• Announcements

    • SFLUFAN

      D1P 2017 Charity Campaign for The Life You Can Save: $740 (as of 07 January)   12/12/2016

      I've decided to extend our charity campaign for The Life You Can Save organizations for the entirety of 2017 so feel free to contribute at any time!  Periodically through the year, I'll have game giveaways for those who have donated to the campaign as a "thank you" for supporting this worthy endeavor!
    • Jason

      Missing icons.   12/27/2016

      Unfortunately, we lost a bunch of user icons due to the data loss a few days ago—but we didn't intentionally remove any so if yours is gone please feel free to just set it back.   -Jason

NextGen

AMD Propaganda Tsar
  • Content count

    14,888
  • Joined

  • Last visited

  • Days Won

    5

NextGen last won the day on February 8 2016

NextGen had the most liked content!

Community Reputation

901

About NextGen

  • Rank
    AMD FineWine™ Technology
  • Birthday 06/07/1988

Contact Methods

  • Website URL
    http://steamcommunity.com/id/ngdot24

Profile Information

  • Gender
    Male
  • Location
    Milwaukee, WI
  • PSN ID
    Canownage
  • Steam ID
    ngdot24

Recent Profile Visitors

33,409 profile views
  1. None of this applies to Ryzen. That's the real deal.
  2. 0℅
  3. The Vega Doom demo was not using either the new variable-width SIMD or the new tile-based rasterizer binning process. There is room to grow performance-wise, but it's not #poorvolta.
  4. That's exactly what I do.
  5. @Reputator Not so much lamenting it, as pointing out the hypocrisy in their marketing. You don't get to run an ad that states this; then turn around and release a card that's only marginally faster than <not the flagship> Pascal architecture a year after the fact. To your next point, I don't consider AMD's GPU's to be bloated at all. Raw compute performance was a very forward thinking way to design a GPU with GCN several years ago. The whole, 'inspiring tomorrow' catch phrase is very true in that regard. What I'm saying is, is if they removed all the "stuff" that could be/might be awesome to have in...2020, and used that <wasted> die space to pack in something like more shaders instead--the gamer would get a better performing card. At the end of the day, that's what we care about. FPS. AMD could always use that "stuff" in the professional workstation cards for workloads that can actually leverage it now with their WX and Insight brands. I don't think AMD is being parsimonious with the Radeon brand, I just don't know how they fix the issue of Nvidia mindshare. It's a really unenviable position for them to be in. and one we might suffer for.
  6. I sure hope not. The compression and relatively high latency compared to a local machine would destroy precision gaming and high fidelity graphics.
  7. That's called Radeon Instinct.
  8. 4K/144hz
  9. Sure, that's the plan. Do you think AMD will be able to launch Vega 10 and Vega 20 a mere 6(ish) months apart for the enthusiast gamer? That doesn't seem realistic to me.
  10. I don't expect Volta until 2018. However, I do expect Vega to be Fury X'd by the 1080ti. That said, a lot of this flies out the window if Vega is aggressively priced.
  11. Firstly, nobody wanted Vega to be an out of the park HR more than I did. So take this ‘report’ from the perspective of a disappointed AMD fan, and PC gamer. It’s not meant to berate any company, throw shade on any company, or generally speaking, belittle anything that a given company is doing. With that out of the way, in order to understand what I’m going to be talking about, it’s going to be critically important to understand some ‘advanced basics’ about the world of GPU’s. Many of you probably already know this stuff, but for the purposes of those who don’t, or just need a refresher, I figured this is the most logical place to start. Physical GPU size. GPU’s are ‘cut’ from a single silicon wafer. These wafers are ordered and purchased by Nvidia and AMD for the purposes of creating your GTX 1080’s and RX 460s and the like. Think of it like a big slice of pizza, cut into little tiny squares by either TSMC or Global Foundries. The littler the square pieces of pizza are, the more slices you’re ultimately able to get out of the singular pizza. However, lets say that the oven you’re using to bake this pizza is terrible, and has a propensity to burn some slices here or there, as well as under cook other slices. So, out of 100 potential slices of the pizza, lets say 50 are cooked properly—or how you intended them to turn out. 25 of the other slices are a little undercooked, but still probably fine as long as you aren’t expecting perfection. 20 of them have burnt crust, but in a pinch are fine, and 5 of them are totally beyond repair and go straight to the trash. Now, not all slices of pizza are created equal, some companies cut their slices a bit bigger than others which limits the amount of slices you’re able to get from a single pizza. That's basically the binning process. It's how you get GTX 1070's and GTX 1060's. Lets take a look at chip we’re probably all familiar with in one way or another though-- the GP 102, or, as it is commonly referred to as, the GTX 1080. The GTX 1080 has a die size of 314mm² and is running with 8GB of GDDR5X. Everything inside the die of the GPU takes up physical space. Transistors, shaders, CUDA cores, ROPs, memory controllers..etc. The smaller you can cram this package together, the more slices you get. It’s why the jump from 28nm (R9 290/GTX 980..etc) to 14nm FinFet and 16nm was so massively huge for the industry. We know the performance of GP 102 in the form of the GTX 1080. We know it’s a great card capable of handling 4K gaming at high settings. But wait, Nextgen, If a 314mm² GP 102 GPU is so amazing, then why isn’t the Fiji GPU, which is 596mm2 faster? After all, who doesn’t want a bigger slice of pizza? That comes down to architectural differences between Nvidia and AMD. Essentially, how they choose to make their pizza, ‘their’ pizza. By way of using different sauces, different ways of preparing the bread, maybe firing it over a wood burning flame instead. You get the idea. A 500mm² GPU from Nvidia, is flat out better than a 500mm² GPU from AMD for the purposes of gaming. It’s why the GP 100 card (The Tesla card) from Nvidia is leagues ahead of Fury X despite physically being only a few mm² larger, at 610mm² This brings us to Vega. The brand new, upcoming GPU from AMD, weighing in at 475mm². If you’ve been paying attention, you’ll notice how obscenely close that number is to the full “fat” GP 102 (The Titan XP) and what we’ve already learned about the die size from AMD to Nvidia in relation to performance in games. Now, Vega is a huge departure from GCN (Like, the R9 390 for example) it’s essentially been completely rebuilt from the ground up! It is still leveraging things like Async Compute and Freesync and all of the great things we’ve come to expect from GCN, but using them more efficiently and effectively to lower power draw, and give us gamers more of those sweet sweet FPS. AMD is calling it NCU. Or Next Compute Unit. Perhaps the biggest change to the architecture is the addition of High Bandwidth Cache, which in all honestly, is a solution to a problem that doesn’t yet exist for high end gaming. HBC will let the GPU intelligently and automatically with no developer input, cache..stuff (textures, polygons, you name it) instead of loading it directly into the framebuffer (VRAM) This is the same principle Radeon showed off with its SSG technology. So while neat that I can cache textures off a network share (no, really, you can) this isn’t a consumer level feature in 2017.What we really care about though is games. We want games to run fast, no hitching or micro stutter at high resolutions. Thankfully, we’ve already seen the performance of Vega in Doom, at 4K, with ultra-settings. AMD actually provided that video themselves it was running at about, 10% faster than a 1080 give or take. Most of that can probably be attributed to the NCU performance gains. 10% faster isn’t markedly better for what is probably a best case scenario running under the Vulkan API. In fact, it’s pretty disappointing. Nvidia, with Volta will be able to further expand the physical size of the GPU as ongoing refinement to the 16nm process node makes for more perfectly cooked slices of pizza, and less burned slices. Lets say for the sake of argument Volta is 596mm² --the same size as Fury X with stacked HBM2 memory and proper Async Compute implementation. Can you imagine a GTX 1080 with nearly a 100% performance improvement? That’s what you could be looking at with a “GTX 2080” IF Nvidia wants to drop the hammer. The best part of it, is they’d be dropping the hammer using AMD pioneered technologies. Beating them at their own game, while beating them in all games. It’s a really bad situation, as a consumer, and for AMD to be in. Edit: fixed weird text formatting.
  12. Or, they wanted to visit the White House and President who was in office when they won the World Series. Stop trying to add divisive politics to baseball.
  13. Seahawks: 31-26 Patriots: 43-13 Steelers: 30-27 Packers: 37-24
  14. Humble Bundle is as legit as it gets, and for a great cause. @Runny