So, recently, the new AMD Carrizo chips have become available (scarcely) and I got my hands on the HP Envy 15Z-ah000 with an FX-8800P APU (review here).
Like many others have discovered, the HP Envy 15Z-ah000 FX-8800P Carrizo APU is limited to 15W cTDP, a limitation that was imposed by HP and cannot be removed easily. The FX-8800P cTDP can be configured to a value between 15W and 35W. The reason for this limitation is to allow manufacturers to use the chip in different kinds of machines with different kind of cooling and to allow differentiation, or something. But, the end result for the consumer is that you get a rather low performance Carrizo APU, compared with its full potential, as a budget gaming chip. It is important to understand that HP did not declared the cTDP in their 15Z, they simply marketing it as the FX-8800P. Without talking morality, this is a misleading practice, common in the marketing world.
Anyway, I did some small tests for the FX-8800P, to see how it compares to Intel’s 15W counterparts, the I5-U and I7-U. Also included are result tests of the FX-7600P AMD Kaveri APU. I’ve used notebookcheck’s tests for those (linked to at the bottom). This is a short one, don’t worry.
These are the CPUs I’ve compared. Brief specifications:
|Intel : GPU / CPU||TDP||Core||EU||Dedicated memory|
|HD5000, I7-4650U||15W||Haswell GT3, 22nm||40@200-1100MHZ||No, shared memory|
|HD5500, I5-5200U||Broadwell GT2, 14nm||24@300-950MHZ||No, shared memory|
|HD5200 Iris Pro, I7-4750HQ||Haswell GT3e, 22nm||40@200-1300MHZ||128MB eDRAM|
|HD6000, I7-5250U||Broadwell GT3, 14nm||48@300-1000MHZ||No, shared memory|
|AMD: GPU / CPU||Core||Shaders/CUs||Dedicated memory|
|Radeon R7 Carrizo FX-8800P 15W||15W (there are 35W variants)||Carrizo GCN 1.2, 28nm||512 shaders, 8CUs (guesstimation), core@300-800MHZ||No, shared memory|
|Radeon R7 Kaveri FX-7600P||35W||Kaveri GCN 1.1, 28nm||512 shaders, 8CUs, core@550-680MHZ||No, shared memory|
Let’s start with some 3D performance in games and synthetic benchmarks
What do we see? That the Intel I5-5200U with its HD5500 integrated GPU isn’t much faster than the FX-8800P, or in some cases, not faster at all. Now, remember we are talking here on two 15W chips, head to head and remember that the AMD APU is far more limited/slower clockwise/IPC.
Remember that the CPU, in games, downclocks to around 1.3GHZ and we already know that AMD’s drivers add considerably API overhead, so I wouldn’t be surprised to discover that the GPU performance here is limited also by the CPU, besides the memory limitations and the core downclock.
Some more numbers:
|FX-7600P 35W (Kaveri)||I7-5250U (HD6000)||I7-4750HQ (HD5200, GT3e)||I7-4650U (HD5000)||I5-5200U (HD5500)||FX-8800P 15W, Radeon R7|
|235||146||325||187||170||194||Luxmark 2.0, Room, GPU only|
|31||31||46||27||26||22.5||Cinebench R15, OpenGL|
|223||260||612||256||260||202||Cinebench R15 MT, CPU|
|5700||9900||4600||4800||4200||3DMark 13, Cloud Gate, Standard|
Pretty nice for a 15W AMD chip based on the old Bulldozer architecture.
Throttling. The FX-8800P is throttling under load. The number are around 300-500MHZ for the GPU core and 1.3GHZ for the CPU core with some jumps in clocks, though the GPU clocks usually set around 300-350MHZ stable after a while. Let’s take a look at the CPU and GPU utilization statistics and clocks (HWInfo):
Besides the fact that there is an obvious throttling here, as we already know, it seems that there is a utilization problem of the CPU and GPU. It might be a result of the memory system being overloaded and even some kind of reading errors. Though I didn’t add those, GFXBench numbers behavior were similar and the CPU and GPU were underutilized too. Prime95 and wPrime benchmarks did result in 100% core usage.
And some additional synthetic FX-8800P benchmarks, including the 3DMark API overhead test:
The API overhead numbers are pretty nice too and are much higher than the Intel number for I7-4770R with its Iris Pro integrated GPU (source) at about 50% for DX12 and around 85-90% with Mantle. It also related to what I’ve mentioned earlier – the Radeon R7 integrated GPU is probably limited by the CPU in games. I guess that DirectX12 / Vulkan games, accompanies by a good, less limited AMD APU, will see a big boost, with the right optimized application.
I’ll add battery performance tests from HP Envy 15Z-ah000, for a 48Wh battery. The FX-8800P could easily do 5-6W consumption for very low loads (like reading/writing) and around 7-8.5W for common use like internet browsing, excel sheets and stuff.
The Carrizo FX-8800P 15W is around as fast as the Intel 15W counter parts for the most part and it seems that AMD has optimized it well for this scenario. Keep in mind we are talking a 28nm APU here, as opposed to Intel’s Haswell 22nm and Broadwell 14nm. Carrizo chips also implement fully HSA 1.0 and as more and more applications will start using it well, the Carrizo (and later chips) will be more cost effective.
As my subjective experience, in “power saver” mode, comparing the HP Envy 15Z laptop to my everyday work laptop, Dell e7440 with an I5-4210U, the HP Envy 15Z felt faster, sincerely, and the E7440 comes with a 250GB SSD vs the 750GB 5400RPM basic HDD in the 15Z. So, this chip has potential and so are laptops that are equipped with it, but not as a gaming chip at 15W, at least not under DX11.
I’m not that it is as good for the 25-35W scenario, but I’m pretty sure the integrated GPU could use another 200-300MHZ core clocks and a steady CPU core 2.4GHZ clocks would have helped the performance considerably.
It remains to be seen what the 25-35W Carrizo FX-8800P can do and we’ll see it soon enough with laptops from Lenovo or others, and maybe some mini-PC systems. I guess that this chip could be great for $400 laptops, if set to a higher cTDP. Remember that it is cheaper than Intel counter parts and laptops equipped with it should be cheaper to (though it is not 1:1).
Sources: 1, 2, 3, 4, 5, 6, 7