AMD CES 2020: 4000 Mobile CPUs

2019 was a huge year for AMD (in partnerships and products) and a year of firsts (First 7nm Consumer GPUs, 7nm CPUs, 12 & 16 Core CPUs on a mainstream platform), and now they are promising a bigger year in 2020.

So here we are, barely into this new year and AMD has lost no time in continuing their all out assault on the x86 world. This time with an extremely compelling laptop computing solution in the form of their new 4000 series mobile CPUs.

Now typically these presentations are a lot of talk and big promises, accompanied with nice graphs (of often hand-picked comparisons) to show these products in the best light possible… and to be honest, this keynote was no exception. 

AMD Breaches Intel's last Consumer Stronghold

The realm of laptops really is the last corner of the consumer market AMD has yet to break into. They kind of existed there, but only barely in recent years, and I believe this will be the toughest for them to take, especially since their 7nm processors are going up against actual new technology and not some 14nm+++++++++++++++  triple refresh Skylake CPUs. When it we're looking at laptops, it all comes down to performance per watt, and AMD has come really come through here. Their 4000 series CPUs offer up to 2x performance per watt compared to the previous generation (4800U vs 3700U in select benchmarks). This doubled performance brings about another world first with AMD's 4800U being the first 15W 8C/16T CPU ever.

Product Stack

They’ve broken down their 4000 Series CPUs into 3 categories.

U-Series

4000U chart

H-Series

4000H chart

Pro-Series

There has been no information on the Pro-Series other than some buzzwords during the presentation and the stock photo above with 2 people looking at what is probably a computer.

What Does it all Mean?

Looking at the above slides there are some interesting things to note. 

Vega cores are being used in the integrated GPUs, not Navi as one would expect. However in the keynote it was noted that these Vega GPUs have faced comprehensive refining to bring them to 59% improvement per compute unit. So when you take into account that there are now 8 compute units vs the 10 before, and assuming the 59% improvement is raw performance per compute unit, the iGPU performance of the new 4800U is 27.2% better than the previous generation. Kind of still wondering why not Navi, but it seems AMD was able to do pretty well with the 4800U’s Vega iGPU especially considering it’s on a 15W product. 

As you can see the integrated GPU on the H series are less powerful than their U series counterparts even with the increased TDP, meaning the focus for the high performance units is on the CPU portion to get that 2.9 GHz base clock (which is pretty incredible for an 8C/16T 45W CPU), as they are no doubt intended to be used with discrete GPUs (like the 5600m). 

Additionally (not in the slides) the new 4000 series CPUs only support PCIe3.0, so no crazy fast PCIe4.0 NVME drives in those laptops.

Why I Hate Keynote Performance Graphs

You may have noted that this article is devoid of competitor performance comparison graphs.

Here is where I may generate some disagreement, because I don’t trust any keynote performance comparison slides. Especially when it comes to laptop CPU performance. This is regardless of what company is posting them.  Unless they are both done within the same laptop chassis, which none of these comparisons have been the comparison is automatically skewed. All the 4800U and 4800H benchmarks have been done in an “AMD reference system” while the Intel ones are in a production laptop. This means the cooling solutions are different, the thermal properties of the laptop chassis are different, the power envelopes are possibly even different. I simply don’t trust the numbers.

Base vs Boost

Additionally, boost clocks to me are meaningless (in a laptop system) unless your workload and/or gaming session is only lasting 1 minute or less. This is because thermals in laptop designs, ESPECIALLY ultrathins are typically quite bad, and can kill your boost clock within seconds of hitting it.  In a desktop, you can always upgrade the cooler to ensure boost clocks are maintained(and/or overclock), and as such they hold some relevance there.

For me, I look at base clocks as the only performance guarantee when it comes to laptops. Especially for any workload that takes more than 60 seconds. For example, the 4800U will boost to the same peak frequency as the 4800H, but it’s really that 1.1GHz (1.8-2.9GHz) advantage on the base clock which will make the biggest difference for extended usage and why it's considered the high performance CPU of the two.

That’s why I don’t really take the boost clock numbers too seriously, unless you’re planning on getting a laptop thicker than 3” or with hookups to an external watercooling system.

This is also why I would expect in an extended workload (>30 minutes) the 4800H would probably outperform even the i9-9980HK even-though it (the i9) has a 800MHz (5.0Ghz vs 4.2GHz) advantage with its boost frequency. This is based on the assumption that they have roughly the same IPC and the 4800H has a 500MHz (2.9GHz vs 2.4GHz) advantage on the base clock.

Really though the only way to know for certain with any performance comparison is to take 2 identical chassis that use the 4800H/U and their Intel competition and put them head to head to get true apples to apples performance graphs.

One More Thing...

And speaking of Apples, I have a hunch, that SmartShift (shown above) coupled with these new processors are going to be AMD’s big selling points that finally convince Apple to release Macbooks, possibly even this year, with AMD CPUs.

SmartShift
SmartShiftgains

Basically this technology allows for an AMD CPU and discrete AMD GPU systems to utilize the laptop's power envelope in the most effective way possible depending on the demands of a given task. So rather than having a fixed power budget for the CPU and GPU as it is in many laptops today, this will allow for power to be sent to where it is needed most.

The "shift" is intended to be nearly instantaneous, and provide a continuously variable amount of power to where it is needed (GPU in graphics intensive workloads and CPU in CPU intensive workloads). This technology (according to the slide above) should provide some noticeable performance improvements in use cases where there is a heavy CPU OR GPU load.

So what do you think? Will we see AMD in Macbooks? Will AMD see similar dominance in the mobile market as we saw in desktop CPU sales in 2H 2019?

It's looking like 2020 is shaping up to be a very interesting year for PC tech, and we're excited to continue exploring it with you all. Until next time, PC Better.

Watch the full Keynote here: