Pretty sure that is not true. See
@seronx post #16. The same "not tied" messaging has been used throughout AMD statements for Zen2. What AMD means by that phrasing is that IF frequency is not 1:1 to memory speed as it was for Zen/Zen+. They added a multiplier/divider and use that to facilitate faster memory speed.
I sincerely doubt that. AnandTech specifically states that
AnandTech said:
AMD says that the infinity fabric is not tied to this memory clock.
"Not tied to" is something quite different than "
tied to, but variable based on a clock divider". "Not tied to" means
no direct relation between. Otherwise AT is either misquoting AMD, or AMD is being more shady than usual with their wording. Neither sound likely. Also,
@seronx's numbers are a bit off. How does 4266 with a 2:1 divider become 1600? Last I checked 4266:2 was 2133. so you'd need a 3:1 divider to keep IF clocks in a normally stable range with 4266MT/s RAM.
My biggest worry is that too much of the power and silicon budget has been moved away from the graphics cores and to the CPU cores. Both previous generations of APU have been perfectly adequate in the CPU department and sorely lacking in the IGP department, ranging from inadequate (Vega3 being as useless as low-end Intel HD) to acceptable-but-underwhelming - in that 720p30 might just be attainable in current titles.
In terms of CPU performance, 15W ultraportable customers really aren't clamouring for more cores. Those products are typically not multitasking mobile workstation powerhouses, lacking RAM, screen size, and storage for many serious workloads. To date, the vast majority of 3700U flagship APUs have been either 8GB or 16GB max, with 512GB NVMe 2x and 1080p displays. They are general-purpose consumption devices that could definitely use a little more GPU power but are typically beyond the point of diminishing returns in the CPU department already.
AMD keep saying that they have added 60% more performance per CU but then have stripped away 30-40% of the CUs The fact that AMD isn't singing praises about the 3D performance of its new 7nm APUs is a pretty bad sign of things to come, especially because the 4800U is likely to command a significant price premium. I would expect the 6CU option to sell at the same pricing tier as the previous 10CU 3700U models, which sucks because the 4800U will likely be price-competing against an MX250 or even faster dGPU options. I can only hope that LPDDR4X is used to its full advantage this generation. My 2700U shipped with a 13W TDP and single-channel DDR4 2400; The only reason I bought it was good cooling (for up to 25W) and an empty DIMM slot for dual-channel.
Is it too much to ask for a 15W ultraportable that provides a meaningful IGP upgrade? The bar is SO low, I'm honestly saddened by the lack of attempts to pass it.
I'm not worried at all about this. One of AMD's biggest points when presenting these is their new system for balancing power distribution between the CPU and iGPU (as well as any AMD dGPU installed in the system) using what they call "Infinity control fabric" (yeah, not very imaginative
). They're claiming significant performance increases due to this as power is supposed to be dynamically divided based on actual demands in real time rather than preset power levels. While this is mainly marketed for H-series chips, the silicon is the same, and the main difference between H and U in this regard is just that H-series chips will mostly have dGPUs attached and thus benefit more from the system also encompassing that. (Also, the lack of dGPU means thin-and-lights don't need the SmartShift setup to dynamically manage CPU/iGPU power.) They're still touting significant increases in mixed workloads like gaming despite there now being 8 CPU cores. Given the work AMD has put into improving the Windows scheduler for optimal function with desktop Ryzen I'm guessing they've brought that work with them so that the system keeps as few cores active as possible while gaming (and in similar loads) so that they can clock high while the rest sleep, freeing power for the GPU. If games suddenly start hammering 10-16 threads I guess the CPU might suffocate the GPU performance, but then you're talking about a game outside the capabilities of a system like this anyhow.
Given advertised performance levels (and leaked benchmarks) the perf/CU increase looks about right, meaning that we'll be getting overall gaming performance increases for all AMD APUs compared to previous generations. LPDDR4X is also going to be used in
a lot of designs, especially thin-and-lights.
Still, the next generation, with RDNA on board, is likely to kick this one's butt. That's a given, especially if it's on 7nm+. But we'll still be getting a decent enough performance increase this go around.
Nonetheless, the 4800U is looking like it'll be both a beast for heavy CPU tasks like video editing or rendering for its class of laptop, while also being the 15W king of gaming. If it indeed delivers ~28% more performance than an Ice Lake 1056G7 like they say (normally I'd be wary of pointing to a canned benchmark, but in this case that's one of Intel's best-case scenarios in terms of optimization), that's a notable jump.
According to NotebookCheck the 1065G7 on average scores ~967 in 3DMark Time Spy, so a 28% increase over that means ~1230 points.
The 3700U scores about the same as the 1065G7.
Although considering what's happening in the APU's lately and what's to come this idea 1080p AAA games High/ultra, doesn't seem so crazy now.
