• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

[rumour] AMD R9 390X, Nvidia GTX 980 Ti and Titan X Benchmarks Leaked

Renekton

Member
I feel embarrassed for the people getting hype over WCCFTech news with fake slides and benchmarks. Videocardz.com is right there with them. Making shit up and not crediting the source of actual viable news they steal from people until they get called out for it.
One correction: they usually do mention the source (with link).

People here know those are rumors.
 

mhayze

Member
Anybody got any ideas as to how much a GM200 cut down would cost?

The #1 factor will probably be the pricing of a competitive product from AMD. If the '980ti' is the cut-down GM200, and it's 10% slower than a $499 390X (completely making up the numbers here), then I can see Nvidia releasing it for $499 as well, or $550 if it has more RAM (like 6GB vs a 4GB 390X)

If on the other hand, AMD releases the 390X at $599, and the perf is faster or really close on the Nvidia side, I wouldn't be surprised by a $650 launch price.

The #2 factor is yield. If these chips are being made in large quantities as surplus from some supercomputer order, I can see that easing pricing a little.

If I had to guess, I would say $599 - $699 at launch with $649 being most likely for a 6GB card, or $599 if it's less RAM.
 
I don't believe the graphs to be true.. 390x should be even faster ;#

Yes, I'm of the same opinion.

Judging from the performance delta between a 290 and 290X, the 290X is only very very rarely shader or TMU bound. So basing performance off of the amount of extra CUs this new GPU has isn't a good idea. Which is what I think these "leaks" are doing.
 

LiquidMetal14

hide your water-based mammals
If AMD has their cooling and power efficiency in check then I will highly consider two of those over the ridiculously expensive alternative nvidia is offering. I'm not biased when it comes to so many things and don't hands my future on something like gsync. I would want about 8gb of memory though.
 
So it's basically a repeat of what happened with the Titan and the 780 Ti? If these numbers hold up in official reviews, then I can't say I'm surprised.
 

El_Chino

Member
There have been rumors that the 960ti will have 4GB of VRAM with a 128MB bus. But it is possible that it could be changed to 3GB, or at least a boost version.
Well considering evga and others have just released 4GB versions of the current 960 it would be a bit odd if it was 3GB
 
As someone who frequently switches between the 2 GPU vendors I want to say a few things:
  • AMD and nVidia essentially have had the same quality of drivers for the past....3 years or so now.
  • nVidia does offer more extra features such as PhysX, G-SYNC, 3D etc.
  • nVidia has been amazingly power efficient for the past 2 years.
I am the owner of GTX 780s in SLI but I don't plan on buying another nVidia card for a while. I can't believe they dropped optimization of the Kepler line as soon as Maxwell was launched. Recent benchmarks puts AMD 7970 Ghz within 10% of a GTX 780 which itself is trailing the 290/290X in games. Extremely disappointed man.

Heh, this used to be amd's shtick
I remember they dropped my 9800 pro like a hot potato when their new line of gpus launched (2 years after I bought my 9800 pro vista released and they never even made a vista compatible driver!)
My 4870 was forgotten about in optimisation 6 months after the 5870 was released.

I guess amd staying with GCN for so long (with 280x etc being rebrands of the 7970) it also meant that the 7970 line kept seeing performance improvements with driver updates for much longer.

It sucks to hear that nvidia are just as bad as amd used to be when it comes to this... (fuck nvidia so much recently, seriously)


On topic: so are nvidia coming out with a 980ti (cut down titan x) alongside the titan x then?
That would be a direct consequence of the amd 390 series' specs then, because they were firmly on track for repeating their gtx 680-> titan -> 780 bullshit all over again.
I was expecting the midrange 980 followed by high end 'fuck you' titan brand and price 6 months later, followed by the actual high end 980 releasing another 5-6 months after the titan x and being called 1080.
This stretched out release schedule really hurts the consumer and general progress in gpu performance.

If amd throws a wrench in that release scheme that is really good for us consumers and gamers as it would put release schedules back on track.
A delayed gm 200 consumer (not titan) price gpu has a ripple effect over the coming years with all subsequent releases being delayed by 6 months, just like gk 110's release delay did.

I'm rooting for AMD with the 390 and 390x.
I was unhappy with the drivers of my hd4870 and 6870 and the lack of features, but at least they've finally sorted the feature issue out with downsampling support on driver level (superior to nvidia's downsampling too from what I've heard).
My other big issue was them dropping gpu driver support really fast (but apparently nvidia is just as bad with this now... shame on nvidia)

A repeat of the insulting gtx 680 ->titan naming scheme combined with the recent false advertising followed by the equally insulting non apology for it have really soured me on the idea of supporting nvidia.

All I ask for is a competent AMD causing proper competition.
It looks like it may happen now (it all depends on the 390 and 390x prices I guess,)
It was starting to look like we were going to go the intel vs amd cpu way in gpu land when the tonka toy amd 285 released. I'm relieved that they have something proper to show now.
 
Well considering evga and others have just released 4GB versions of the current 960 it would be a bit odd if it was 3GB

They have? I am off the ball on this one. That has still got to be some sort of Boost model though, as the performance falls between the 960 and 970.
 
On topic: so are nvidia coming out with a 980ti (cut down titan x) alongside the titan x then?

That wouldn't leave them with anywhere to go for the 1000 series later on in the year. Unless they're prematurely launching it with this slightly cut down Titan X, which is possible, but not all that likely. The Titan is coming out in like <1 week, the 390x is probably a month or two away. Leaving AMD in the lead for a couple of months until the 7th or 8th month of the year wouldn't be out of character.

The probability that these benches are legit and they have access to a card 6 months away from launch is pretty low. Like we needed more doubts about it the source.
 
There have been rumors that the 960ti will have 4GB of VRAM with a 128MB bus. But it is possible that it could be changed to 3GB, or at least a boost version.

That is normal 960 with 4GB vram

Is anyone else wondering what that GTX 9** 3GB is?

Probabably the long rumored 960ti based on the more cut down version of 970

I don't believe the graphs to be true.. 390x should be even faster ;)

There's more to GPU than bandwidth

I am the owner of GTX 780s in SLI but I don't plan on buying another nVidia card for a while. I can't believe they dropped optimization of the Kepler line as soon as Maxwell was launched. Recent benchmarks puts AMD 7970 Ghz within 10% of a GTX 780 which itself is trailing the 290/290X in games. Extremely disappointed man.

I hope You do realise that "unsupported" GTX 770 has been faster in some recent games than 290 - it's a question of engine used and you are probably thinking of Evolve where Cry-engine is heavily optimised for GCN.
 
780Ti was the opposite of cut-down I think?

Titan had some SMXs disabled so 780Ti (then Titan Black) got them all running to put the 290X in its place.
well a 980 ti would be similar to the 780 (non-ti) in what it is (big die maxwell, but a bit cut down from the titan version)
They can hardly call it a 1080 so soon after the 980 launched hence why I called it 980ti

In the end the name doesn't matter, all that matters is if it's a proper high end not crippled big die gpu and if it's priced correctly for that bracket, they could call it the 'smellmyfarts' for all I care (exactly the same thing as calling gk104 gtx 680 and gm 204 gtx 980, the name has no relevance to what the hardware is anymore:p)

That wouldn't leave them with anywhere to go for the 1000 series later on in the year. Unless they're prematurely launching it with this slightly cut down Titan X, which is possible, but not all that likely. The Titan is coming out in like <1 week, the 390x is probably a month or two away. Leaving AMD in the lead for a couple of months until the 7th or 8th month of the year wouldn't be out of character.

The probability that these benches are legit and they have access to a card 6 months away from launch is pretty low. Like we needed more doubts about it the source.

:( I just saw a mention of a cut down (non titan) gm200 in those benchmarks and was hoping that meant the gm200 nvidia consumer card (whatever they call it) would be coming out alongside the titan.

There have been tons of leaks and rumors about the 390-390x and their specs right? (and performance could be inferred from those specs) so even if (when) these benchmarks turn out false the 390-390x results shown elsewhere still stand I hope.

But yeah if you assume they (nvidia) won't do that (gm 200 card alongside titan) then this rumor is highly unlikely =(
 
There have been tons of leaks and rumors about the 390-390x and their specs right? (and performance could be inferred from those specs) so even if (when) these benchmarks turn out false the 390-390x results shown elsewhere still stand I hope.

But yeah if you assume they (nvidia) won't do that (gm 200 card alongside titan) then this rumor is highly unlikely =(

There have been a lot of leaks/rumours but they don't all line up perfectly with each other. This exact bench was posted earlier on GAF in some other thread, iirc. Best just to wait for official confirmation or at least reputable sources before getting overhyped.
 
There have been a lot of leaks/rumours but they don't all line up perfectly with each other. This exact bench was posted earlier on GAF in some other thread, iirc. Best just to wait for official confirmation or at least reputable sources before getting overhyped.

Oh don't worry I keep my expectations in check.

Just some good tidings would be nice, ya know.
Those tonga rumors were depressing (and turned out to be true in the end), I like positive rumors better.

These rumors sound a lot better than the skylake rumors.
 
Oh don't worry I keep my expectations in check.

Just some good tidings would be nice, ya know.
Those tonga rumors were depressing (and turned out to be true in the end), I like positive rumors better.

These rumors sound a lot better than the skylake rumors.

What are the rumors about Skylake?
 
Only a 20 percent cpu performance increase at best, both due to a lower TDP (seriously WHY) and once again more die space to the iGPU (which anyone buying a k series skylake is not going to care about)
edit: misremembered: the lower TDP part of the rumors is for broadwell
The performance increase is really poor considering it's a new architecture and is on a new process node.

If true then the performance/dollar gains between late 2015 and 2011 (sandy bridge) are going to remain pathetic:\

At least there is still supposed to be a k processor and they won't be soldered onto the motherboard (like rumors 1 year ago suggested for broadwell and skylake), that would have thoroughly sucked.

Of course skylake is still rumored to just come in 2 and 4 core options... 6-8 cores will still be kept out of the mainstream just because (there is no competition), until you eventually get a skylake-e for a thousand bucks a year down the line...

edit: and another rumor is on die edram (that takes up a bunch of die space, making it costly) for the iGPU, which again is a complete waste of die space for anyone who cares about IPC and not about some iGPU they will never use.
When you buy 300 euros worth of die + a brand new motherboard (because ofcourse they change the socket again) and half of it is thrown away on iGPU that you don't need it's pretty dire.

tldr: it looks like intel is going full mobile direction and can't be bothered to release a performance desktop class die for gamers/power users.

Imo the best thing we can hope for is that VR's refresh rate requirements will put some pressure on intel to release more capable cpus in the future. It's a pretty sad situation
 

mrklaw

MrArseFace
I'm curious about power, heat and noise. My 7950 was hideously noisy and hot, whereas my 970 is cool and quiet. I can't see myself switch back to AMD unless their cards are way quieter.
 
I'll be surprised if there's even a 20% CPU perf increase with Skylake. We should all know by now that Intel are going for efficiency over pure power gains and they have been doing so for a goid number of years.
 
I think "20% improvements" is going to be the new normal, Stephan. Even if this thread's rumor is correct, the "high end" AMD card taking advantage of the brand new high bandwidth memory technology is a mere 25% improvement over a "mid range" card from 6 months ago. And it's more like 20% from 980 -> Titan X according to those graphs. The 600,700,900 transitions seem roughly in line with this.

CPU stuff has stagnated years ago and GPU tech just isn't improving as fast as it used to be. If you do the math, a 20% Year on Year improvement for each generation yields a 5 year total improvement of 2.5x the original card's power. 3.5x after 7 years. That's pathetic and represents a clear trend of slowing GPU tech.
 
I'll be surprised if there's even a 20% CPU perf increase with Skylake. We should all know by now that Intel are going for efficiency over pure power gains and they have been doing so for a goid number of years.

The two are not mutually exclusive (never have been)

efficiency has always gone up...
It was always going to be more power efficient , they are just choosing to sell you smaller dies (or more die space to an iGPU) for the same amount of money. Nothing more, nothing less.


If you asked anyone in 2011 (after sandy bridge released) what 250$ cpus would be like in 2015 they would have said 6-8 core cpus with much better IPC and would be pretty appaled by the current situation.

(also if you warned me abou this in 2011 I would have sold my phenom II and bought an i5 2600k immediately, knowing it's the best I was going to get for at least 5-6 years at that price)

I think "20% improvements" is going to be the new normal, Stephan. Even if this thread's rumor is correct, the "high end" AMD card taking advantage of the brand new high bandwidth memory technology is a mere 25% improvement over a "mid range" card from 6 months ago. And it's more like 20% from 980 -> Titan X according to those graphs. The 600,700,900 transitions seem roughly in line with this.

CPU stuff has stagnated years ago and GPU tech just isn't improving as fast as it used to be. If you do the math, a 20% Year on Year improvement for each generation yields a 5 year total improvement of 2.5x the original card's power. 3.5x after 7 years. That's pathetic and represents a clear trend of slowing GPU tech.
Except northern islands -> GCN was a far greater jump than 20 percent

The limited jump from the 980 to the 390x is understandable as nvidia seem to have the better architecture here (it's still supposed to be a 40-50 percent jump from the 290x, right, or did that rumor change again) and both maxwell and the new amd gpus are stuck on 28nm.

Despite being stuck at 28 nm, there has been a much bigger jump between early 2011 and early 2015 (3x performance in 4 years, not 3.5 in 7 years) and we would have had our proper full jump if 20nm fabs were ready.
The problem in gpu land is not that high end gpus are no longer being made, they very much are (500-600mm^2 dies with 8 billion transistors, now 1024bit busses and massive bandwidth increases incoming).
The problem in gpu land is that prices have doubled (because fuck consumers we'll still buy it anyhow, right nvidia?)

That situation is still infinitely preferable to the cpu one where not only do we not get high end cpus, we don't get any gains at all..
Intel is many die shrinks ahead of the fabs amd and nvidia has and they've had many architecture improvements to enable more performance/watt and more performance/die space.

But instead of seeing a transistor budget growth we just keep getting smaller dies and more of that die used on iGPU rather than cpu.
Core i7 (Quad) 731,000,000 2008 Intel 45 nm 263 mm² (i7 920)
Quad-core + GPU Core i7 1,160,000,000 2011 Intel 32 nm 216 mm² (sandy bridge)
Quad-core + GPU Core i7 1,400,000,000 2012 Intel 22 nm 160 mm² (ivy)
Quad-core+GPU Core i7 1,400;000,000 2014 Intel 22nm finfet 177 mm² (haswell)

See the pattern here?

Smaller dies, more of it dedicated to igpu negating any performance benifits from new architectures and die shrinks.
From the rumored 20 percent IPC increase + edram + 14nm intel fabrication process we can probably expect a similar or smaller die size compared to ivy/haswell, and all the extra transistor budget thrown away on the EDRAM for their precious igpu
If there is no edram then expect a smaller(cheaper) die, at the same price of course.

edit: rumor is the base clocks will be higher so there migth not even BE any IPC increase... in which case definitely an even smaller die.

And ofcourse they don't want to give you a full cpu dedicated die as the performance of it would (rightfully) make their apus looks like garbage in comparison...

The cpu situation is very different to the gpu one and it is much much worse.
The problem isn't just that the quad core skylake from the rumors should've been the celeron by now , but that there in fact IS no i5-i7 performance cpu.
 
Why can't they just make more efficient versions of the current cards they have? and not just keep dropping new unrefined tech that never gets taken advantage of?

If the current set up I have right now, which is pretty much mid range beyond maxes out every game at 2K, I would just be pissing money out the window upgrading anything.

I mean they refine the shit in consoles for 5-6 years until they are unable to get more out of them, where as it feels like my GPU and CPU are hardly getting used?

I don't know, all seems like redundant shit to me.
 
I'm curious about power, heat and noise. My 7950 was hideously noisy and hot, whereas my 970 is cool and quiet. I can't see myself switch back to AMD unless their cards are way quieter.

This is more a function of the cooler and not AMD or Nvidia, unless you have a reference card (there are no reference 970s though, so).
 
This is the first time I've heard of the Skylake.
Is it going to be the next generation of intel CPU past i7 series?

It's the name of the architecture.

2000 series - Sandy Bridge
3000 series - Ivy Bridges
4000 series - Haswell
5000 series - Broadwell
6000 series - Skylake
7000 series - Cannonlake

Will still most likely have the i3/i5/i7 split as usual.
 

dumbo

Member
CPU stuff has stagnated years ago and GPU tech just isn't improving as fast as it used to be. If you do the math, a 20% Year on Year improvement for each generation yields a 5 year total improvement of 2.5x the original card's power. 3.5x after 7 years. That's pathetic and represents a clear trend of slowing GPU tech.

Big advances typically tend to come from new "process nodes".
In 2007 the first 65nm GPU appeared.
In 2008 it was 55nm.
In 2009 it was 40nm.
In 2011 it was 28nm.

In 2015, it's still 28nm :(.
 

DieH@rd

Banned
With the arrival of low-level multicore-friendly graphical APIs on PC market, I have zero wish to upgrade my FX8350. It will remain highly viable for years and years.

Hopefully majority of the AAA games will adopt DX12.
 
I hope Skylake sees hex core CPUs on a non-enthusiast socket, but I'm not holding my breath.

Nope - not on currently leaked roadmaps. I want one so I was collecting rumors and at the moment we are looking at:

Broadwell-K release - around June, IPC increase below <5% + unknown impact of L4 cache from Iris pro on cpu operations, unknown frequency, 65W TDP

Will be supported by most of not all Z97/H97 mobos.

//I'm afraid it might clock low and come with higher than normal price tag due to Iris Pro

Skylake is coming after Broadwell but we don't know if 2015 releases include K- SKU

on positive side K-SKU has 88 or 95W tdp so it will be normal cpu.

There was very recent leak of engineering sample that puts IPC gains at 10% vs Haswell single threaded and 20% multi threaded.

Rumors suggest DDR4 and low voltage DDR3 support.



And if someone cares latest rumour is Broadwell-E in early 2016.
 

AllenShrz

Member
As an AMD user for years, I find this claims of terrible drivers or heat problems laughable, and whenever I buy a new one I always buy the high end-single core.
 
I hope Skylake sees hex core CPUs on a non-enthusiast socket, but I'm not holding my breath.

As long as people are buying their quad cores at 250-350 euros it's not going to happen.. (I seriously wonder who keeps buying them, how can there be any demand when there is no noticable upgrade from any 4 year old intel cpu, demand WILL slow down for this shit eventually, it has to)

edit; if you give it some thought it's not even funny what a scam the newer intel cpus are...
Why do people upgrade these days? Not for cpu performance but because they want new motherboard features (a new sata standard for newer faster SSDs that saturate old sata ports, more pcie lanes for newer higher end gpus, usb 3 for their external hdd or digital camera etc etc)
But instead of offering you a sandy bridge mobo with those features you're forced to get a z97 mobo , which means a new cpu... which is going to do nothing for you that your old cpu wasn't already doing.

It's like the reverse of the previous situation, where people wanted that cpu performance and were forced to buy a motherboard they didn't want because the socket changed every 6 months.

It still changes, just for different reasons, heh...

That's the cost of using a PC these days... you have to give money to a terrible pseudo duopoly of anti consumer hardware manufacturers (who have spent the last few decades pulling their competition out by the roots with illegal anti competitive measures, while writing off the massive court fines for them off as operating costs/investments into controlling the market) to get access to a pro consumer software platform.
It's like punching a bag of kittens just to keep your local animal sanctuary open. And it feels just as bad.
 
Top Bottom