3/2     How is an Opteron different from an Athlon64?
        \_ Most Opterons require registered DIMM (940pin) and most
           have 3 HT links.  1xx series only have 1 usable one, though.
           There're some 939pin Opterons, which are basically identical
           to Athlon 64.  All Athlon 64s have 1 HT link only and use
           unbuffered DIMMs.  Internally, they're identical other than
           cache size and missing HT link logic.  DRAM controller is
           actually the same, but depending on which pads get routed to
           the pins, it turns into A64 or Opteron.  -- AMD employee.
        \_ one has more vowel than the other
           \_ Thanks!  Can you answer one more question though?  I thought
              the 939 pin opterons (165,170,175) had 3 HT links (not that
              it matters) and the only difference between those and 940 pins
              was the pin count and memory required?  Is there any difference
              between any of 165,170,175 and the same speed/cache X2? -!op
              \_ There is practically no difference between 939pin
                 Opteron and 939pin A64.  Apparently, MANIDs are
                 different, though, which some BIOS may not like.  I
                 believe Opteron 939pin is multiplier unlocked, but
                 don't quote me on that.  FX parts are all multiplier
                 unlocked.  All 939pin parts have only 1 HT link, so
                 939pin Opterons can only be used on 1 socket systems.
                 Also, in case anybody cares, the biggest win for 939pin
                 is that vendors can build cheap 4-layer motherboards for
                 them.  940pin requires more layers.  Layers on motherboard
                 are relatively expensive.  Also, unbuffered DIMMs are
                 cheaper.  All K8 is capable of supporting ECC memory, but
                 not many BIOS in the consumer boards enable it.
                 \_ I've seen references to layer counts in motherboards at
                    review sites and all they ever said was "more is better!".
                    What's it really mean?  Thanks again.
                    \_ This is pretty simple.  Think all those tightly
                       packed pins.  Now, you need to route them a
                       certain way without them colliding.  Obviously,
                       if they collide, that's an electrical short.  Any
                       short you didn't intend is bad.  There are only
                       so many ways to route them in our 3-dimensional
                       world.  In fact, it's more like 2-dimension +
                       however many layers you add.  More layer really
                       shouldn't imply better quality other than the
                       fact that you can isolate the power and ground
                       planes better.  You can also route things in
                       such ways to reduce noise (add more inductive
                       loops and keep certain signals away from each
                       other).  If you're current limited (crazy
                       overclockers and what not), more may be of some
                       use, too.  More layers can also be bad since
                       there will be more points of failures,
                       particularly the layer-to-layer connections.
        \_ http://tinyurl.com/mcgr3 (amd.com)
           Opteron has more HT links (3 v. 1) and more cache than a
           Athlon64.
           \_ I think the above link is dated. The reason I think so is
              that when comparing prices I found the Athlon64 FX60 to
              cost more than the equivalent dual-core Opteron. Why is
              that? Also, along those lines, anyone have links comparing
              the performance of a Opteron 280 to an Opteron 250? They
              are the same clock speed, but the 280 is dual core. Are you
              better off with a 254 (unicore but 20% faster CPU)? This is
              for pure floating point computing, let's say monte carlos
              running threaded.
        \_ You can unscrew an Athlon64!