GPU Performance

The 4th generation iPad integrates a quad-core PowerVR SGX 554 (MP4). The 554MP4 doubles USSE2 count over the previous generation PowerVR SGX 543MP4 used in the iPad 3, while keeping ROP and TMU counts the same. The result is a pure doubling of peak theoretical shader performance:

Mobile SoC GPU Comparison
  PowerVR SGX 543 PowerVR SGX 543MP2 PowerVR SGX 543MP3 PowerVR SGX 543MP4 PowerVR SGX 554 PowerVR SGX 554MP2 PowerVR SGX 554MP4
Used In - iPad 2/mini iPhone 5 iPad 3 - - iPad 4
SIMD Name USSE2 USSE2 USSE2 USSE2 USSE2 USSE2 USSE2
# of SIMDs 4 8 12 16 8 16 32
MADs per SIMD 4 4 4 4 4 4 4
Total MADs 16 32 48 64 32 64 128
GFLOPS @ 300MHz 9.6 GFLOPS 19.2 GFLOPS 28.8 GFLOPS 38.4 GFLOPS 19.2 GFLOPS 38.4 GFLOPS 76.8 GFLOPS

The theoretical numbers validate Apple's "2x faster GPU" claims, but as always we'll turn to Kishonti's GLBenchmark to see how achievable that performance increase is.

We'll start out with the raw theoretical numbers beginning with fill rate:

GLBenchmark 2.5 - Fill Test

GLBenchmark 2.5 - Fill Test (Offscreen 1080p)

The peak fill rate test shows a ~16% increase in performance over the previous generation 543MP4. Since there's no increase in number of TMUs we're seeing the results of a higher clocked GPU in the iPad 4's A6X.

GLBenchmark 2.5 - Triangle Texture Test

GLBenchmark 2.5 - Triangle Texture Test (Offscreen 1080p)

There's a pretty hefty improvement in triangle throughput - we're seeing more than a 60% gain compared to the iPad 3.

GLBenchmark 2.5 - Triangle Texture Test - Fragment Lit

At native resolution the fragment lit triangle texture test shows a big gain over the iPad 3 (~80%).

GLBenchmark 2.5 - Triangle Texture Test - Fragment Lit (Offscreen 1080p)

In both of the final triangle throughput tests the iPad 4 manages a 40 - 45% increase in performance over the iPad 3:

GLBenchmark 2.5 - Triangle Texture Test - Vertex Lit

GLBenchmark 2.5 - Triangle Texture Test - Vertex Lit (Offscreen 1080p)

With the synthetics out of the way, we can look at simulated game performance using the Egypt HD and Egypt Classic benchmarks. Remember the on-screen tests are run at native resolution with v-sync enabled, while the offscreen tests are run at 1080p with v-sync disabled for an architectural apples-to-apples comparison.

GLBenchmark 2.5 - Egypt HD

Despite sub-2x gains in a lot of the synthetic tests, Egypt HD shows us what's possible in a simulated game: the new iPad is roughly twice the speed of the previous gen model when running at the panel's native resolution. How we've seen this implemented in many cases is with titles finally running at native resolution on the iPad 4 vs. some lower, scaled resolution on the iPad 3.

GLBenchmark 2.5 - Egypt HD (Offscreen 1080p)

The Egypt Classic test is a much lighter workload, as a result most of these devices hit 60 fps at their native resolution:

GLBenchmark 2.5 - Egypt Classic

Although Egypt HD is a bit overkill for today's games, Classic undershoots by a good amount. The offscreen test however does provide some guidance as to whether or not these devices would be able to hit 30 fps on an appreciably heavier workload:

GLBenchmark 2.5 - Egypt Classic (Offscreen 1080p)

CPU Performance & Memory Bandwidth PowerVR SGX 554MP4 and iPad Retina Display: A Balanced Platform
POST A COMMENT

113 Comments

View All Comments

  • prdola0 - Tuesday, December 11, 2012 - link

    http://www.anandtech.com/show/6440/google-nexus-4-...

    Just an example. Looking at many articles back, this bias is clear. It does not matter if you find one exception 6 months old, out of 40 cases. The trend is still there.
    Reply
  • Observerrr - Thursday, January 03, 2013 - link

    iPad 4 scores better than you've posted in the article (as example, 52 fps for Egypt Hd offscreen, 42 for onscreen, 132 for egypt classic offscreen and etc.).
    Please, fix it if you can. A6x gpu is a beast.
    Reply
  • SanX - Sunday, January 06, 2013 - link

    When you're using these tests DIFFERENTIATE what specific subtest gave what. May be just the PDF viewer part was lagging whole the test for one processor and giving super-advantage to another Reply

Log in

Don't have an account? Sign up now