![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
In case of Windows, there is a well documented disrepancy. Hell, Wx7100 beats Vega FE in those benchmarks, are you happy?
Do quote it because i want to laugh at how silly you will look with your stupid numbers when gaming benchmarks hit and instead of being 75 times faster as you are peddling here, it is not even 40% faster (over 1070).
lol
The 330gflops of fp16 are for packed math, SM6.1 (Gp102, GP104, GP106) has one vec2 fp16 unit per SM so as you can imagine it's only used by sensible developers who don't subscribe to the absurd view that you should just use FP32 math at that point. Nonsense.
I know, after all nobody stops one from turning fp16 into fp32 and then back.
Well, it also serves as a way to milk fanboys for silly margins.
What efficiency gains?
Funny you say that, because the only efficiency gains over last gen they promised were with Polaris. They did not deliver on those either, but they never promised them for Vega.
Links/pictures/slides whatever, where are their efficiency claims they promised?
Budget conscious prosumers purchasing workstations with minimal graphics and replacing them with Titans and consumer drivers was the target AMD stated. How is the comparison invalid as that's exactly what AMD claims consumers are doing already? This card really shouldn't be used for CAD and content creation.
The purpose is to enable application developers. Hence no certified drivers and just hardware. I keep saying that, but you keep missing that point. The card is intended as a development kit where bugs and crashes are to be expected. Not caked into a server for deep learning. The extent of deep learning use should be to write the code to implement it.
There's also a dual Vega10 in documentation as a pro card to accompany it.
For whatever reason people get mad when you compare FP16 to FP32 performance because peak flop rates aren't comparable or something.
Nope, simply FP16 is kind of irrelevant outside of compute and mobile games.
Nice price tag for being a beta tester, huh.
There's still a difference in memory capacity, which is significant. So that's not having it both ways. For professionals with large datasets capacity would be significant and amplified by HBCC.
Or games with HDR, particles, hair, lots of simple physics, etc that make up a significant amount of work. Stuff found in high end and enthusiast gaming.
Well clockspeeds and price would be significant. Say 15% faster and a third the cost for gamers. The gaming version as Raja said will be faster, most likely from higher clockspeeds.
You mean besides new consoles? A platform with a specific path that covers the vast majority of the gaming market revenue? So yeah it seems all but guaranteed FP16 will see extensive use. We've already seen the feature used. Devs will just have to spend time dumbing down their code to accommodate Nvidia in the PC market. Not that different from disabling async for Nvidia already.
Can't say we've seen them used for deep learning or server accelerators yet.
Budget conscious prosumers as shown by AMD focus on CAD related would just go with the cheaper P4000 as Ieldra keeps mentioning.
WTF did I just watch??? Measured in Centuries? Is that how long AMD is going to take?
I do like that you can run Pro drivers on the card if you have specific applications that can take advantage of it. Issue is would one be switching back and forth with drivers if you want a more diversified setup? We just need some solid, down to the point reviews. Also would like to know if AMD VR has been brought up to better standings or works better.
Well if you make content using professional software such as Adobe Products to make games and play them - Use game engines like in the video using the Unreal engine (game) and then SteamVR, lol, it would be better if AMD just made the drivers do both well or have easy configuration change from one type of work environment to the next. Do agree having specialized rigs is more a mark of a professional plus computers today are cheap compared to 20 years ago costs. Back in 1989 I bought a whopping 20meg (yep meg) Supra hard drive with 2mb (yep mb) of EDO ram for $950 for my Amiga 500. Inflation over that time would make that like $1500 today. So a $1200 Titan Xp or FE Vega for what you get is peanuts compared to back then. We just have to see once folks can start reviewing them to figure out WTH. RTG is so vague with weird off the wall points of almost any discussion with Vega it becomes pointless to even pay attention to RTG.
Well that is exactly what I get from RTG so far, evasive, non-answering, told to wait etc.
http://pro.radeon.com/en-us/product/radeon-vega-frontier-edition/
.
Well well, didn't see this:
http://pro.radeon.com/en-us/product/radeon-vega-frontier-edition/
So looks like the drivers will have modes optimized for different work loads. Sounds good if they deliver. Maybe I should read more but usually end up hating it
Tiled rasterization reducing the workload 15-40%? That one software feature accounts for nearly all their lead in poorly optimized titles.
It's a development card. Any pro creating models or using CAD shouldn't be using this card. Maybe devs making CAD applications. Engine development or experimentation with any GPU acceleration then sure. Artists, outside indie devs, have other options. Best analogy is that FE looks like a XBox development kit.
Maybe on Nvidia products where they lack the capability. Game devs have already started using FP16 on console. Next generation of engines furthering that use on PC. Based on all the people I've seen arguing against it, I'm guessing consumer Volta will lack it as well.
Indie devs with third party engines and low budgets are more of a thing now.
Well well, didn't see this:
http://pro.radeon.com/en-us/product/radeon-vega-frontier-edition/
So looks like the drivers will have modes optimized for different work loads. Sounds good if they deliver. Maybe I should read more but usually end up hating it