skip to Main Content

Post by Rob Collie

Good Computers for Power Pivot / Power BIA Lot More Work Than Expected!

Last month I posted a survey of computer performance for Power Pivot and Power BI usage.  I underestimated how much work it would be, to synthesize the results into something useful for the community.  At bare minimum, this has been five different tasks:

  1. Throwing out untrustworthy outlier results
  2. Cross-referencing with CPU benchmark sites
  3. Finding computer models that contain those CPU’s
  4. Verifying that those models have good RAM
  5. Pulling together a view different price points
  6. Repeating for Desktop vs. Laptop

So for now, I’ve only managed to pull together options from HP.
(I will add other manufacturers later, but I have always liked HP hardware, especially their laptops, so it’s a great place to start).

Desktops – High End Options

If you want a true beast of a computer that chews through DAX workloads, you might consider something like THIS monster:

 The Z840 Workstation is an Obvious Candidate.
(But the Next One Below is Better AND Cheaper.)

16GB of RAM is more than enough for Power Pivot and Power BI workloads.  Seriously, 8 GB is gonna be enough for most of your needs until/unless you start transitioning into SSAS Tabular because you have too much data for Power Pivot.

Note that this workstation above offers much faster RAM than most other machines, at 2133 GHz.

But it also carries the ultra-premium Xeon E5 2620 v3 CPU – which the test results and benchmarks indicate is a LESS effective processor (for our purposes) than many lesser models of Xeon, such as the E3 family.

So, for $1300 less, you could have the following workstation, which in all likelihood performs even BETTER for our needs:

The Z230 Workstation:  $1300 LESS than the Z840
(And Still Probably FASTER for Our Needs.  I Now Want One.)

Desktops – Midrange/Value

In desktop hardware, I discovered that there isn’t much difference between “value” and “midrange.”  You can get a VERY fast and capable computer in the $1000-$1200 range, but once you start trying to shave cost after that, you quickly start losing lots of functionality.

Similarly, when you look in the $1500 range, you don’t get much more than you do in the $1000-$1200 range.  You have to “jump” into that $2k range before you start to see a big difference.

So it doesn’t make too much sense to go above OR below this $1k price band, unless you are prepared to go all the way to $2k.

With that in mind, here are two good options:

The ENVY Phoenix 810-460:  16 GB RAM and a Great CPU for $1200
(Very Good Price/Power Ratio)

The ENVY Phoenix 810-470:  12 GB RAM and a Slightly-Lesser CPU for $1000
(For Perspective, its CPU is Still Better than Any of the Machines I Currently Run)

Laptops – High End Options

Laptops are a much more-finicky prospect. More factors to consider, and the CPU market seems even more fragmented than the desktop CPU market.

Let’s start out with a Beast again:

The Zbook 17 G2 – Monster CPU, 16 GB RAM, and a full 1TB of Hard Drive for $2649.
(You Could Run ANYTHING on This Weapon of Mass Calculation,
and It’s Reasonably Priced for Such Specs in a Laptop)

ZBook 15 G2 – Same CPU and Hard Drive, Smaller Screen, Lighter, 8 GB RAM
(And $500 Less.  Better Option For Most Of Us I’d Think)

Laptops – Midrange/SSD

If you want a lighter/slimmer laptop with improved battery life, Solid State Drives (SSD’s) are a great alternative to traditional hard drives.  The tradeoff is that you get less storage space for the dollar with SSD’s, so don’t expect to find affordable laptops with 1 TB SSD’s.  256 GB SSD’s are much more standard, and for perspective, that’s what my laptop has.  It’s been plenty of space.

Elitebook 840 G2 – 8 GB RAM, Slim, Light, 256 GB SSD, Solid CPU for $1749

ENVY 17-k118nr:  Same CPU & SSD, but a MONSTER Screen AND 16 GB RAM, On Sale for $949!
(Hard to Find This Much Laptop at Sub-$1k Prices, Especially if You Like 17″ Screens)

Laptops – Value

Depending on your feedback, I can pull together some options in the $1k range.  But given that the last one above is now in that price range (originally was $400 more), if you don’t mind the size that comes with a 17” screen, that’s the one I’d be looking at first.  Let me know.

Is this Useful?

Let us know if this is helpful!  No sense in us pulling together future iterations of this unless it actually helps.  So leave us a comment please, let us know even something as simple as useful/not useful, or how we should improve it.

Rob Collie

One of the founding engineers behind Power Pivot during his 14-year career at Microsoft, and creator of the world’s first cloud Power Pivot service, Rob is one of the foremost authorities on self-service business intelligence and next-generation spreadsheet technology.

This Post Has 12 Comments
  1. I have a dell (Xeon) that works unbelievely well. My models are huge and I wouldn’t want to change the pc.

  2. Rob,

    I think I would be interested in some basic statistical data, median, mean, std dev for the entire dataset, maybe a stem and leaf. It would be helpful to differentiate the cost to speed ratio in the above. How much speed does the extra $$ potentially buy me? At some point it would be good to also be able to benchmark responses on SP.
    In my company we can only choose between Dell and Apple, which reminds me I was going to run it on a couple of Macs, can I still send you the data if I remember to do it?

  3. I found a significant increase in speed when moving to SSDs on the C drive using Excel 2013. In working with large models, Powerpivot appears to unpack the data into temporary files on that drive – dozens of them – and manipulates them when the model changes. For one large model (~1gb) it took about 5 minutes to calculate after a model structure change, which dropped to about a minute on SSDs.

    1. any idea where this files are unpacked? I have 16GB of RAM and I’m using a RAMdrive for temp (that lowers the wear level of my SSD and is much faster) but it looks Excel/PowerPivot doesn’t use the system’s TEMP folder to do what you mention. Thanks!

  4. I just got the 840 G2 last week as my new work laptop. I can honestly say this thing is a beast! Haven’t done anything too intensive yet but for my normal work functions I am literally twice as productive now from the complete lack of lag in anything I am trying do now.

  5. Second Carl’s comments about a statistical breakdown… and maybe show how performance relates to the two simple variables of amount of memory and CPU model rather than in relation to specific models of computers

  6. Since the formula engine is the typical bottleneck in Tabular/Power Pivot speed, there is no reason to consider a Xeon. The formula engine is single threaded (though this is looking to change in a future release based on what I’m hearing) and an i7 will always outperform a Xeon on single threaded tasks due to a higher base clock speed.

    It seems that Tabular faces a two-part optimization problem – memory bandwidth and single core speed. The newest, fastest i7 will get you single core power.
    Memory is a bit trickier – you will need to minimize CAS latency and maximize frequency. I cannot say whether DDR4 would be worth the expense, but it would be your best bet for fast memory.

    Moving forward a couple years, High Bandwidth Memory – HBM (currently being pioneered by AMD for graphics cards, with Nvidia right behind) – will be the solution of choice. What seems especially promising is something in AMD’s 2-year gameplan – a new APU with HBM on die with the CPU. AMD is pursuing an SMT architecture that should hopefully put it back in the game with Intel on IPC. If they pull that off and put HBM on the same die with it, it should be unbeatable for a compute and memory intensive workload like Tabular.

    1. Totally agree except for one point: Xeons support ECC memory and I have been experiencing some crashes now and then when working without ECC memory 🙁

  7. Just what I was looking for! Our IT group standardized on HP so I’m limited to this selection. (But I’m a building architect; no connection to HP.) Thanks for this info, but it would be great to see a graph or scatter chart of performance for a large dataset on various pieces of hardware. Probably more than you want to volunteer for, but thanks for posting this.

Leave a Comment or Question