h1

Shanghai Economics 101 – Continued

May 4, 2009

Let’s look at some more real world applications of what we’ve learned from the VMmark results for Nehalem and what it means in a practical comparison. We’ll award Nehalem-EP’s SMT a 25% bonus for in our comparisons when vCPU/core count is taken into the measurement. In a 6:1 consolidation, this means 60 vCPU’s for 2P Nehalem and 48 vCPU’s for Shanghai. Using this bias, the following cost characteristics are revealed for VM’s with average memory footprints of 1.5GB, for the Nehalem-EP 3.2GHz system:

Nehalem-EP Configuration Street $ 1536MB VM’s, 1 vCPU’s Max vCPU’s (6/c) Cost/VM
2P/8C, Nehalem-EP, W5580 3.2GHz, 6.4GT QPI with 24GB DDR3/1333 $7,017.69 13 60 $539.82
2P/8C, Nehalem-EP, W5580 3.2GHz, 6.4GT QPI with 48GB DDR3/1066 $7,755.99 28 60 $277.00
2P/8C, Nehalem-EP, W5580 3.2GHz, 6.4GT QPI with 72GB DDR3/800 $8,708.19 42 60 $207.34
2P/8C, Nehalem-EP, W5580 3.2GHz, 6.4GT QPI with 96GB DDR3/1066 $21,969.99 57 60 $385.44
2P/8C, Nehalem-EP, W5580 3.2GHz, 6.4GT QPI with 144GB DDR3/800 $30,029.19 60 60 $500.49
2 x 2P/8C, Nehalem-EP, W5580 3.2GHz, 6.4GT QPI with 144GB DDR3/800 $60,058.38 120 120 $500.49

We’ll compare this to a Shanghai 2P system at 3.1GHz vs. the Nehalem-EP system:

Shanghai 2P/HT3 Configuration Street $ 1536MB VM’s, 1 vCPU’s Max vCPU’s (6/c) Cost/VM Savings per VM Savings %
2P/8C Shanghai, 2393 SE, 3.1GHz, 4.4GT HT3 with 32GB DDR2/800 $5,892.12 18 48 $327.34 $212.48 39.36%
2P/8C Shanghai, 2393 SE, 3.1GHz, 4.4GT HT3 with 48GB DDR2/800 $6,352.12 28 48 $226.86 $50.14 18.10%
2P/8C Shanghai, 2393 SE, 3.1GHz, 4.4GT HT3 with 64GB DDR2/533 $6,462.52 37 48 $174.66 $32.68 15.76%
2P/8C Shanghai, 2393 SE, 3.1GHz, 4.4GT HT3 with 80GB DDR2/667 $8,422.12 47 48 $179.19 $28.14 13.57%
2P/8C Shanghai, 2393 SE, 3.1GHz, 4.4GT HT3 with 96GB DDR2/667 $11,968.72 48 48 $249.35 $136.09 35.31%
2P/8C Shanghai, 2393 SE, 3.1GHz, 4.4GT HT3 with 128GB DDR2/533 $14,300.92 48 48 $297.94 $202.55 40.47%
2 x 2P/8C Shanghai, 2393 SE, 3.1GHz, 4.4GT HT3 with 128GB DDR2/533 $28,601.83 96 96 $297.94 $202.55 40.47%

And against a 4P Shanghai 3.1GHz system vs. the Nehalem-EP system:

Shanghai 4P/HT3 Configuration Street $ 1536MB VM’s, 1 vCPU’s Max vCPU’s (6/c) Cost/VM Savings per VM Savings %
4P/16C Shanghai, 8393 SE, 3.1GHz, 4.4GT HT3 with 48GB DDR2/800 $13,789.58 28 96 $492.49 $47.34 8.77%
4P/16C Shanghai, 8393 SE, 3.1GHz, 4.4GT HT3 with 64GB DDR2/800 $14,189.58 37 96 $383.50 -$106.50 -38.45%
4P/16C Shanghai, 8393 SE, 3.1GHz, 4.4GT HT3 with 80GB DDR2/800 $14,589.58 47 96 $310.42 -$103.08 -49.72%
4P/16C Shanghai, 8393 SE, 3.1GHz, 4.4GT HT3 with 96GB DDR2/800 $17,512.87 57 96 $307.24 $78.20 20.29%
4P/16C Shanghai, 8393 SE, 3.1GHz, 4.4GT HT3 with 160GB DDR2/667 $26,413.87 96 96 $275.14 $225.34 45.02%
4P/16C Shanghai, 8393 SE, 3.1GHz, 4.4GT HT3 with 256GB DDR2/533 $33,410.47 96 96 $348.03 $152.46 30.46%

Note the relatively high start-up cost of the 4P system cause some problems economically for 4P in the 64 and 80GB configurations. This shows where “modern consolidation rules” favor 2P systems where low consolidation ratios (high memory-to-CPU ratios) prevail. In fact, the economics for real world applications of 4P consolidations do not prove advantageous until 96GB of memory is allocated in an Microsoft SQL Server consolidation. Note that the cost savings associated with license reuse become much more significant in these systems (not accounted for in this material cost comparison.)

Shanghai 4P/HT3 Configuration Street $ 12288MB VM’s, 4 vCPU’s Max vCPU’s (3/c) Cost/VM Savings per VM Savings %
4P/16C Shanghai, 8393 SE, 3.1GHz, 4.4GT HT3 with 96GB DDR2/800 $17,512.87 7 48 $2,501.84 $636.73 20.29%
4P/16C Shanghai, 8393 SE, 3.1GHz, 4.4GT HT3 with 160GB DDR2/667 $26,413.87 12 48 $2,201.16 $2,088.73 48.69%
4P/16C Shanghai, 8393 SE, 3.1GHz, 4.4GT HT3 with 256GB DDR2/533 $33,410.47 12 48 $2,784.21 $1,219.69 30.46%

Disclaimer: the following is for comparative analysis and represents pricing that has neither been established nor disclosed in any way to SOLORI. These numbers have been derived by taking the current price of Opteron 8393SE and adding 15%. Please do not e-mail me looking for Istanbul processors.

Assuming Istanbul ships at a 15% premium, clock-for-clock, the “SQL consolidation” example, the “value” of Istanbul becomes fairly apparent:

Istanbul 4P/HT3 Configuration Street $ (est.) 12288MB VM’s, 4 vCPU’s Max vCPU’s (3/c) Cost/VM Savings per VM Savings %
4P/24C Istanbul, 8493 SE, 3.1GHz, 4.4GT HT3 with 96GB DDR2/800 $19,340.68 7 72 $2,762.95 $375.62 11.97%
4P/24C Istanbul, 8493 SE, 3.1GHz, 4.4GT HT3 with 160GB DDR2/667 $28,241.68 13 72 $2,172.44 $2,117.45 49.36%
4P/24C Istanbul, 8493 SE, 3.1GHz, 4.4GT HT3 with 256GB DDR2/533 $35,238.28 18 72 $1,957.68 $2,046.21 51.11%

Perhaps a more interesting real world case is consolidation of Windows Server 2008 requiring 6GB RAM and 2 vCPU per VM. This makes a good use case for Istanbul 2P systems vs. Nehalem-EP in Server 2008 consolidations:

Istanbul 2P/HT3 Configuration Street $ (est.) 6144MB VM’s, 2 vCPU’s Max vCPU’s (6/c) Cost/VM Savings per VM Savings %
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 32GB DDR2/800 $6,294.04 4 72 $1,573.51 $765.72 32.73%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 48GB DDR2/800 $6,754.04 7 72 $964.86 $143.14 12.92%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 64GB DDR2/533 $6,864.44 10 72 $686.44 $105.21 13.29%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 80GB DDR2/667 $8,824.04 12 72 $735.34 $56.32 7.11%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 96GB DDR2/667 $12,370.64 15 72 $824.71 $639.96 43.69%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 128GB DDR2/533 $14,702.84 20 72 $735.14 $570.47 43.69%
2 x 2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 128GB DDR2/533 $29,405.68 41 144 $717.21 $588.41 45.07%

The economics of consolidation systems does not appear to favor 4P systems with the exception of some very special cases where access to multiple-threads and multi-gigabytes of memory are needed by the virtual machines. All other workloads seem to favor 2P systems in $/VM computations. Since 2P systems can be completely turned-off in DRS-driven DPM (VMware) the power saving promised in “better” 4P systems has dubious application.

Perhaps this explains the absence of road-mapped 8P systems from AMD and Intel for their new architectures. If socket G34 delivers 2P and 4P within the same $/CPU cost, these equations will change in favor of 4P systems. Why? The dollar cost of memory, and sliding performance scale as memory buses are loaded, allows 4P systems to out-perform 2P systems with the same memory footprint. Hence, G34 is AMD’s future “performance oriented” platform.

What’s more, it looks like Intel has really done their homework – zeroing-in on the price/performance/capacity sweet-spot that makes the economics most favorable (their 72GB x 2P platform). However, using some statistics provided by the University of Chicago still shows a $/VM advantage in Opteron’s favor – skewing today’s usage towards the 32GB platform.

Istanbul 2P/HT3 Configuration Street $ (est.) 1569MB VM’s, 1.45 vCPU’s Max vCPU’s (2.33/c) Cost/VM Savings per VM Savings %
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 32GB DDR2/800 $6,294.04 18 27.96 $349.67 $190.15 35.23%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 48GB DDR2/800 $6,754.04 19 27.96 $355.48 $129.27 26.67%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 64GB DDR2/533 $6,864.44 19 27.96 $361.29 $182.98 33.62%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 80GB DDR2/667 $8,824.04 19 27.96 $464.42 $79.84 14.67%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 96GB DDR2/667 $12,370.64 19 27.96 $651.09 $722.04 52.58%
2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 128GB DDR2/533 $14,702.84 19 27.96 $773.83 $1,102.99 58.77%
2 x 2P/12C Istanbul, 2493 SE, 3.1GHz, 4.4GT HT3 with 128GB DDR2/533 $29,405.68 38 55.92 $773.83 $1,102.99 58.77%

21 comments

  1. There is one big problem with your analysis. You chose the most expensive Intel workstation processor possible which is more powerful than two Shanghai systems. You should be comparing processors of similar performance. The Intel Xeon E5520 2.26 GHz ($407) is as fast or faster than the AMD Shanghai 3.1.

    Like


  2. Here’s a much better comparison of price.

    Using the HP D380 G6 with Intel E5520 and 72 GB of RAM, you can build a system for $5289. This system is as fast or faster than the AMD Shanghai 3.1, and it gives you the lowest cost per VM.

    Prices can be verified below.

    http://h10010.www1.hp.com/wwpc/us/en/sm/WF06b/15351-15351-3328412-241644-241475-3884082-3907766-3907767.html

    http://shop.kingston.com/partsinfo.asp?promo=PRCGRBR&ktcpartno=KVR1066D3Q4R7S/8G

    Like


  3. To be fair, we chose top-priced processors from both Intel and AMD. Likewise, we attribute the Nehalem-EP with a 25% increase in processor loading (see 60 vCPU’s vs. Shanghai’s 48 vCPU’s). We feel that this empirical advantage more than offsets the implied performance differences from VMmark.

    Likewise, since our study is based on VMmark results, only processors with VMmark scores are being compared. The exception is “Istanbul” which, as disclaimed, is assumed to operate as Shanghai (factoring no improvement in HT3 efficiency) with only the addition of 2-cores per processor.

    Like


  4. The Kingston part is the one we use at $1,168/DIMM. That means your system with 72GB would be $10,500 in memory alone. With 18x4GB, the part we us is:

    http://kingston.pricegrabber.com/rd.php?pg=p~~7&r=2541&m=716393872&q=n&mt=~~138.00~138.00~~~0~~y~~n&k=3b37cc98993bc368771d5f6e80ffb355&dl=1&search_id=4674a5ab073e404976e90798dcf2f83f&set=1241030397&source=mlink

    and each is $138. That’s $2500 in memory alone, but more comparable. However, that’s also a DDR3/800 which will drop the system performance considerably (in which case I’m sure we’d be attacked for stifling the performance to make our case…)

    Like


  5. http://www.dailytech.com/Article.aspx?newsid=15036&red=y#439304
    I had VMmark scores here comparing Nehalem 2P to Shanghai 4P. It essentially shows close to a 2-to-1 advantage clock for clock core-for-core. Therefore your comparison is seriously flawed since a 2.26 GHz Nehalem most likely performs better than a 3.1 GHz Shanghai with a similar number of cores and processors.

    Like


  6. VMmark results are pretty public. You can find them here.
    http://www.vmware.com/products/vmmark/results.html

    Here’s the fastest 2P system based on a 2.93 GHz Nehalem-EP with a score of 24.14@17tiles
    http://www.vmware.com/files/pdf/vmmark/VMmark-Cisco-2009-04-21-B200M1.pdf

    The best 2P AMD score is still based on a 2.7 GHz Shanghai system from HP with a score of 11.28@8 tiles
    http://www.vmware.com/files/pdf/vmmark/VMmark-Dell-2008-11-12-R805.pdf

    They do have a 4P 3.1 GHz Shanghai system scoring 22.11@15 tiles.
    http://www.vmware.com/files/pdf/vmmark/VMmark-HP-2009-04-24-DL585G5.pdf

    Now when we can build a 2-Processor 2.93 GHz Nehalem-EP system that performs better than a 4-processor 3.1 GHz Shanghai for less money, that’s pretty amazing. It also shows how ridiculous it is for you to claim that a 2P Nehalem-EP 2.26 GHz is equivalent to a 2P 2.5 GHz Shanghai. You do realize that a 2.26 GHz Nehalem-EP is identical to a 2.93 GHz Nehalem with except of a 29.6% clock speed increase right? That means we would expect a 2.26 Nehalem-EP to perform within 77+ percent of a 2.93 GHz Nehalem-EP system.

    Like


  7. And, as has been my point, the Nehalem-EP turns-in a much less worthy score with 18x4GB DIMMS (i.e. running DDR3/800) on a more “typical” platform:

    2P, 2.93GHz/3.2GHz Turbo = 14.22@10 tiles.

    Logically you’d be forced to compare that result with the HP 385G5 score of 11.28@8 tiles. Clearly, the other scores represents some tuning in vSphere that take better advantage of Nehalem-EP’s SMT that does not exist in ESX 3.5.0 U4.

    If you look at the per-tile ratios of the results, another thing is clear: Nehalem’s tile ratio is between 1.37 and 1.45 where Shanghai’s is 1.40, so processor efficiency appears to be predominantly in the SMT side of the equation.

    Therefore, Istanbul’s 6-core drop-in upgrade should fair pretty well against Nehalem with vSphere – not bad for a 3-year old platform. You can not say the same for the millions of dollars in obsolete Xeon gear out there…

    For today’s loads, Nehalem and Shanghai are pretty well matched in price-performance: even if it takes 4 Shanghai servers for every 3 Nehalem servers. Istanbul – as you’ll see in my conclusion blog – changes that again.

    Like


  8. I do find it laughable that you have to rely on comparisons to Netburst based Xeon from 3 years ago to make your point. I’m not sure how that’s remotely relevant for someone buying a server today.

    You’re also relying on a sub-optimal Nehalem result to the best possible AMD result which is misleading. The fact is that a 4P Shanghai gets blown away by a 2P Nehalem with LOWER CLOCK SPEED. That’s more than a 2 to 1 blowout which is a pretty hard to spin.

    Now there’s no doubt that Istanbul will significantly close the gap. It will probably narrow to around 70% the performance of a Nehalem rather than Shanghai’s 50% of Nehalem. Then Intel will widen that gap again with Nehalem-EP.

    Like


  9. You’re way of base and out of bounds, George. The only inkling of “Netburst” bashing is simply fact: obsolete and redundant. The same can not be said for those who invested in Opteron in the last three years: they still have good value left in their systems and will continue to have an excellent eco-system foundation.

    As for the “sub-optimal” Nehalem, I’m sure if they could have tuned it better we’d have seen a better score. Don’t you agree that this represents fact, not hyperbole? If everyone listened to you, they’d come away believing every Nehalem-EP system performs “just like the benchmark.” This is unrealistic at best and intentionally disingenuous at worst.

    Your rationale and assumptions must include the Supermicro benchmark as well as the Cisco benchmark or it is insanely flawed. We’ve shown where the economics of today are in Shanghai/Istanbul’s favor – and we’ve pointed out that will not last long. That’s economics, not hyperbole.

    Soon we’ll revisit the tables and see where things stand. All it will take is better than expected performance and pricing from AMD, a rapid increase in DDR3/1066 supply or a change in AMD’s 4P strategy to make a huge shift in market (4P still has power savings to consider).

    Like


  10. Of course all Nehalem systems don’t perform like the benchmarks, just like all Shanghai systems don’t perform like the benchmarks. What you DON’T do is compare optimum Shanghai results to sub-optimum Nehalem results and claim that’s a fair comparison. It’s just plain dishonest of you to do that and it makes your information suspect.

    What you have shown is that AMD might be able to justify a slightly higher price for their CPUs because they use slower cheaper DDR2 memory. What you have not shown is how they are competitive with Nehalem.

    Right now, we’re seeing 2P Nehalem blow out more expensive 4P Shanghai systems across the board. That’s 2 Intel CPUs against 4 AMD CPUs. We’re seeing this on:

    SPECint_rate2006
    SPECfp_rate2006
    SPECjbb_2005
    SPECweb_2005
    SAP
    VMware VMmark
    and probably a bunch of other server benchmarks

    That’s just the sad but true facts today.

    Like


    • What has been done is to compare the only known ESX 3.5 result for Nehalem-EP against Shanghai 2P in ESX 3.5. How that is “super vs. sub-optimal” is illusive to me. In fact, it is more “real world” in application, and again supports our judgment that optimizations in vSphere – not available in ESX 3.x – contribute to Nehalem’s success. We are NOT drawing the conclusion that these same optimizations will similarly benefit Shanghai/Istanbul, as the Shanghai benchmarks seem to indicate otherwise.

      Since vSphere is not available to the public, the “forward looking” results have value, but are not purchasing decisions that would affect implementations right now. By the time vSphere is out, Istanbul will be too, changing the equation yet again.

      This is a virtualization blog – not a general computing blog – and we emphasize issues that are directly related to virtualization. It’s hard to recommend a system switch on the basis of 2 additional VMmark tiles in ESX 3.x (available today.) Likewise, since there is sufficient time before its availability, Istanbul’s value proposition is an important and timely one.

      Like


  11. Oh, so it’s OK to take the absolute best world record result for and AMD Shanghai and compare it to a Nehalem result that’s about 40% lower than the best result and call that “real world”. I guess that’s what you would call fanboy logic.

    Judging by the facts we have today, Istanbul 6-core isn’t going to be taking the lead against quad-core Nehalem-EP much less 8-core Nehalem-EX.

    Like


  12. What’s been done is to compare the TOP Shanghai and TOP Nehalem VMmark scores in 2P/ESX 3.x. That’s not “fanboy” logic, just logic, but clearly upsetting to some. I do not know why more “optimal” results are not listed for Nehalem, but that is not SOLORI’s doing. Perhaps someone in the Intel camp would push for more ESX 3.5 VMmarks to round-out the results?

    As for your last comment, I’ll let you read our conclusions to see if that’s where we stand: judging from your comments I’m sure you’ll be surprised. Again, from a “real world” virtualization-centered approach…

    As for 8-core Nehalem-EX, that will be against Magny-Cours, expanded HT3, 4-channel DDR3, AMD-i, etc. and is much too far in the future to predict. I suspect the results will converge fairly close with some pros and cons on either side.

    Like


  13. If you’re trying to suggest that the 4.x scores don’t reflect 3.x, then that’s a flawed argument. If the 3.x scores for the Nehalem system had the same memory count as the 4.x scores and still suffered a huge performance deficit, you might have a point.

    But I did in fact compare apples to apples with ESX 4.0. In fact, I compared a 4P Shanghai 3.1 GHz with 16 physical cores to a 2P Nehalem 2.93 with 8 physical cores running VMware ESX 4.0. Here are the links again.

    2-Processor Cisco system based on a 2.93 GHz Nehalem-EP with a score of 24.14@17tiles running ESX 4.0
    http://www.vmware.com/files/pdf/vmmark/VMmark-Cisco-2009-04-21-B200M1.pdf

    4-Processor HP system based on 3.1 GHz Shanghai system scoring 22.11@15 tiles running VMware ESX 4.0.
    http://www.vmware.com/files/pdf/vmmark/VMmark-HP-2009-04-24-DL585G5.pdf

    Now if someone is buying new servers today (which is what we’re talking about), and even if they are running ESX 3.x TODAY, you better believe that people are going to care about ESX 4.x performance. And generally, new servers are the perfect time to change the software too. The fact of the matter is, most vendors will focus on benchmarking 4.x for their AMD or Intel based systems and that’s a good thing.

    Like


  14. Gotcha! At 72GB (4GBx18 DIMM) the systems price-performance is pretty much the same! Now you’re back to the systems with $14,000-21,000 worth of memory, and Shanghai price/performance beats the crap out of Nehalem at 64,80 and 96GB!

    You need to get the extra memory to get the extra tiles, and that BLOWS-UP the price per VM… The 72GB tile score is the only one that applies in a reasonable $/VM ratio and any extra consolidation capability Nehalem-EP may have is priced out of the equation!

    Now, try to argue that a 1P Nehalem-EP will beat a 2P Shanghai in similar tests and you’ll ALSO be hung by the low memory constraint (or high cost) of the platform. Likewise, you’ll rob yourself of 1/2 of the potential memory bandwidth the 2P can offer.

    So again, my price-performance figures are accurate given the memory configurations compared. These configurations are chosen based on $/VM (as stated in my research) and provide an appropriate comparison of the products given the available resources and bound by economics.

    Now, unless you are back to advocating spending $14-20K for memory just to have a Nehalem-EP system, we’re back to my original argument: you can’t REALIZE Nehalem’s potential without VERY expensive memory configurations. Without VERY expensive memory configuration, Nehalem only represents a 25% bump in capacity-per-core: again, modeled by my economic comparisons.

    Just to be fair, I will post a VMmark comparison chart that details the “cost of performance” to make it simple for everyone…

    Like


  15. NOTE: To get the most out of Shanghai configurations at 64GB, 80GB and 96GB, you’ll need a 4P system to enable DDR2/800 and the additional memory bandwidth. Nehalem’s memory bandwidth is potentially up to 45% greater than 2P Shanghai due to Nehalem’s DDR3/1333 and 6.4GT QPI vs Shanghai’s 4.4 GT bus and DDR2/800 limit.

    Like


  16. Like I said, it’s extremely expensive to buy 8 GB DIMMs on either the AMD platform or Intel platform. But if you did spend $18K on 144 GB of memory for a high-end Nehalem-EP 2P system, the whole system would still cost you less money than a 4P Shanghai system that probably sucks about 4x more electricity. Then there’s that little issue of per-socket licensing where socket licenses cost $10K to $100K.

    If we’re talking about cheaper 72 GB configurations, then you could actually just use a 2.26 GHz Nehalem-EP because a 2.93 GHz Nehalem-EP would be under-utilized due to the memory limitations. Such a system would beat a 3.1 Ghz Shanghai system with comparable memory and the Intel system would cost less.

    Now this is really application dependent and you have to look at what ratio of CPU performance to memory capacity your application requires. Some applications might favor a weaker CPU and more memory in which case you might get a good value out of Shanghai. Some apps might not need as much memory and require more CPU. But again, AMD’s DDR2 advantage doesn’t really help much you in the case of 8GB DIMMs.

    Like


  17. LOL at your last comment. You do understand that a 4P platform costs about $15K more right not to mention insane licensing costs. Right now, I can’t really see any advantage to 4P AMD systems. Their only play right now is in the 2P value segment.

    Like


  18. George regarding your statement here:
    “I do find it laughable that you have to rely on comparisons to Netburst based Xeon from 3 years ago to make your point. I’m not sure how that’s remotely relevant for someone buying a server today.”

    If you find Collin’s point laughable then you should call out Intel for their ridiculous ROI claims they made when launching Nehalem about consolidating 9 four-year old, single-core Netburst servers into one Nehalem server. Intel is obviously “relying” on Netburst comparisons still. How is that “remotely relevant” for someone buying a server today? http://blogs.intel.com/technology/tag/roi

    Like


  19. […] Shanghai Economics 101 – Continued […]

    Like


  20. Note: We have indicated numerous times the exact differential in 8GB/DDR3-1066 DIMMs ($1,160/ea.) and 8GB/DDR2-667 DIMMs ($500/ea). Clearly, this represents the most significantly differentiated piece of the CAPEX equation for equipment given similar prices in chasis, mobo’s and processors.

    On the ROI side, faster = better within the same OPEX constraints. Again, we see no significant power savings for benchmarked systems given their build-outs and application. Catch-Twenty-two? Maybe, but economics is what we’re discussing, and the relative cost of compute reasources is a significant factor in purchasing decisions.

    Now, with the price of memory poised to rise by 15% in the short term:

    http://www.digitimes.com/news/a20090505PD218.html

    http://press.dramexchange.com/en/node/213

    http://www.cpu3d.com/news/7646-1/dram-spot-prices-up-significantly/story.html

    http://www.digitimes.com/news/a20090508PD224.html

    According to these sources, it seems that when short-term supplies are exhausted, the vender prices will climb – not drop. This could indicate that a pivotal quarter lies ahead and DDR2 is on the value side of the adjustment…

    Will OEM stockpiles be sacrificed to sell systems, or will OEM’s profit-take from their newly-found position of advantage? Short-term: system sales will prevail, making the larger OEM’s more competitive agaings the smaller ones. Meanwhile, Istanbul will launch on DDR2 platforms and Shanghai’s current value proposition will likely hold.

    Like



Comments are closed.

%d bloggers like this: