Nowdays there isn't a huge amount of scaling difference between small medium and large dies. GP102 GP104 and GP106 scale pretty similarly, and all reach about 2 to 2.1ghz maximum before hitting the voltage wall, small dies have the benefit of having less chance of having a part of the die be worse than the rest, hence usually higher overclocks in return.
Could take Fiji's massive 596mm^2 die again and compare it to the other GCN cards, all hit a wall at about 1200 to 1250mhz on a good bin. Good Polaris 20 chips on the RX 580s can hit 1500mhz on watercooling, If Vega is tuned for higher clockrates I don't think it's unlikely it can hit those speeds as well at stock.
Usually architecture matters just as much as the node when it comes to clockrates, just remember that the shaders on old cards like the 8800 GTX could hit 1700mhz, and that's on a 90nm node from 2006.
That has got to be some sort of LN2 benchmark.. You were lucky if your card could do 850MHz, and that's with strapping an aftermarket cooler; pretty much all cards in that day came with those leaf-blower coolers
I'm talking about the shader frequency, I remember my old 8800 gtx with a zalman aftermarket air cooler did 1750mhz on the shaders. Old Tesla based GPUs had 3 frequencies, core/memory/shader.
Stock a 8800 gtx did 575 core 1350 shader. I'm not sure if you could do the same in 200 series card, Fermi for example had double pumped shaders, so the shader clock was always twice the core clock.
I'm not sure how i feel. With videocards, they improve so rapidly, and are so expensive, it is becoming harder and harder to justify purchasing them. I kind of want a new videocard soonish, but its late in nvidia's 10 series and AMD doesn't have anything worth the upgrade over my 970. If i have the willpower i will probably wait until the next gen from either company that forces a price drop (so basically you get more for your money, if a high end vega card is comparable performance to the 1080 or 1080ti and comparable price, i'm not going to buy it) and get something in the $400-$450 range, and then if all things are equal i'll get the AMD card.
Generally, I consider upgrading whenever the current card isnt quite cutting it anymore for whatever game occupies my time; although, while it was bit of an unconventional leap, I did recently upgrade from a 970 to a Fury(but that was more to do with nVidia's poor displayport support) 970's are still relevant today. If I were in your position, I'd likely look to upgrade whenever a card exists that will solidly handle games at 4k+ with whatever arbitrary refresh-rate my future monitor can output.. Or whenever the 970 begins to noticeably chug.
the problem is i'm used to 144hz and its going to be awhile before a $500 card can handle 4k at that. More like i have to buy videocard to keep my 1080p 144hz smooth until its possible to run 144hz 1440p or 4k
That's nothing new, it will be quite the technical achievement if AMD manage to get the RX Vega which is 12.5tflop (higher then titan btw) to perform like a 1070
We can all stand up and clap about that blunder, it would take special effort!
I´m not really sure why the RX580 is stated as outdated in the op? The polaris architecture is still build upon 14nm Finfet. Which isnt really that outdated atall.
Yeah maybe thats what he means idk. But Polaris has never really been intended to be flagship. It´s ment to be competing in the midrange market. The current flagship gpu architecture on AMD is basiclly still Fiji (FuryX). Which in that point of view is indeed a bit dated nowdays. But Polaris isnt really outdated as far as i´m concerned.
Yeah AMD is now working on Vega which i'm personally pretty excited about. Some of the leaked rumored specs do look promissing compaired to Fiji. But we will only know how they perform wenn they are out. Some competition in the highend gpu market would be nice.
Yes, me too. Maybe they'll just do an announcement at Computex, then limited availability at the end of June. Plenty of rumors of limited availability. We'll see what's what soon enough.
I think certain stacks of HBM2 is rare, atm 4GB stacks are probably good to go which is why we are seeing a likely lineup of 8GB cards coming and not 16GB (2 stacks per GPU).
So NVIDIA still has bragging rights as their top cards are 11GB, lets hope this magic efficiency in HBM2 actually amounts to something.
Its worth noting that I have seen quite a few 4k benchmarks now eat 9-10GB of VRAM, so it will be interesting to see how 8GB HBM2 fairs because If you've ever played a game where vram got limited, fps stutter galore.
