AMD Teases Upcoming Video Card

75 Comments

View All Comments

• CiccioB - Sunday, May 24, 2015 - link

  And you are connecting those separated memory pools with? PCI 3.0 at 16GB/s?
  Good luck for doing anything useful in a 3D game.

• ES_Revenge - Tuesday, May 26, 2015 - link

  If you're using AFR (which I think is the most common way SLI and Crossfire are implemented these days) then no you don't really have double the RAM *effectively* as each GPU only has access to its own VRAM (3GB x 2 or what have you). In SFR though I think it works out that you have more VRAM because instead of each GPU rendering each entire [alternating] frame like in AFR, each GPU is only rendering half the frame. So it's only needing to use half the VRAM it would otherwise (approximately, I mean if one is rendering the sky and the other a lot of textures then it's not exactly 50/50)... So I *think* it is possible to have "double the VRAM" but since AFR is the most common method, it doesn't work like that in reality. And yeah PCIe is far too slow to effectively share RAM over, in a pool.

• Alexvrb - Tuesday, May 26, 2015 - link

  I really do not care for AFR. I always wanted to see widespread use of a tiled multi-GPU rendering implementation - it would divide work up more evenly. Oh well... at least with the new lower-level APIs there might be some creative use of multiple GPUs that doesn't rely on AFR.

• CiccioB - Sunday, May 24, 2015 - link

  AMD is already very late with this GPU, I expect the X2 coming too close to 14/16nm new GPU release to have any appeal.

• RussianSensation - Thursday, May 28, 2015 - link

  They are not really that late in the context of the market's competitor. The Titan X launched for $1K in mid-March 2017 and 980Ti is only being launched June 2nd, 2015. Fiji XT is launching this June by all accounts, which means it's barely behind 980Ti. Where AMD is late is with GTX970/980 competitors but one could argue that R9 290/290X were good enough performance wise and they certainly offered superior price/performance. They got just tarnished brand image since nearly every professional site ignored the existence of cool and quiet after-market R9 290 series of cards. There are rumours that Hawaii enhanced will just be a slightly clocked version with 8GB of GDDR5 to better compete against 970/980, with possibly some improvements in perf/watt. Other than perf/watt, 970/980 barely moved the needle when it comes to high-end performance from 290X, which is now approaching 1.5 years old. Where AMD really needs better cards is in the mobile dGPU space and > $400 level (Fiji Pro and XT).

• CPUGPUGURU - Saturday, May 23, 2015 - link

  AMD's hot watt wasting ReBranded Tonga glued to 4K gaming inept limited to only 4GB HBM1 is going to beaten like Lame Llano filled pinata by, highly over clocked performance per watt champion Maxwell GTX980, TitanX and upcoming GTX980ti.
  
  read weep and cry me a amazon river,
  
  The water cooled nature of this card has other tertiary benefits as well. While EVGA doesn’t unlock anything spectacular for overclockers (+87mV and +25% for the voltage and Power Limit respectively), our sample hit some impressive levels, running at 1620MHz for hours on end without the smallest hiccup. As you can imagine, the card is also very, very quiet due to the 120mm fan running alongside the low-RPM blower.
  
  There are so many positive points about the EVGA GTX 980 Hybrid that its potential downfalls may be overlooked. From a compatibility standpoint, there shouldn’t be any problem getting this card to fit into nearly any case on the market. However, any system that’s already been equipped with an All in One cooler may find itself with some limitations when installing the Hybrid since there are only so many accessible 120mm ports on some cases.
  
  With the GTX 980 Hybrid, EVGA has created an awesome graphics card that performs at an extremely high level, runs cool, grants acoustic-minded individuals a quiet environment and provides an impressive amount of overclocking headroom. It may not be able to perform up to the level of a TITAN X, EVGA’s iteration costs significantly less while still delivering some of the highest framerates around. What more is there to ask for?

• formulav8 - Saturday, May 23, 2015 - link

  Are you posting on the correct article? Your post looks to be a bunch of gibberish.

• CPUGPUGURU - Saturday, May 23, 2015 - link

  AMD's hot watt wasting ReBranded Tonga glued to 4K gaming inept because its limited to only 4GB HBM1 = AMD HBM1 390, I think that's what's pictured so its the correct article. But I do wish for a edit option as my point is that the 390 is 4K gaming cripple, and a ReBranded Tonga GPU that Maxwell highly over clocked 980, 980ti, and Titan X will be beating like a pinata.
  
  AMD 390 too late, too lame (hot watt wasting Tonga), too little (only 4GB memory) to 4K game.
  
  We will see if I'm right, AMD has already lost massive market share to performance per watt champion Maxwell and since all of AMD's 3xx series GPUs are ReBranded with one glued to HBM1 the market share beat will go on and on.
  
  Have a Great Memorial Weekend

• testbug00 - Sunday, May 24, 2015 - link

  From an engineering standpoint, you can have over 1GB of HBM1 per 1024-bit bus. It's just very expensive. And an engineering challenge, but, AMD has plenty of engineers who can do the job.
  
  The extra projected revenue from it likely is lower than the cost to do it. And, memory usage depends on the game and resolution and settings. Not the resolution. TWIII for example...
  
  not to mention that due to moving things in and out of memory, you can have a 4GB and 12GB card with the same settings and have different RAM utilization amounts. It becomes a question of how much RAM the game needs to use if it only has what it needs in RAM. That amount will vary drastically game to game, in all likely-hood.

• Hicks12 - Sunday, May 24, 2015 - link

  People seem to keep banging on about 4GB not being enough but AMD has thrown several developers onto this issue, compression will most likely be implemented and ensure less is more :P.
  
  Wait and see, should be interesting to see how it pans out !