MHz, but has 512 (!) FPUs you can use in parallel along with an embedded PowerPC and 10Mb of memory and plenty of reprogramable gates to connect everything. Number crunchers like the oil and gas industry are (in theory) salivating at this idea. Standardization through HT will hopefully result in all sorts of neat things in addition to higher volumes for chip sets, something we haven't seen work in the marketplace since FPUs were integrated in CPUs.
To finish this riff, HT will soon be the only advantage AMD has, since Intel has finally gotten its act together and is about to start shipping some very good cores designed in Israel. And AMD will probably not beat Intel in manufacturing, especially in the long run.... Basically Intel has always been very strong in manufacturing (whereas if you read e.g. the Wikipedia AMD entry you'll note that several times AMD wasn't able to get high enough yields to take advantage of a superior CPU), and there's the ugly fact that in revenues Intel is ten times bigger than AMD.
Of course AMD is laser focused on just the CPU market (they bailed out of the flash market within the last year), but it's hard to fight that sort of scale disadvantage.
It will be interesting to see how AMD will reply to this new challenge; it will be rough for a while, since they clearly have nothing significant in the works for this year.
Harold's comments are much appreciated as a contribution to the discussion. Thank you Harold!
I will end this update by saying that while I encourage this turn of events, I am not too chuffed until Dell pushes some more incentives and possibly the desktop market into the deal. Also, AMD must remain wary of Intel using this as a tactic to weaken AMD's case against Intel.
You can contact Harold at hga-amd@ancell-ent.com.
Karma: 416 [Add to karma]