Hacker News Reader

Comment by frou_dh | Hacker News Reader

Consumers tend to be limited by not having many PCIe SSDs in RAID0?

Who are these people and what do they need galactic disk bandwidth for?

Enthusiasts tend to be gamers, and once you have a high end machine, most games tend to be limited on disk bandwidth, which affects load times.

By consumers, I meant enthusiasts, since that's who the Threadripper (X399) and Intel's Extreme Edition (X299) platforms are aimed at.

Enthusiast here - can confirm, an array of nvme's makes a huge difference over single standard SSD. I set this up last year, check the benchmark results, it's night and day.

http://imgur.com/a/N4FpD

note: be sure to note the scaling at the bottom. 4-500MB/s old vs 3000+ MB/s new

Do games actually load any faster? I went from a Samsung SSD(I think the 850 pro) to an Intel nvme ssd, and load times in games didn't seem much faster. I definitely notice it when booting up VS though.

Gamers might the most vocal high-end PC users out there but they are not the people with the most money.

The threadripper is not targeted at gamers. It is targeted at people who make money from using their machines, these people can, therefore, afford to upgrade.

Consider a user who does video work. (most modern films and tv is filmed in 8k know) if this is in Raw format as it comes off the camera for a 90 min episode that is 6000 GB. Working on this data (tone mapping, cutting, etc) needs all this data to be fed in and out of the (CPU and or GPUs) so all these PCI lanes are there so you can have very fast access to the raw data and still have space for 2 or 4 full speed GPUs to do that work for you.

Forget games no game developer optimizes their games for 4 GPUs these days even Nvida have said that they dont realy want people to do more than 2GPUs in SLI.

I do a lot of compiles of a codebase which is pretty big (Netflix version of FreeBSD). A full recompile takes me ~1 hr on my box (Xeon E5-2630 v3, 8 cores, 16 threads, 32GB DDR4). Most of this is a parallel make that is 100% CPU bound. So I'm probably going to update to thread ripper when it comes out. Doubling my core count should come close to halving my compile times.

I probably will do storage via a ZFS array of small-ish SSDs or NVME drives. I started with a ZFS array of spinning drives when I built this machine ~2 years ago, with an SSD for L2 ARC. I was occasionally IO bound, so I moved most of the build stuff onto its own SSD. That's probably enough, but I want redundancy (without spinning drives to back up to), and I suck at cabling, so I'm leaning towards just using NVME drives in PCIe slots in order to avoid the cables.

> A full recompile takes me ~1 hr

Could you explain why full recompiles on a single machine are necessary? Are you testing different compiler settings?

I work remotely, and it is easier to develop and compile on the same box. I normally compile just the kernel, but occasionally have to rebuild the entire thing.

What makefile/build system r u using ?

I am curious since i did some work on parallel builds and I always wonder how much of this is (easily) transferable to other setups...

A modified version of the standard FreeBSD build, with bsd make. Eg, a script driven 'make -j X buildworld && make -j X buildkernel'

if you had access to a hosted build system, which is x10 faster. lets even assume some sort of transparent gradual sync of your source tree, and the binaries back to your system.

would u find value in such a setup or is it not worth the hassle ?

(lets assume the output is veritably binary-identical to your local build)

I use the system more for work than games but one game it did make a very noticeable difference in was Minecraft believe it or not. Long-view distance + pre-generated map (world border fill) = move/fly fast without chunks taking friggin forever to load :-)

That's a very clean looking system! Kudos. :)

I have only one of the 512 950 PROs in mine; they are awesome. :) http://i.imgur.com/sydrBvv.jpg . Pic taken before 2nd 980 GPU was put in; 6850k CPU.

A lot of processing happens when games load. Once you have sufficient storage bandwidth and latency, load times flatten out. It seems that "sufficient" is SATA; above that, there is little difference in game load times:

https://techreport.com/review/29221/samsung-950-pro-512gb-ss...

I wonder if game developers are actually disincentivized from qualitatively improving load times past a certain acceptable threshold, lest their AAA game be perceived as lightweight.

I doubt it. Gamers have been complaining about long load times since almost the dawn of time, and I think one with as little as technically possible could make it a selling point.

The poor market penetration of PCIe SSDs and their high price per GB is a strong disincentive for game developers to store data in lightly-compressed formats that require minimal CPU processing to load and use. We might see some improvement in using multiple threads to load game data now that CPU core counts are on the rise at the high end.

Reminds me of the first time I played Kings Quest from a RAM drive. I first played the game from a floppy disk, and it ways really nice to have everything load instantly!

That sounds frivolous versus a single high-end PCIe SSD. Wouldn't be surprised if the improvement in real-world game load times wasn't even perceptible.

>That sounds frivolous

Maybe, but that's what being an enthusiast is all about :) also assets (models & textures) need to be loaded from disk during game-play which can result in stutter when gaming. Faster IO would help here.

Our baseline of a single high-end PCIe SSD isn't slow, though. This comparison is ultra high bandwidth vs ultra ultra high bandwidth.

I think gaming and ARM cannot be used in the same context. The extra IO is useless to the gamer unless there is an Nvidia GPU on the other end.

I've been shopping for a future proof workstation (10Gbps Ethernet 8x + 2 16x video cards) that couldn't quite fit on a Xeon E3.

What is really nice about the full Zen line is they are supporting ECC memory. It always felt like Intel had removed this on lower end chips just to force users to buy a Xeon even if the did not need the rest of the Xeon features.

Yeah, the Xeon-E3 lines lack of PCI-e lanes is really underwhelming. I've got a LSI 9200-8e and Mellanox Connect-X 2 in my ProLiant ML10 used as my FreeNAS box, leaving a total of 4 PCIe lanes available - thankfully I don't see any need for more AIB's in this machine, but if I used it as a workstation that needed a GPU + another board or two it'd be a really awful situation to be in.

Considering how hard Intel is pushing Thunderbolt it makes little sense how hard they're gimping their consumer and workstation CPU's when it comes to PCIe support.

Reply on news.ycombinator.com