Data Analysis

So we’ve taken a bunch of data on both average FPS and 99% frametimes on a few different games.  Let’s take a look now on how the R9-295×2 does versus the Fury X.  We would expect the R9-295×2 to be somewhere around 150% of a Fury X.  Let’s first look at the difference in average FPS comparing the R9-295×2 as a percentage of the Fury X’s performance:

A higher result means the R9-295×2 did better than the Fury X.  On average the R9-295×2 gives 28% more FPS than the Fury X.  This isn’t as high as our 150% expectation – although a few games in a few modes are getting up there.  It seems there are a few games that could use some crossfire tweaks.

If we look at 99% frame times however the picture as usual looks worse.  Despite frame pacing improving AMD’s performance it is still not working as well as it should:

A higher result means the R9-295×2 did better than the Fury X.  Here we see results all over the map.  The average rolls in at only 18% better frametimes for the R9-295×2 than the Fury X, and in three of our test cases the 295×2 was 20% worse than the Fury X.

So then we can say that despite Crossfire’s inefficiencies and it’s driver problems that overall the 295×2 still performs better on both critical metrics.

But let’s also take a look at VRAM.  AMD made a big deal of HBM and we’d like to investigate some of those claims to see if HBM really is everything that was claimed.

Claim #1 – You don’t need as much HBM as you do GDDR5

Let’s take a look at video memory used – now bear in mind that how much is used is not always the same as how much is needed.  Sometimes drivers just don’t take stuff out of memory unless they need it.  However it should give some indication as to the truth of the matter:

Consistently time and time again the R9-295×2 uses slightly less memory than the Fury X.  There is absolutely no indication that the Fury X HBM has super powers that GDDR doesn’t have.  4GB is 4GB – don’t be fooled by marketing.

Now the good news is that very few of the games really seem to be really running out of memory.  Only Thief at 4K really seems to be limited.  So let’s use that to verify the next claim.

Claim #2 – HBM’s faster bandwidth will not suffer from VRAM induced stutter

So let’s take a look at the Thief 4K frametimes again as that game had by far the highest VRAM usage:

The R9-295×2 is noticeably good even at high percentil frame rates.  In fact neither card seems to be stuttering at all.  This is quite bizarre.  In fact the only time we really see this claim played out is in GTA V:

However while GTA’s VRAM usage is high – it doesn’t look high enough to be causing stuttering due to VRAM.

Given that we only have one case which definitely is VRAM limited and that this case doesn’t show any improvement for the Fury then we have to conclude that this claim seems false also.