Of the 1.5x perf/watt improvement over Vega, 60% of that is attributed to the new architecture, and not the node shrink. As in, if Navi were on 16nm still at the same clocks as Vega, it'd have around a 30% improved perf/watt over Vega.
It is likely that launch drivers are going to leave a lot of performance left to be gained as they'll need newly optimized shader compilers.
RDNA CUs are more different to Vega ("GCN 1.4") than Turing SMs are to Pascal, when Turing SMs changed quite a lot.
Don't worry it is still terrible when compared with 2070.
Connor Nguyen
>inb4 it goes toe to toe woth the rtx 3080 in a few years
Jonathan Evans
I look forward to Just Wait™ing until the drivers are good.
Mason Martinez
You must have never gotten to experience FineWine™ >Buy 7970 >It's okay at launch. Much better than Fermi. Just couldn't know then that it was the best GPU ever made, relative to the time it was released. >Kepler comes and is competitive, but then drivers make the 7970 you already got better than it and Kepler refresh >More years later, it still runs all the new stuff great, thanks for improved drivers and games being more compute heavy, and it having enough VRAM which Nvidia cards always skimped on >Sell it to a miner for 1/3rd of what you paid, 6 years later
>Buy Ryzen 1000 series >At launch, just okay. Better than $250 for 4c/4t stutterlake, even if averages are sometimes worse >Next year, microcode upgrades greatly improve memory compatibility, and also improve IPC about 6%. Now it's even better >Another year later, Windows Scheduler gives another up to 20% improvement >Don't even need to upgrade to Zen 2 like you thought because it FineWine™'d
It's not that it's bad on launch. 7970 wasn't bad on launch, it was still better perf/$, perf/watt, even if drivers had weird bugs. Ryzen 1 wasn't bad on launch. It just gets even better with time.
Jason Diaz
Windows update "allegedly" fixed 1% fps drop while gaming. Wait for real tests and not just MS shilling.
Connor Evans
Here comes the wait and see bullshit again.. Anyone remember vega? I actually bought a card and put my money on the line.. Yeah, the wait never paid out.. Drivers never came. Features were never enabled. Just a half ass'd power inefficient architecture that delivered sensible middle ground performance and nothing more.
Carson Roberts
You are definitely meming if you think vega didn't deliver. V56 is like 1070ti performance and when undervolted runs around 200w and less. V64 is good too and trades blow with 1080 depending on the game. Yes they do take more energy but their performance was good.
Wyatt Peterson
Updated my X470 chipset, latest BIOS and the Windows 10 May Update. My 2700x is absolutely a monster. Everything is snappy and smooth and it already was very nice. Glad to see MS partnering more with AMD. New consoles are Zen 2 and Navi. Second hand market is really good right now and AMD is luring Nvidia into a price war. The future is bright! Not to mention SSDs and RAM are getting cheaper! Can't decide on the 3800x or the 3950x.
Oliver Green
10 Advanced Micro Dollars have been deposited into your account.
I actually kinda wonder whether the performance/power improvements they gave on the presentation were for Navi which is supposedly a GCN/RDNA hybrid or actually for future full RDNA architecture. I'd say it's a valid question because the slide with those numbers was about RDNA in general and not about Navi specifically
Justin Allen
Meanwhile I still use my win7 installation I did on 3570k now on my r5 1600 and everything was snappy all along. Too bad win10 requires so much tinkering, probably still laggy compared to win7.
Kayden Garcia
and already the "just wait, fine whine" has started before the new gpus have even released. Share holders must be really worried about amd stock prices.
Josiah Turner
They go with the Fine Wine™ bullshit while at the same time bashing nvidia for bad drivers. How shit must AMD's drivers be if they take away that much performance at launch.
Jack Price
Cope. :)
Joshua Diaz
considering CUs now are using IF its gonna take a lot more time and NOT just drivers sadly
this effectively means that each dual CU acts like a core and they are all basicly on a crossfire situation
in a tl dr version
this is bulldozer GPU that somehow is faster
Gabriel Powell
RDNA appears to have made optimizations and improvements over GCN across the board. I don't get why the power requirements is still higher than the competing gpu made on an inferior process.
because amd wont ditch the hardware sc no matter what (there is no way of emulating async especially a heavy load on a software sc nvidia is a prime example of it at WWZ)
remember how the power dropped like a stone from maxwell to pascal? 780ti has similiar tdp to navi also
Eli Kelly
Could've gotten the same perf/w back in 2016 with the GTX1080
Nvidia OBLITERATES AMD in WWZ now with a single driver release
John Phillips
kek you fell for raja's poor volta vega is trash, there is no defending it. v56 is mostly fine, but mediocre. v64 is too slow to command its msrp when the gtx1080 exists.
>With a single exception, there also aren’t any new graphics features. Navi does not include any hardware ray tracing support, nor does it support variable rate pixel shading
KILL YOURSELF FAGGOT :^)
Bentley Johnson
what is all this meme shittery you pretend to care about
Nathan Allen
That's ok user, all those are useless and shit. Next year when AMD finally starts to support them they'll be groundbreaking must-have features though.
Tyler Taylor
>+18% for 2080ti >+14% for 2070 >+12% for 2060
oh yeah they barely reach amd let alone beating them LOL
Caleb Carter
>doesnt support it
thats weird considering its a software feature i wonder if the drives support it he will just eat his hat off
Andrew Roberts
>225W HOUSEFIRES on 7nm finally reaching GTX 1080 performance the only valid point here, jensen.
Brandon Jenkins
If Poovi supported it AYYMD would have stated it in the PDF, it doesn't and it's not software, it's required to be supported in hardware
I'm more curious about their mid / low end tier gpus desu. I have a nvidia gtx1050ti and I want to see if amd can match it's performance with a new gpu.
Kayden Smith
are they even going to make a mid / low end?
Brayden Bailey
250mm2 used to be mid for $300-330.
Nicholas Kelly
what the fuck made you buy that over an rx 570? shitty prebuilt/low profile?
Parker Allen
was cheaper, the 1050ti costed 140 while the rx470/570 was almost 100 bucks more expensive IIRC. It's been so long but money definitely was the deciding factor.
jesus fucking christ Jow Forums doesnt have any sense of technology
Jason Flores
Because I would've had the performance your shit company is trying to sell me now for 3 years. You won't be recieving dividends from me Jow Forums amd nigger.
Camden Thompson
Navi does not support VRS and this has been confirmed by many tech websites
Can't get any simpler than that but you insist on arguing about it like a faggot
Kevin Lopez
Super SIMD when
Jack Cooper
navi LACKS software you moron the HARDWARE is there rasterization of pixel shaders exists since 2xxx
Jaxson Bailey
>As in, if Navi were on 16nm still at the same clocks as Vega, it'd have around a 30% improved perf/watt over Vega. They should have shoveled out tons of them cheap using 12nm or 16nm. A vega64 sized chip would probably be decent and still cheaper than 7nm.
Nicholas Jackson
>Could've gotten the same perf/w back in 2016 with the GTX1080 In reality the 1080 is 7% slower than the 2070, and the 5700 XT is 6% faster than the 2070. The 5700 XT is Vega 64 +15% perf.
Based. Keep dabbing on the AMDrones here, most everyone hates these cards.
Jack Collins
For real though, I think navi is gonna be pretty okay desu. This is coming from a GTX 1070 owner who wants to upgrade.
If we take a look at the cheaper $379 36 CU Rx 5700 you'll notice how it wipes the floor with the RTX 2060 by about 10% on average. Does that put it closer to the RTX 2060 or 2070?
Because it it gets really close to the 2070, say 95% as fast then it properly competes with it in my book. That's not taking the driver updates into account like OP said btw.
Personally I really look forward to essentially a ~doubling of FPS at 1440p with the 5700XT from my 1070 without having to pay the RTX tax of the 2080.
>hurrrrr double the framerate what is the actual improvement in frametimes though
Angel Wood
Also remember the $499 price tag was comfirmed fake, Rx 5700XT will be $449. That's a pretty sweet deal if it gets ~95% within RTX 2080 performance which retails for ~$800 on average.
I'll be honest here: I honestly EXPECT AMD to fuck up day 1 drivers but that's okay in my book too because this is what polaris and vega were like at launch. AMD does all the hard grinding with the hardware but then buffs everything out later with drivers.
I've been waiting 3 years for an upgrade and I don't mind waiting another few months after I get my 5700XT for frametime problems to be smoothed out (probably just power limit that can easily be fixed with slight UV).
Luis Taylor
>~95% within RTX 2080 more like Radeon VII
Hudson Martin
Which is 90% as fast as the 2080 2070 is 80% as fast as the 2080 see So throw 5% on top with driver updates and you have 95% 2080 performance in the end. Not bad.
Connor Gutierrez
>1070 user >upgrade to 5700 well, suit yourself. i don't see a reason to do so. 1070 is fine for 1440p (exactly my setup right now) but not high refresh rate
Brody Harris
the only game that runs bad on it right now is shadow of war on ultra everything else is pretty good at 1440 on 1070
David Hall
>The 5700 XT is Vega 64 +15% perf. At practically the same price of Vega 64. Horrible pricing. 349.99 would have been a lot better.
Joseph Harris
Nah, 5700XT. I have some IC diamond tubes laying around which I plan to use to lap the heatsink with and use as TIM to OC as high as possible. My 1070 only went up to 1720 MHz but maybe I'll get a lucky 100-200 MHz OC with navi.
Also seeing how big of a gap there is between the base and boost, there seems to be a lot of untapped performance on this little demon. Just needs more juice and better cooling, OC will just be icing on the cake.
That's what Nvidia were doing since maxwell. That's why Radeon were mostly slower when compared to Nvidia unless the game were optimized for the GCN arch.
Lucas Hughes
>7970 wasn't bad on launch, it was still better perf/$, perf/watt It was literally so bad at launch and getting its ass kicked by the 680 a few months after that AMD pushed out the factory OC ghz edition refresh that's exactly the WHYY AMD OVERVOLT THEIR CARDS, ITS OUT OF ITS COMFORT ZONE meme.
If you want to know how GCN should actually be clocked look no further than 7970 vs 7970 ghz edition
Michael Taylor
I see the slides, AMD didn't lied, this time they really changed the architecture. No wonder, despite the higher clocks, that 40CU is outperforming 56CU.
Nathaniel Sanders
I told you guys it was a wavefront saturation problem
Charles Carter
More like 20% if it outperforms the 2070 by 10% which by itself is 10% faster than V64. Right?
Jaxson Cox
navi is power locked like nvidia now. it won't go above 225w TDP spec.
better binned limited edition for $500, also it's 1980MGhz
Jaxson Foster
ah yes source is GN, they asked AMD directly about unlocked BIOS, it's all limited now I think it's a good think, retarded journos won't show you 500w power draws anymore
Carson Clark
Yeah, Navi is fucking epic. I bet nvidia doesn't even bother reacting.
Jaxson Diaz
compress shader quality to achiev higher fps.
Colton Lee
It is everywhere. That is really sad because Overclocking Radeon is really fun. Power tables, modded bios and all of it is now gone.
fuck that. nvidia already makes games look worse with aggressive culling
Evan Foster
The only thing wrong with Navi is the price.
Grayson Young
What about wattman's power limit offset thingie? Not even like 10%. I'm starting to reconsider navi desu. I just wanted rtx 2080 perf without the rtx price tag.
David Murphy
GTX 970 -398 mm2=$329 GTX 1070 -314 mm2=$379 RTX 2070 -445(120-130mm2 of it is cheap cheap tensor cores) mm2=$499
>thinking the yield would be 70% You people just blindly parrot made up bullshit from AdoredTV. The Zen2 chiplet won't even have yields that high, let alone a fucking 200mm2+ dense logic IC.
Mason Long
AMD goes full SIMT, this solved a lot unused capacity, and making better for future Raytracing hardware.
Brody Williams
Comming from a 970, how big of an upgrade would the 5700xt be?
Cameron Morgan
x2
Nolan Long
The 5700 XT is massively faster than Vega 56, and ~15% faster than Vega 64 which launched at $499.
Logan Rivera
There would be no reason to put an 8 pin plus 6 pin on the 5700 XT unless they were going to allow significant power offsets in wattman. They had to lock the BIOS for DRM reasons.
Robert Wright
its fucking jewish to raise price for increased performance between generations as this is expected to happen however as all the nvidiots cheered for that i say suits them right not getting the benefits of a price war right, there might be some massive OC headroom >75 Watts: None >150 Watts: One six-pin connector >225 Watts: Two six-pin connectors >300 Watts: One eight-pin + one six-pin connector >375 Watts: Two eight-pin connectors
Noah Stewart
>Using yield numbers from a literal scam artist who makes shit up to grift NEET bux on Patreon Lmoa
Hudson Stewart
>chiplets half the size of the previous one that had insane yields wont have the same and even more now..
Jow Forums ladies and gents
Asher Campbell
7nm is more expensive than 14nm.
Ryan White
>underage retard kid thinks he understands semicon industry 14LPP is not 7HPC, you retard. 14LPP is not an immersion process. TSMC's 7m node is. It has radically more lith masks, more multiexposure, complexity is exponentially higher. Die size alone does not determine defect density, you retarded little kid. TSMC's 7nm lines will never produce high yields.
Jace Sullivan
>Another clueless dumbfuck believing Scottish NEET Patreon bux shill lies Stop listing to people who make everything they say up to scam money out of dumbfucks like you.
Leo Wood
whats your point? anyway amd priced 5700XT and 5700 so badly that it looks like intentional sabotage, I Have no idea why they think they can be more greedy than fucking nvidia who are hated wide and far for their greedy margins
Brandon Watson
>>thinking the yield would be 70% First off that is the Worst case scenario, not best. all those components are taken at their highest possible prices Second, it's not AdoredTV dingus. >The Zen2 chiplet won't even have yields that high, let alone a fucking 200mm2+ dense logic IC. You have no idea what you are talking about. Then again maybe you do and just have stocks in AMD.
Jacob Parker
so? 970 was massively faster than 770 it didn't cost arm and leg 390 was barely faster than 290 they still asked 970 price plus premium.