All new Quadro family

[lateralustool]

Today nVidia introduced new lineup of their Quadro family:

 

http://www.nvidia.co.uk/object/quadro-desktop-gpu-specs-uk.html

 

New gpus:

 

K5200 - 8 GB

K4200

K2200

K620

K420

 

Price range £109 (K420), £1,759 (K5200)

 

Their are still Keppler based, not Maxwell. So I think it is far future from maxwell quadros.

 

I don't expect much better viewport performance in 3ds Max, probably still better choice would be GeForce.

Edited August 12, 2014 by lateralustool

[Dimitris Tolios]

Outside the K4200 that is now 256bit - a shy over 192bit K4000, I think the utility of having "double the RAM" with identical specs on all other models is nearly futile - doubt K2200 can utilize 4GB RAM over a 128bit bus for viewport purposes - forgetting how pathetic its compute is regardless of buffer capacity.

 

The K5200 could be interesting for compute at just 150W and having 8GB of RAM, but will probably cost more or "as-much" as 2x GTX Titan 6GB cards, with around 1/3~1/4th the raw performance of those.

 

Having experience on how impressive Maxwell compute is - even more Watt/actual compute performance - with my 750Ti, I would refrain from getting any Kepler card for this purpose right now. Medium/High grade Maxwell based GTXs are allegedly coming out as soon as next month.

[Francisco Penaloza]

mmm I also though that the new Quadros will come with Maxwell, actually my good old Quadro 4000 just dies a week ago, three weeks after warranty runned off LOL hour IT was very "Happy"

I was telling him to wait for the new upcoming Quadro line up, now I am not sure how good deal it will be, well more memory is always welcome so maybe it will be a 25% better than the K4000?

I had two video card in my Workstation the Quadro 4000 and GTX 580 for RT purposes, now I am running only with the GTX and everything seems fine. Those Fermi cores are pretty good still.

[Dimitris Tolios]

I don't know if there is a reliable formula to directly "predict" viewport performance based on CUDA cores / VRam and/or VRAM bandwidth.

For sure the K4200 is "more serious" than the K4000: 1344 cores & 256bit indicates a GK104 based card (GTX 670/760Ti) instead of the mediocre 768 core / 192bit K4000 based on GK106 (ala GTX 650Ti). Vram capacity was not the limiting factor - at least for viewport.

 

The K4200 should handle lots of things better, perhaps being well above 25% faster than the K4000.

Has the grunt to do well in D3D apps too. If that was still a $700-800 card, would be far better value than the K4000 was for the same money. If.

 

To be honest, this should be NVidia's K quadro line from the get-go. I can see that cheaper 8 GBit GDDR5 (1024MB chips) make it "easy" to boost VRam, but there was little excuse to cut back in GPU cores utilized so that they would "just outperform" 3-4yo Fermi cards.

 

Taking a better look @ the K2200, this is also a potentially a great update over the K2000, and with 4GB of RAM it might even be a decent compute card for RT! Why? It is a GM107 card - yes, the 1st Maxwell desktop Quadro!

The GTX 750ti is doing amazingly well in GPGPU (it can match a GTX780 in Luxmark with 1/4 the wattage, lulz), so having that kind of efficiency paired with 4GB of VRam might be interesting.

[RyderSK]

The GTX 750ti is doing amazingly well in GPGPU (it can match a GTX780 in Luxmark with 1/4 the wattage, lulz), so having that kind of efficiency paired with 4GB of VRam might be interesting.

 

Is it because of some OpenCL inneficiency in nVidia's implementation ? Or how is it possible ?

 

High grade Maxwell based GTXs are allegedly coming out as soon as next month.

 

I can't find any info when the 8GB GTX880 is supposed to come out. Only that Gigabyte leak, but they talk about 4GB version ?

[Dimitris Tolios]

8xx and "mid-sized" maxwell cards are coming either by the end of September according to some sources, or by November according to others. We don't know. But its coming "soon enough" for me not to be getting anything after selling my GTX Titan, and sticking with my peppy 750Ti for a couple months more.

 

Since nVidia appears to be using both 4GBit & 8GBit (=1024MB) RAM chips in their latest cards, we will definitely be seeing 4GB versions with their 256bit cores - probably we will be getting 4GB versions using 4x 1GB chips, easily upgradable (in the production line, not buy the user!) to 8GB with 8x 1GB chips in "clam-shell" mode - just like 6xx/7xx cards were with their 2 & 4GB versions (using 512MB chips).

 

Maxwell Architecture & OpenCL: I don't know what's the issue with Luxmark.

 

Are the packages sent by the program to big for the small Kepler SMX pipeline that was revised with Maxwell?

 

All I know is that the li GTX750Ti is reliably pulling scores that put shame to any GK104 card, and for the "medium" complexity setting it actually beats even the GK110 GTX Titan...

 

http://pcfoo.com/luxmark_v2_scores/

[elipan]

I stick with Geforce. Can afford myself another Quadro heartbreak. Another one will definitely kill me.