4 ms·
More ram = more performance if there's a cost in going across a bridge to keep the pipeline full. 16gb of _unified_ RAM that can be accessed by the CPU and GPU
by Damogran6 4y ago
More ram = more performance if there's a cost in going across a bridge to keep the pipeline full. 16gb of _unified_ RAM that can be accessed by the CPU and GPU without a performance hit in going to the motherboard is a factor in some workloads. Whether you agree or not, it's an effective test comparing two architectures.
So the question is: are you spec benchmarking, or are you interested in the performance of the workload, because it's the actual work that matters, not how the bits are stored under the keyboard.
- gbil 4y agoI'm interested on not swapping which is a simple matter of RAM size, regardless if that is unified or not - which I believe I touched in my post
- Damogran6 4y agoOnly if your datasets, in aggregate, extend beyond what can be compressed into 16Gb...an what's the performance of the device you're swapping to? There's a GREAT BIG DIFFERENCE between an RLL 80 Gb drive and What you're swapping to on the motherboard today... (In fact, you could consider the on-die RAM to be a really big Cache and the Flash storage to be REALLY REALLY big Non volatile addressable RAM.)