Interesting Memory Blog

[Greybeard]

We have recently published a DRAM blog underlining the benefits of high speed RAM in a system that is based on a Haswell platform. Check out the blog here and our Facebook page here.

"DDR3-1600 is quite simply no longer enough for modern chips outside of Ivy Bridge-E and Vishera. "

[kalniel]

Thanks Greybeard, but can you please refrain from using non-zero axis graphs, it really exaggerates the significance of a few frames per second difference if you're only zooming in to the 70+ fps range, and gives the impression you're trying to mislead rather than present an informative article.

(I don't think you are trying to mislead at all, but non-zero axis is exactly the sort of thing we see from marketers who are)

[Agent]

It's nice to see the video encoding example, but is there any reason you've gone with a multipass route?
Most people I know seem to go for a target quality.

[herulach]

It'd also be interesting to see how this scales with memory size. If I understand it correctly windows 8 is pretty good at using ram to cache things. A test run with 8 or 16 gb might be a better indicator of how much people can expect to see a difference in real world use. 32GB of 2400 is going to run you as much as a second high end GPU.

[Greybeard]

Thanks Greybeard, but can you please refrain from using non-zero axis graphs, it really exaggerates the significance of a few frames per second difference if you're only zooming in to the 70+ fps range, and gives the impression you're trying to mislead rather than present an informative article.

(I don't think you are trying to mislead at all, but non-zero axis is exactly the sort of thing we see from marketers who are)

Non-zero axis graphs just seemed like a better way to present differences. We're unfortunately just not clever enough to be wantonly misleading; with some scales (like memory bandwidth) the two would be on top of each other even with practical differences in place.

[Greybeard]

It's nice to see the video encoding example, but is there any reason you've gone with a multipass route?
Most people I know seem to go for a target quality.

The x264 video encoding example is a synthetic benchmark. Practical, single pass x264 encoding would be the Adobe Media Encoder CC – AVCHD test, which was set

It'd also be interesting to see how this scales with memory size. If I understand it correctly windows 8 is pretty good at using ram to cache things. A test run with 8 or 16 gb might be a better indicator of how much people can expect to see a difference in real world use. 32GB of 2400 is going to run you as much as a second high end GPU.

Memory size…is a very good idea. Something that we look forward testing in the future. Once we get a Haswell testbed in the office up and running we may just start playing with memory capacity as well.