Saturday, January 26th 2019

Anthem VIP Demo Benchmarked on all GeForce RTX & Vega Cards

Yesterday, EA launched the VIP demo for their highly anticipated title "Anthem". The VIP demo is only accessible to Origin Access subscribers or people who preordered. For the first hours after the demo launch, many players were plagued by servers crashes or "servers are full" messages. Looks like EA didn't anticipate the server load correctly, or the inrush of login attempts revealed a software bug that wasn't apparent with light load.

Things are running much better now, and we had time to run some Anthem benchmarks on a selection of graphics cards, from AMD and NVIDIA. We realized too late that even the Anthem Demo comes with a five activation limit, which gets triggered on every graphics card change. That's why we could only test eight cards so far.. we'll add more when the activations reset.
We benchmarked Anthem at Ultra settings in 1920x1080 (Full HD), 2560x1440 and 3840x2160 (4K). The drivers used were NVIDIA 417.71 WHQL and yesterday's AMD Radeon Adrenalin 19.1.2, which includes performance improvements for Anthem.

At 1080p, it looks like the game is running into a CPU bottleneck with our Core i7-8700K (note how the scores for RTX 2080 and RTX 2080 Ti are very close together). It's also interesting how cards from AMD start out slower at lower resolution, but make up the gap to their NVIDIA counterparts as resolution is increased. It's only at 4K that Vega 64 matches RTX 2060 (something that would be expected for 1080p, when looking at results from recent GPU reviews).

We will add test results for more cards, such as the Radeon RX 570 and GeForce GTX 1060, after our activation limit is reset over the weekend.
Add your own comment

134 Comments on Anthem VIP Demo Benchmarked on all GeForce RTX & Vega Cards

#126
bajs11
At 1080p, it looks like the game is running into a CPU bottleneck with our Core i7-8700K (note how the scores for RTX 2080 and RTX 2080 Ti are very close together).
is 144+fps really needed for playing rpg?
and those who have gtx 1080Ti and RTX 2080 or better should really play games at higher resolutions than 1080p anyway
unless they are making a living off CS

As I keep telling people in Toms forums we still don't need super expensive cpus like i9 9900K or even i7 8700K/9700K to play the latest triple A games
just save your money for a good high res monitor and a good GPU, unless of course you want to play games "professionally"
Posted on Reply
#127
rvalencia
efikkan
What is your point? RTX 2070 is far beyond Vega 64, with nearly comparable fillrate.
Your claim is still incorrect, Vega 64 is not ROP limited, nor is it bandwidth or Gflop limited.

GCN's problem is saturation of resources, which is why it struggles more on lower resolutions than higher.


Don't mix in compression, that's not relevant for this.
Wrong, If AMD is promoting async compute which uses TMU as read/write units, then there is ROPS bound issues.

efikkan
GCN's problem is saturation of resources, which is why it struggles more on lower resolutions than higher.
This is not correct, lower resolution with high frame rates requires lowest latency and highest narrow parallelism performance .

NVIDIA's higher clock speed leads to lower latency with the workload loop


At higher resolution, there's a higher chance for very wide parallelism workloads which has higher benefits for AMD's lower clock speed GCNs.


Zubasa
GCN so far are Geometry bound, GCN has been stuck on 4 Geometry Engines since Hawaii / 290X.
Vega was suppose to have new features to by-pass that limit, but RTG never managed to get it to work, so it ended up just being a higher clocked Fiji.
One of the reasons why Polaris performs as well as Hawaii with half the ROP etc, it is still limited to 4 Geometry Engines (other optimizations aside), which is much less of an issue on a mid-range GPU vs a high-end.
Raja Koduri joined AMD in 2013 and under Koduri administration, RTG has focused on TFLOPS increase without increasing quad raster engine unit count.

Hawaii doesn't have Polaris delta color compression improvements.
Hawaii doesn't have Polaris 2MB L2 cache improvements. Hawaii has 1MB L2 cache for it's TMUs.

It's most likely VII is stuck in quad raster engine unit count.
Posted on Reply
#128
psyph3r
efikkan
Well, that's always the thing with AMD's GPUs isn't it? We're always waiting for that big optimization to arrive and finally unleash the performance. I remember it was claimed at the launch of both Polaris and Vega; don't judge it yet - it will become much better over time!

Don't get me wrong, AMD should certainly prioritize making the best drivers possible, and there might be some marginal gains to be had here. But their focus on "optimizing" is by manipulating the games (which Nvidia also does), and I'm sure that if they reallocated these resources on actual driver optimizations, they could get some gains.

But still, any driver optimization to unlock a massive gain is highly unlikely.
Massive gains over time are a regular occurrence for amd cards. Just look at where all the amd cards sat a year ago vs now.
Posted on Reply
#129
Frutika007
psyph3r
Massive gains over time are a regular occurrence for amd cards. Just look at where all the amd cards sat a year ago vs now.
They are still sitting there as they were a year ago. No performance gain over time.
Posted on Reply
#130
efikkan
rvalencia
Wrong, If AMD is promoting async compute which uses TMU as read/write units, then there is ROPS bound issues.
Not at all. The reason why GCN get larger gains with some async compute workloads is underutilized resources. Nvidia is much better at saturating their GPU cores, and therefore the gain is much smaller.
Posted on Reply
#132
efikkan
Wake me up when AMD releases a driver which gives 15% across the board. "Up to 7%" in one game is not going to cut it.
Posted on Reply
#134
johnnyfiive
EarthDog
O.H........


Hai neighbor (Cbus)!!
I.O!!!
Posted on Reply
Add your own comment