AMD R7 (A10-7850K) vs FirePro v4900

Recently, Unshockable Engineering upgraded a workstation in our Victoria, BC office. The old Dell had an issue and wouldn’t work anymore.  Luckily, no data was lost.  We went with an AMD system with built-in graphics. The long story short, this was a mistake. Even a small graphics card simply slaughters the performance of the R7 APU.

 

The R7 APU crashes when doing even mild 3D tasks — a backtrace on a PCB can’t be rotated without causing a crash. This performance was obviously UNSAT, so a new graphics card was ordered immediately.  A FirePro v4900 was selected simply because it was about $200 on sale at newegg, had decent specs including a gig of DDR5, and didn’t require extra power hookups.


We realize that this test would have been better with 4-5 cards.  4-5 cards aren’t sent for free to engineering companies. The work we do is almost all electronics so the 3D is minimal, with a little board previewing in KiCAD.  We aren’t porting our code to Rhino to get a few thous differential on a G-code run.  (The video card for that was about $1500 a few years ago.  The V4900 dwarfs it now. )


Now, to check improvement we had to first measure.  For that, we had to get some free benchmarking tools.  3DMark is $1500 for commercial use, and that would buy a VERY nice video card.  All respect to Tom’s review of video cards, but we’re not testing multi-thousand-dollar cards here.  If your corporation is picking up the tab for a $2000+ card, go for it.


The first tool was Novabench.  It was rated as light, fast, and free.  We had some doubts about grabbing a couple of “free Internet progams!!” especially when they’ve got Russian mirrors, but hey, beggars can’t be choosers… and we use a robust real-time backup system.  The first results were all right. The workstation got a Novabench score of 792 overall, which is lighter than the 3 month average of 913. I’m unclear as to why the drive score is so low – this workstation uses a SATAIII interface and SS drives, so 113MB/s is quite poor.  This may be a Windows 8 issue, but it’s not pertinent to the graphics card issue.

Novabench before
NovaBench score of 792 with built-in R7 graphics.

The test setup is the same for all the tests. 8G of GSKill RAM, an A10-7850 processor, and a SSD running SATA III. It’s not the biggest, gruntiest computer you’ll see, but it’s a good enough machine with a decent set of results. It’s an engineering workstation and won’t be used for gaming.  That’s an important distinction.  Gaming machines… you know what, if you’re reading this, you already know what the difference is.  SIS told a very similar story in terms of the benchmarking, but it would not complete the tests properly with the V4900.


AMD was chosen for two reasons:  Primarily, having a second architecture to check values against. Ideally, engineering work gets double checked with a separate architecture. (It’s a long story.)  There’s already a couple of Intel machines here, and having an AMD machine will make everything just a little safer.   Nevertheless, the point is that the machine is unchanged except for the video card.


After upgrading the machine and re-running the tests, the FirePro makes a noticable difference. The new score is 918, just at the 3-month average, which for a workstation computer running up against high-FPS gaming machines is perfectly acceptable.

Novabench after
NovaBench score of 918 with dedicated V4900 graphics

 

More RAM is freed up for general use (That 1G is no longer “shared” with the APU), the overall score is improved, and you get the ability to rotate multi-layered PCBs.  Now, this was expensive 2133MHz RAM to keep that APU happy, but it wasn’t even nearly enough for the type of minimal 3D work we’re doing.


So the question is: who is the R7 is for?  The extra cost of the chip is handily destroyed by the same price point in a low-end dedicated card.  Any gamer is going to buy a big honking gaming card, and workstations can’t use it for even the simplest tasks.  Don’t bother with the Kaveri unless you’ve got something else in mind.  The performance isn’t low — it’s literally not there at all.  For comparison, here’s a year-old Sony Vaio laptop:

personal_laptop
For comparison, here’s a one-year-old Sony Vaio laptop.

As an aside, if you are using a Linux build and upgrade your video card, press ctrl-shift-F1 when you get to the “can’t start X server”, and delete \etc\x11\xorg.conf and reboot.  This will force the OS to rebuild the configuration and let the X server restart.