Alledged Benchmarks Titan X ; 980Ti; 390X

As this is WCCFTech, take with a grain of salt but if this is true, AMD has a winner as it is better than the Titan X and much cheaper and they are saying the 390X will launch with 8GB of the stacked VRAM.

http://wccftech.com/amd-r9-390x-nvidia-gtx-980ti-titanx-benchmarks/




Edited by Dax Doomslayer - 14 Mar 2015 at 9:27am

NVIDIA has publicly acknowledged that HBM-style memory will be used on the next generation of chip, Pascal, which won't come out until sometime next year. If the 980 Ti does pop up, and if so it will probably be this week at NVIDIA's GTC, in all likelihood it will be a cut-down GM200 chip from the Titan X with half the VRAM.

Dax,

Get the Aventum 3 with one or two NVIDIA cards fully water cooled. When the R9 390X cards, get one or two of them and add them to the rig and take advantage of the heterogeneous compute abilities of DX12. (I'd talk to DS about the water blocks on the cards to make sure you can bridge between them.)

So, if you are going to go with air-cooling for the GPUs, keep in mind that the only motherboards out there that allow for an extra slot of spacing between the GPUs in a 3-way set-up are the EVGA boards (you'll have to sacrifice 8 PCI-e lanes to do it, but with the interface bandwidth usage right now, that isn't as big a deal as cooling). Otherwise, with the ASUS cards, you will have two cards right next to one another with no spacing. While you can do this, you'll get some high temps in the top cards (or bottom card when the motherboard is flipped in the Velox or Aventum).

Originally posted by Dax Doomslayer Yes the 390X looks to be a pretty strong card if this is accurate and NVIDIA won't be migrating to HBM until 2016 and Pascal.  It does look like there may be a Ti card now which somewhat surprises me giving the fact that the Titan X doesn't do D.P. well and is more of a gaming card this time around.  That said, I am now torn as I was just going to go with x3 Titans but if I wait, I'm sure this card will be cheaper (and better).  I'm not sure I can wait until June.

even w/o DP the Titan X is still a prosumer card as opposed to "gamer" card.  it's really niche, filling the workstation+gaming user who doesn't want to shell out lots of cash for more expensive setups and who also likes to game.  however, as i've stated in another thread this card doesn't make a whole lot of sense to people who build their rig around gaming.  there are cheaper options (yes, even with a single card) for handling sub 4k gaming resolutions, and there are better options for handling good (which this card struggles at) 4k gaming experiences with dual card setups.  it's still unclear how R9 390X will handle 4k as a single card solution, so until we see gaming benchmarks it's all theoretical. 

i don't think you can even classify this card as an "enthusiast's card" because enthusiast's don't target sub 60 fps at 4k res.  when you launch a "flagship card" (at $1,000) it had better hit that magic number on the highest res currently available (on ultra settings).  because, why game at 4k on medium settings?  that's a fail imo. 

tldr: unless you need cuda cores and lots of GDDR5 memory for work i'd pass on this expensive niche card. 


Edited by db188 - 19 Mar 2015 at 11:18pm

Yes, the WS board could do it, but they're using a couple layers of PCI-e switches, including a PLX with routing capabilities. That should provide the same bandwidth (16 lanes to the top card and 16 lanes to the bottom two cards) but at the expense of a couple hundred nanoseconds (with processor clock cycles in the GHz region, each cycle is sub-nanosecond). The WS board is much better suited for compute applications that feed a ton of data to the GPUs instead of having tons of draw calls as there are with graphics processing. The EVGA boards (as well as most consumer boards) only use a single layer of switches which don't have routing capabilities, so they're much lower in latency.