New build, new concerns (can't test for 2 weeks) -

[mbox_bob]

This album shows some single player VRAM usage numbers at the Alliance garrison in Shadowmoon Valley. You can subtract ~430MB off of those numbers since my idle desktop eats up that much before loading up the game client.

They're all 2560x1440 (based off of the monitor you've purchased) and they're without any anti-aliasing. From the looks of it, and assuming you want to run a resolution somewhere in the ballpark of 2560x1440, you may have to mix and match your APIs since DX11 on 10 clients is likely going to use over 4GB of VRAM.

And yes, in that album, DX9 on Ultra was running terrible. I do suggest using DX11 these days whenever possible since you're bound to get better framerates, and you may have to end up disabling SSAO altogether since it tends to eat a lot of VRAM and GPU.

Probably worth noting that as you load more clients they can reuse the loaded textures, so you don't necessarily need 10x the vram indicated. Not to say that you'll be sweet running Ultra settings on 10 clients, but you may be able to play comfortably with the backgrounders on lower details. This is definitely one of those, try it out, and see what gives. I haven't seen any reports back on the Fury/HBM and multiboxing settings (not that I've been looking hard though). Think of yourself as the guinea pig, and we will await your report .

Also the AMD Fury does gain some advantage in the HBM, purely because it can shift data faster, so when it offloads to the system ram buffer, it moves more data, so can still perform even with less. Of course moving data from vram to system ram then means the bus is moving data in the wrong direction, but it at least means the perf drop isn't quite as hard hitting while it does it.

Probably not worth it since SSDs are fast enough on their own, but if you were going to buy another SSD then it'd be useful as a drive just for the OS or just for games (however you want to look at it).

Too true. If you really want "faster daddy, faster" SSD, then you are better off getting an M.2, like the Samsung SM951 (or the SM951nvm, if it happens to be out). M.2's are generally in the region of 2 - 3x faster than a SATA SSD, although you need to make sure they are the new gen (those Gen2 on PLE) and not the crossovers to get the big perf increase.

[MiRai]

Probably worth noting that as you load more clients they can reuse the loaded textures, so you don't necessarily need 10x the vram indicated.

As much as I wanted to believe that back over a year ago, it just doesn't seem to be true with World of Warcraft. I ran another quick test just now because Blizzard did update their rendering engine with the last expansion and I wanted to see if something had changed, so here are my results.

1.4 GB Idle Desktop (Lots of things running in the background)
2.2 GB 1 Client
3.0 GB 2 Clients
3.8 GB 3 Clients
4.7 GB 4 Clients
5.5 GB 5 Clients

So, each client at the location I was at, which differs from my location in my prior post, eats up about 825MB, and the math works out almost perfectly.

5 Clients x ~825MB = 4.125 GB + 1.4 GB (Idle Desktop) = ~5.5 GB Total VRAM

Other games may handle the textures they store in VRAM a bit differently, but WoW definitely seems to end up being a simple addition problem (or multiplication if you want to get fancy) when trying to figure out total VRAM usage.

Also the AMD Fury does gain some advantage in the HBM, purely because it can shift data faster, so when it offloads to the system ram buffer, it moves more data, so can still perform even with less. Of course moving data from vram to system ram then means the bus is moving data in the wrong direction, but it at least means the perf drop isn't quite as hard hitting while it does it.

I agree. The Fury X was able to keep up with both the 980 Ti and Titan X in games where it should have been losing due to having its VRAM overloaded (e.g. Shadows of Mordor, GTA V), but I certainly don't know how it's going to handle multiple game clients pushing it beyond its limit, if it comes to that (but I'm also looking forward to feedback from people).

Too true. If you really want "faster daddy, faster" SSD, then you are better off getting an M.2, like the Samsung SM951 (or the SM951nvm, if it happens to be out). M.2's are generally in the region of 2 - 3x faster than a SATA SSD, although you need to make sure they are the new gen (those Gen2 on PLE) and not the crossovers to get the big perf increase.

I definitely want to try out an M.2 for my next build, but it's difficult to justify the price. Yeah, it's faster than a standard SSD, but I have SSDs hooked up to both SATAIII and SATAII ports (as well as the RAMDrive I had used before), and the only difference I notice between them is on CrystalDiskMark.