Some time ago Objective Analysis ran nearly 300 standard benchmarks on a PC with varying amounts of flash and DRAM and found that a dollar’s worth of flash provided a greater performance boost than a dollar’s worth of DRAM once the DRAM size grew above a certain minimum (1-2GB) depending on the benchmark.
You might wonder how this could possibly be true. Everyone knows that best way to improve any computing system’s performance is to add DRAM main memory. How could flash, which is orders of magnitude slower than DRAM, provide a bigger performance boost than DRAM?
It all makes sense if you think of the DRAM of something that is there only to make the HDD look faster. More is better, but if you can use a little less DRAM and add a large flash memory layer then disk accesses appear to speed up even more.
The benchmark data and the price/performance findings that are based on these benchmarks are contained in a report called: How PC NAND Will Undermine DRAM. This post’s graphic shows one of the series of benchmarks from that report. The benchmark, PCMark Vantage Gaming, was run on a variety of DRAM sizes from 1GB to 8GB, shown across the bottom of the chart. The depth axis of the chart represents the amount of NAND flash, from zero (HDD-only) at the front to an SSD-only system at the rear. The vertical axis is the benchmark performance – higher is better.
The steep red part at the far left shows where DRAM really does a better job than flash, at the 1-2GB level. After that the chart increases more steeply in the depth direction than in the crosswise direction, showing the impact of adding NAND flash to the system. Other charts in the report map this data to cost to provide a clear comparison of the cost/performance benefits of adding either flash or DRAM to the system. Except at the very left-hand side of the chart, flash is always a better option.
Today a gigabyte of NAND flash costs about 1/15th as much as a gigabyte of DRAM, so for every gigabyte of DRAM that you don’t add to a system you can add almost 15 gigabytes of NAND flash. If you replace a little of the system’s DRAM with a big huge chunk of flash at the same cost you can improve performance significantly.
The argument is statistical, and is something I discussed in detail in my presentation at the 2015 Memcon conference. (Slide 21 summarizes everything.) If you add a few gigabytes of DRAM you will make a few accesses extremely fast, but the accesses that go outside of the DRAM to the HDD will be about one million times slower. If, instead of those few gigabytes of DRAM, you add fifteen times as many gigabytes of NAND flash, then you make a whole lot of accesses moderately fast, and you significantly reduce those phenomenally-slow HDD accesses.
Now none of this would be an issue if consumers simply converted from HDDs to SSDs in their PCs, but this won’t be happening anytime soon. I have been explaining why for over a decade. Consumers want a bigger “disk” for a lower price than they had in their last PC, and they can’t get there with an SSD.
The argument also goes directly against the widely-held, but completely erroneous precept that HDDs will all be replaced by SSDs in the near future. Back in 2006 I was shown studies that contended that the entire HDD market would be replaced by SSDs as early as 2008. Clearly that didn’t happen! The truth is, HDDs aren’t threatened by SSDs, DRAM is threatened by SSDs!
The SSD Guy didn’t invent the concept of using flash to reduce DRAM requirements – it’s a trick that has been used for over five years by data center managers to reduce their DRAM use in servers. They can add a caching SSD instead of more DRAM to make their servers meet their performance requirements. That’s good, because DRAM burns a lot of power, creates a lot of heat, and is very expensive.
But even though PCs behave like servers do, this approach has not yet found its way into the PC. I expect for that to change, but the industry is moving very slowly.
If you want to see the benchmark data in detail you can purchase the report for immediate download from the Objective Analysis website. If you want a preview then look at the Memcon presentation on the link above. Since I use a lot of animation in my presentations certain of the Memcon slides will be somewhat confusing. Rest assured that we have taken pains to ensure that the report itself is very clear.