[Leaplist] AnandTech on latest LGA-771 Xeon Quad-Core DDR2 v. "old" S940 Opteron Dual-Core DDR ...

[Kyle Gonzales]

If think it people will read the article, the takeaways are:

1) Stop buying anything based on the P4/Netburst architecture, and

2) The buying decision between Intel Core and AMD Opteron is getting
more difficult.

On Fri, 2006-11-10 at 18:43 -0500, Bryan J. Smith wrote:
> [ Since this covers SSL and MySQL benchmarks, I'll post to LEAPLIST ]
> 
> AnandTech took an "old" 2.4GHz Opteron dual-core from the Socket-940
> era with only DDR333 (PC2700), and _not_ the newer LGA-1207 (Socket-F)
> Opteron with DDR2, and pitted it against the latest'n greatest LGA-771
> Quad-Core Xeon monster from Intel with 8MB L3 cache.
> 
> Once again, *ABSOLUTELY*NO*I/O* benchmarking was done.  That removes
> the overwhelming majority of Opteron's advantage -- it's partial mesh
> interconnect (well beyond "switched," let alone Intel's "bus").  So
> this was 100% computational and memory interconnect testing -- and the
> memory affinity (especially for memory mapped I/O) was largely gone
> for Opteron as well.
> 
> Anand even threw the -O3 switch to ultra-optimize Intel with its
> inprecise lossy math SSE units over the Opteron's precise FPU
> operations (which it also uses for SSE).  So Intel Core should blow
> Opteron away at SSL and MySQL, right?
> 
> *WRONG*!
> 
> The Opteron still _trounces_ Intel at SSL, handily!  I was a bit
> surprised there.  I thought Intel's inprecise, lossy math SSE pipes
> would boost it.  Maybe the -O3 switch does not use Intel's SSE pipes
> as liberally as I thought it did -- or the GCC team changed the
> optimization (because of all the data loss due to doing so for 64-bit
> FP operations reported in the past)?  I'll have to research this a bit
> more.
> 
> At MySQL operations, they were the same MHz for MHz, although Intel
> has the 800MHz lead at 3.2GHz to only 2.4GHz.  Intel will offer up to
> 3.4GHz in the initial offering, while AMD does offer the newer
> LGA-1027 (Socket-F) in up to 2.8GHz.
> 
> Pricing is not too much more for Intel's solution.  But considering
> the SSL performance, why would you buy it?  I'd also like to see a
> well-designed I/O solution pushing the data for the SQL tables --
> something you could seriously tip in design in Opteron's favor even
> more.
> 
> http://www.anandtech.com/IT/showdoc.aspx?i=2872
> 
> 
-- 
Kyle Gonzales <kyle.gonzales at gmail.com>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
Url : http://lists.leap-cf.org/pipermail/leaplist/attachments/20061111/026cb680/attachment.bin