% | $
Quotes you view appear here for quick access.

Advanced Micro Devices, Inc. Message Board

  • pluscard_sucks pluscard_sucks Dec 16, 2008 8:52 AM Flag

    Dual Nehalem crushes Dual Shanghai in SAP


    At this point we have no idea how it is possible that a 3 GHz Nehalem outperforms the latest Opteron by a margin as high as 80% and more.

    Even without HT, Nehalem is almost twice as fast!

    AMD server market under greater pressure than ever!

    SortNewest  |  Oldest  |  Most Replied Expand all replies
    • Agreed! Can kiss goodbye the niches held by AMD.

    • from a well respected poster on fool

      Recommendations: 6

      Lol! I've been working for several hours on putting this non-story in context. Anandtech blew it pretty badly. I don't know where they got their Opteron numbers, but below are the most recent 20 SAP test reports, which make this a total non-story.

      First, there are two Shanghai numbers, which Anandtech forgot to mention. In fact they forgot all 8 way Opteron systems. So AMD is not in deep trouble, quite the opposite. When the data are sorted by SAPS score, the new Intel machines end up tenth and eleventh. Seventh through ninth are results for Intel's recent Dunnington 6 core per chip processor (4 sockets, 24 cores). Opterons are in second, fifth, and sixth with 8 socket 32 core systems. Of course the actual numbers matter more than position on a list of the twenty most recent submissions.

      The Shanghai system from Sun running Solaris, was in second place and scored 39270, the other Shanghai system (from HP) came in at 35400. (The system in first place was an IBM Power system with 16 chips and 64 cores.) More interesting is the Sun system with 4 Barcelona Opteron 8360SE CPUs, at 29670. If we scale for clock speed, 2.7 vs. 2.5, the Sun result for Shanghai shows an architectural improvement of 22.6%. Not as great as the Intel dual CPU results, but this is at the top end of the chart--and these systems use HT 1.11 (or if you prefer 2.0 ;-) Switching to HT 3.0 system designs should reduce interprocessor and NUMA latency significantly. Note that switching from Barcelona to Shanghai resulted in a 9600 increase in SAP score. The new dual Xeon system when released, will increase performance by 13400 over existing Intel dual socket systems. But you will probably have to replace the entire system. (New CPU, chipset, motherboard, and memory type.)

      For a reality check, people who buy high-end database systems are not that concerned about price/performance ratios. They care deeply about RAS issues and whether the system can handle their maximum load. And they don't change processor families (such as x86 to SPARC or the reverse), operating systems, or database vendors lightly. I'd say everything else is in tenth place, but total power consumption is creeping up. So the newer Intel systems have a big advantage over the old. (DDR3 vs. FB-DIMMs).

      Will Intel be a lot more competitive in the low-end database market with these new systems? Sure. They may even have a presence in the current mid-range systems (currently four-socket x86) market. But until Intel has quad or eight socket Nehalem system, they won't touch Opterons primary market. On the other hand, the improvements from Barcelona to Shanghia, make AMD more competitive with SPARC and IBM Power at the top end. Montreal and HT 3.0 should cement that.

      • 2 Replies to kallenijs
      • So, in other words AMD has nothing to worry about because the 8P opteron systems score higher SAPS than the 2P nehalem?

        Nothing you have written in these two verbose and rambling posts refutes the significance of a 2P X5570 system beating a 4P Opteron 8360SE system by a significant margin.

      • continuing:

        Oh, one other comment and then on to the chart. Note that all of the top results did not use SQL Server, or for that matter Windows 2003 Server Enterprise Edition. They both run out of steam around 30,000 SAPS. In fact, I'm pretty impressed that HP scored 35400 on SuSE Linux. (The Sun AMD results used MaxDB 7.6 on Solaris.)
        Techno SAP OSRelease RDBMS ServerName CPUs Cores Threads CPU Speed MemoryIBM 72250 AIX Version 6.1 DB2 9.5 IBM Power 570 16 32 64 POWER6 4.2 Ghz 252144Sun 39270 Solaris 10 MaxDB 7.6 Sun Fire X4600M2 8 32 32 Opteron 8384 2.7 Ghz 131072Sun 37650 Solaris 10 Oracle 10g SPARC T5440 4 32 256 UltraSPARC T2+ 1.4 Ghz 131072Fujitsu 37650 Solaris 10 Oracle 10g SPARC T5440 4 32 256 UltraSPARC T2+ 1.4 Ghz 131072HP 35400 SuSE Linux ES 10 Oracle 10g HP ProLiant DL785 G5 8 32 32 Opteron 8384 2.7 Ghz 131072Sun 29670 Solaris 10 MaxDB 7.6 Sun Fire X4600M2 8 32 32 Opteron 8360SE 2.5 Ghz 131072IBM 26550 Windows 2003 DE DB2 9.5 IBM System x3850 M2 4 24 24 Xeon X7460 2.66 Ghz 65536F/S 25720 Windows 2003 EE SQL Server F/S RX600 S4 4 24 24 Xeon X7460 2.66 Ghz 65536IBM 25850 Red Hat Linux 5.2DB2 9.5 IBM System x3850 M2 4 24 24 Xeon X7460 2.66 Ghz 65536HP 25000 Windows 2003 EE SQL Server HP ProLiant DL380 G6 2 8 16 Xeon X5570 2.93 Ghz 49125F/S* 23650 Windows 2003 EE SQL Server F/S* TX300/RX300 2 8 8 Xeon X5570 2.93 Ghz 49152Dell 21430 Windows 2003 EE SQL Server Dell R900 4 24 24 Xeon X7460 2.66 Ghz 122880HP 13780 Windows 2003 EE SQL Server HP ProLiant DL385 G5p 2 8 8 Opteron 2384 2.7 Ghz 32768NEC 10050 Windows 2003 EE SQL Server NEC ECO CENTER 2 8 8 Xeon L5420 2.5 Ghz 24576NEC 9780 Windows 2003 EE SQL Server NEC 5800 320Fd-MR 2 8 8 Xeon E5450 3 Ghz 24576Wipro 7620 Windows 2003 EE SQL Server Wipro NetBlade 1600 2 8 8 Xeon E5440 2.83 Ghz 32768Dell 5730 Windows 2003 EE SQL Server Dell M605 2 8 8 Opteron 2356 2.3 Ghz 32768Sun 4130 Solaris 10 Oracle 10g SPARC M3000 1 4 8 SPARC64 VII 2.52 Ghz 16384Fujitsu 4130 Solaris 10 Oracle 10g SPARC M3000 1 4 8 SPARC64 VII 2.52 Ghz 16384Dell 2570 Virt. Windows** SQL Server Dell R900 4 16 16 Xeon X7350 2.93 Ghz 90112
        * Fujitsu Siemens
        ** Windows 2003 Server running on two virtual CPUs on Windows 2008 Hyper-V

    • From the author in the comments section:

      "I have been a hardware journalist for 10 years now, and I never seen this. A new CPU + platform doubles the performance over a previous one without: 1) Using new instructions 2) a newer process technology 3) large jump in clockspeed or 4) running a very exotic benchmark that stresses only a very small part of the CPU."

      Looks like Intel's lead is increasing...


    • Are smilejens and jf3110 playing fangen today?


    • 1. When is the Xeon 5570 available?
      He's comparing an AMD 8 socket product out now in the market with an 2 socket Intel product in the future.
      2. What would the comparison look like with HT 3.0 and 3.0ghz Shanghai due out soon?
      3. What's the power/energy comparison?
      4. What's the price/performance comparison?
      5. What the comparison of #2 with other benchmarks?
      6. What the comparison when AMD switches to its HKMG gates?

      AMD can't take Nehalem lightly, but at least for now, Shanghai is out and Nehalem server is not. Energy looks good, overclocking indicates 3.0+ ghz is achievable in the near term, 8 socket is running, HT 3.0 is not far off, and HKMG like gates are forthcoming


    • Why would an IT manager buy anything except a Nehalem based server? Does AMD have anything coming in the next two years that could possibly compete with these Nehalem monsters? The AMD roadmaps would suggest that they don't!

      Of course, Intel won't exactly stand still for the next two years; instead, Intel is expected to deliver both another Tick and another Tock in that time! Will Intel be winning the benchmarks by 400% in two years?

      Will AMD even be able to survive a $2B/year burn rate for the next two years?

    • Well, it isn't like the AMD idiot moron fanboy longs weren't warned the Nehalem would throw AMD back to the slide-rule age....

      This isn't even the top bin.... AMD's presence in server is 3-6 months away from being completely removed. They will need to sell their boxes at Walmart during the red light special.

    • I think I hear the fat lady singing!

    • telecomcompdude_sucks telecomcompdude_sucks Dec 16, 2008 4:52 PM Flag

      Even quad Barcelonas lost to dual Nehalem. Imagine the power saved with only 2 processors.

7.49+0.48(+6.92%)4:00 PMEDT