Fury X vs. Titan X benchmark

Status
Not open for further replies.
Brent_Justice said:
It is up to AMD and NVIDIA to provide the kind of experience they want out of their cards to gamers. It is our job to evaluate that experience. We take that, at its default, and evaluate the gameplay experience. If we were to change anything in the driver we would be artificially changing that experience and creating an unfair comparison.

To keep it fair, we leave the driver settings alone, and at default, on both AMD and NVIDIA. If we detect some measure of ill looking image quality we bring this to the attention of AMD and NVIDIA.
Click to expand...

so if amd make the tessellation level switch say 16x by default and tessellation level controlled by game the option
you would be ok with that and review that way ?
 
KENNYB said:
This definitely use to be true for ATI. I recall back in the 9700/9500 days, I could definitely see a difference, and I always preferred ATI. Both companies were fooling around with optimizations, and still are of course. I preferred what ATI was doing over whatever the hell nVidia was doing. I doubt this is still true.

I say that having owned many nVidia cards.
Click to expand...

This was the only time it was true, was back in the NVidia GeForce FX fail days as NVidia was doing a lot of driver trickery and cheating and processing everything at FP16 vs. FP24/FP32 as the SM2.0 standard dictated. The cards only supported 16 or 32 and doing the shaders at FP32 caused a massive perf. hit. Meanwhile ATI's cards from that generation just did everything at FP24 inline with the spec. The problem is that ever since this people still like to bring up that ATI's hardware has 'better IQ'. It doesn't, because NVidia is now processing everything at the same standard.

I'm not sure about the picture on the first page in question. All I will say is that no review site has mentioned it, and if it is indeed happening I would venture to say the person in question forgot to adjust the default texture filtering setting in the NVCP. By default NVidia's drivers are set to a lower quality texture filtering setting. One of the first things I do whenever I update/install NVidia drivers is always adjust that setting to high quality.
 
it wasn't even fp 16, fp 16 still should give good results, the pixelation was of the textures/shaders was do to int 8. Once the issue was fixed after the 6800 was released, fp 16 was being used properly.

Any case the only person that seems to have the issue is the actually guy that took the screenshots and videos in the first place, default settings are set to application, so I think he might have had it set to AF x0 in the driver and forgot about it when he was doing his first video and screenshots.
 
razor1 said:
it wasn't even fp 16, fp 16 still should give good results, the pixelation was of the textures/shaders was do to int 8. Once the issue was fixed after the 6800 was released, fp 16 was being used properly.

Any case the only person that seems to have the issue is the actually guy that took the screenshots and videos in the first place, default settings are set to application, so I think he might have had it set to AF x0 in the driver and forgot about it when he was doing his first video and screenshots.
Click to expand...

Yes, i'm sure he "forgot"

Not even sure why people would fool around with AF these days. The perf. hit is so small that any benchmarking being done should just be at 16x high quality AF period.
 
Wow, this thread took an incredible turn, I can't believe I just read all of it...

I was going to ask the OP to test Project Cars... but since I'm here,
I can see this being a quality priority over speed thing for AMD, just like in the ATI days where their trilinear and aniso filtering were better than Nvidia. But, I find this interesting because, I for one, turn everything on in the drivers, as my default, except for AA. AFx16, Transparency AA 8x, TF Qual HQ, Negative LOD Clamp, same thing if I'm running AMD hardware, CCC is going max everything.

I thought everyone here did that? I want the drivers as optimized as possible, in game settings can take care of the rest. The performance hit isnt something even noticeable. Poor aniso, is.
 
WorldExclusive said:
You're still asking to go from stock settings review, to a variable settings review.
There's no way to set both control panels to render the same quality level to accurately come to a conclusion which card is faster.
Both sides will continue to cry that the AA was too high, or Triple Buffering should be turned off etc.
Click to expand...

The point is to remove as many optimizations from both sides as possible that come at the cost of IQ, not to enable extra features like AA driver side. Triple buffering should be the same between camps, and to do otherwise would not be an apples-to-apples comparison at all. To think it's acceptable to have triple buffering enabled for one vendor and not the other is absurd (this is what I'm assuming you are arguing for, if not, then your comment about triple buffering is irrelevant).

It's pretty darn obvious that in terms of settings like texture filtering quality, vendors could be (and most likely are) aiming for different targets for every setting *except* the maximum quality options, which in all reasonable cases should be aiming for just that -- maximal quality. Any optimizations at this level should be incredibly hard for the human eye to detect.

Another thing to note is that as enthusiasts, it's not like anyone posting in this thread is running texture filtering at anything lower than the highest quality setting.

Rather than rely on defaults, which could be tuned for wildly different scenarios, what makes the most sense is to go for the maximum quality options, which one would expect to be tuned for the same scenario across vendors. Extra "add on" IQ features like special AA modes should remain disabled. This seems like common sense, but appears to be alluding many posters here.

While I've seen many people commenting that they favour defaults and asserting this is the most "fair" (or alternatively that anything else would be "unfair"), I haven't seen one person attempt to explain the logic behind this position. The logic behind favouring maximum quality on the other hand, is very clear -- that maximum quality should have the least difference between vendors as we can reasonably expected them to aim for very similar targets at these settings.

With all that in mind, I've seen some conflicting evidence that the default Nvidia settings may not actually look like that. I'll give Nvidia (and the [H] results) benefit of doubt for now, but either way the methodology should be changed in the future to something more scientific than "We'll just leave it at default and hope that both vendors optimize those settings comparably."
 
Last edited:
The logic behind using the default is very clear. It's what is recommended by the manufacturer for the card. It is that simple. You say just set it all to max, but that won't work because setting everything to max will tank most cards performance in ways that will make the arguments over what is on or off worse than what is in this thread. To put it another way. If we need to put it all to max, then we need to set tessellation levels to max cause it wouldn't be "fair" that Nvidia doesn't provide for a "tessellation limiter" in their driver to match AMD's driver. If that's true then AMD cards would completely tank in performance and we would get the same arguments that it wasn't a fair test, even though we are advocating setting drivers to "Max". It's better to let both manufacturers simply set up the default settings, benchmark it with screenshots and let the reader decide. That way the answer to any argument is "contact manufacturer about their recommended default settings if you want to complain about some setting being a problem".
 
Interesting read... More things change, the more they remain the same.

So with no overseeing body or agreements between corporations on any set of guidelines; default settings are completely arbitrary and don't mean anything really.

nvidia defaults more towards performance, AMD more towards quality.
 
LurkerLito said:
The logic behind using the default is very clear. It's what is recommended by the manufacturer for the card. It is that simple. You say just set it all to max, but that won't work because setting everything to max will tank most cards performance in ways that will make the arguments over what is on or off worse than what is in this thread. To put it another way. If we need to put it all to max, then we need to set tessellation levels to max cause it wouldn't be "fair" that Nvidia doesn't provide for a "tessellation limiter" in their driver to match AMD's driver. If that's true then AMD cards would completely tank in performance and we would get the same arguments that it wasn't a fair test, even though we are advocating setting drivers to "Max". It's better to let both manufacturers simply set up the default settings, benchmark it with screenshots and let the reader decide. That way the answer to any argument is "contact manufacturer about their recommended default settings if you want to complain about some setting being a problem".
Click to expand...
since the "tessellation limiter"is off by default and not used by reviewers, your whole point is meaningless
 
Osjur said:
-In the past ATI had better VGA quality because they used better RAMDAC
-They started using 10bit internal LUT from 5xxx series onwards which gives you better monitor picture quality with calibrated monitors because you get less banding. Nvidia still uses 8bit LUT on their consumer cards.
-In 2014 they upped the internal LUT to 12bit, meaning even less banding with calibrated monitors. I can now calibrate my 10bit Direct Drive HP ZR30w monitor (no internal LUT / scaling board) and get no banding at all.
-And now it seems they even have better texture quality at default settings.
Click to expand...

This. Seems most don't know or care about 10 bit or better support, but for users of those panel types, especially for productivity, it's a valuable addition when considering AMD. These days, I notice banding more than used to. When it's virtually eliminated in day to day use, it sticks out like a sore thumb.

Throw this default shenanigan IQ/performance differences on the fire too, red team seems even more appealing this round.

Would love Hardocp to start including image quality comparisons again. And not stop including them, or Nvidia will do some more greasy, questionable crap like usual.
 
Mchart said:
By default NVidia's drivers are set to a lower quality texture filtering setting. One of the first things I do whenever I update/install NVidia drivers is always adjust that setting to high quality.
Click to expand...

Well i'm sure a lot of gamers would manually set it for HQ or 16x as a best practice, but the point here is that reviewers leave it on default for benchmarks.
 
N4CR said:
This. Seems most don't know or care about 10 bit or better support, but for users of those panel types, especially for productivity, it's a valuable addition when considering AMD. These days, I notice banding more than used to. When it's virtually eliminated in day to day use, it sticks out like a sore thumb.

Throw this default shenanigan IQ/performance differences on the fire too, red team seems even more appealing this round.

Would love Hardocp to start including image quality comparisons again. And not stop including them, or Nvidia will do some more greasy, questionable crap like usual.
Click to expand...

Are you sure NV doesn't support 10bit? There is an option to change the color depth in the NV control panel on Maxwell cards at least. My monitor only supports 8, but here is a screenshot showing 10:

10bpp.png
 
x3sphere said:
Are you sure NV doesn't support 10bit? There is an option to change the color depth in the NV control panel on Maxwell cards at least. My monitor only supports 8, but here is a screenshot showing 10:
Click to expand...

You are correct, however 10 bit is only supported in Direct X, as per Nvidia, confirmed in 2015 by CS, including Maxwell/Titan.

http://nvidia.custhelp.com/app/answ...-bit-per-color-support-on-nvidia-geforce-gpus

NVIDIA Geforce graphics cards have offered 10-bit per color out to a full screen Direct X surface since the Geforce 200 series GPUs. Due to the way most applications use traditional Windows API functions to create the application UI and viewport display, this method is not used for professional applications such as Adobe Premiere Pro and Adobe Photoshop. These programs use OpenGL 10-bit per color buffers which require an NVIDIA Quadro GPU with DisplayPort connector. A small number of monitors support 10-bit per color with Quadro graphics cards over DVI.
Click to expand...
 
N4CR said:
This. Seems most don't know or care about 10 bit or better support, but for users of those panel types, especially for productivity, it's a valuable addition when considering AMD. These days, I notice banding more than used to. When it's virtually eliminated in day to day use, it sticks out like a sore thumb.

Throw this default shenanigan IQ/performance differences on the fire too, red team seems even more appealing this round.

Would love Hardocp to start including image quality comparisons again. And not stop including them, or Nvidia will do some more greasy, questionable crap like usual.
Click to expand...

Because 10bit gaming is so popular right now. :rolleyes:

If true 10bit color is needed, it's more likely I will have a workstation card paired with a true 10bit monitor. Both of which are awful for gaming.

If you're judging consumer cards based on workstation features, I can see why you're POV is terribly skewed.
 
Bowman15 said:
Did AMD kick your dog, get your sister pregnant or are you just a fan boy?
Click to expand...

I'm guessing got the sister pregnant. Those AMD Marketing peeps are smooth customers. Slicker than cum on a gold tooth if you will. They said overclocking beast and it was panties around the ankles for sure!
 
RAutrey said:
I'm guessing got the sister pregnant. Those AMD Marketing peeps are smooth customers. Slicker than cum on a gold tooth if you will. They said overclocking beast and it was panties around the ankles for sure!
Click to expand...

Do you mean how its currently voltage locked and we've already seen the memory can be overclocked as well? Voltage unlock will be coming. Whether it will be a great overclocker has yet to be determined.
 
Yep. I actually think there is a lot yet to be determined about this card.
 
Bowman15 said:
You words mean absolutely nothing if you don't know how many cards there actually were on release. My point is you are just pulling numbers out of your ass to justify your own argument as you do not know. Did AMD kick your dog, get your sister pregnant or are you just a fan boy?
Click to expand...


When AMD says they have low supply they have low supply, its pretty easy to understand that isn't it?

Why would AMD admit they would have low supply and you are saying something to the contrary? That means with regular amounts of cards being produced it wouldn't have been supply constrained, either way it means they aren't selling. The Fury cards aren't going to make anything for AMD as far as marketshare. Niether does the 980ti or the Titan X for nV, its the mid range cards that they need, and they are getting blown out of the water, performance is good for AMD's midrange but power consumption sucks, and mid range PCs this is much more important then high end PCs. AMD will get some marketshare back from nV because at least they are in the competition again to some degree.

It doesn't matter what the number is, the first batch of cards were supply constrained,

Bowman, you should take your own post to heart.
 
Bowman15 said:
Do you mean how its currently voltage locked and we've already seen the memory can be overclocked as well? Voltage unlock will be coming. Whether it will be a great overclocker has yet to be determined.
Click to expand...

Care to show us where the memory has been overclocked without hardware/BIOS modding and that it improves anything?
Memory overclocking has been debunked by AMD as not possible.

A fair amount of optimism is required for the voltage to be unlocked and that it will allow much.
The isolation of the VRMs from cooling, even from natural airflow, prevents them from taking much more.
If the unlock surfaces I cant see it allowing more than a tiny voltage increase.
 
bill_d said:
since the "tessellation limiter"is off by default and not used by reviewers, your whole point is meaningless
Click to expand...

Read the post previous mine that I am replying to, they are advocating Max Settings. So yes the point is completely valid. By setting Nv to max settings means Tessellation is set on and is at a relatively high level. So if you want to make it "fair" then AMD's Tessellation setting should be set to max also which will tank their performance even more. You understand now? This doesn't only apply to the "tessellation limiter" it also applies to any new settings and fx that might be introduced over time with new cards. Using "max" settings as part of the main review is a BAD idea. That's why using, manufacturer default in the drivers is the best option.
 
Bowman15 said:
Do you mean how its currently voltage locked and we've already seen the memory can be overclocked as well? Voltage unlock will be coming. Whether it will be a great overclocker has yet to be determined.
Click to expand...


And that was stupid marketing on AMD's part, saying FuryX will overclock like a dream and well nothing to show for it......

That's the type of hype that kills AMD. What if when they finally unlock the voltages for the GPU and memory we find out that you have to over volt Fury X to overclock? Will that be good? Its going to suck so much more power, it might actually be ridiculous amounts of power.

Edit:

What would be interesting is when we see the air cooled version of the cut down Fury and it uses around the same power as the Fury X, then we will know that Fury X's water cooler wasn't a nice add on, it will be at the point noted as a necessity to remain somewhat competitive still is lacking.
 
Last edited:
Nenu said:
Care to show us where the memory has been overclocked without hardware/BIOS modding and that it improves anything?
Memory overclocking has been debunked by AMD as not possible.

A fair amount of optimism is required for the voltage to be unlocked and that it will allow much.
The isolation of the VRMs from cooling, even from natural airflow, prevents them from taking much more.
If the unlock surfaces I cant see it allowing more than a tiny voltage increase.
Click to expand...

You must not read anything outside of [H]forum. There are a number of posters everywhere talking about the OC mem bug in CCC, where ever so often when you boot it is available. With those that have tried, a 9%core 20%mem resulted in 20% firestrike score increases.
 
Nenu said:
Care to show us where the memory has been overclocked without hardware/BIOS modding and that it improves anything?
Memory overclocking has been debunked by AMD as not possible.

A fair amount of optimism is required for the voltage to be unlocked and that it will allow much.
The isolation of the VRMs from cooling, even from natural airflow, prevents them from taking much more.
If the unlock surfaces I cant see it allowing more than a tiny voltage increase.
Click to expand...

Google it...
 
durquavian said:
You must not read anything outside of [H]forum. There are a number of posters everywhere talking about the OC mem bug in CCC, where ever so often when you boot it is available. With those that have tried, a 9%core 20%mem resulted in 20% firestrike score increases.
Click to expand...

No, they'd rather sit in here and shit on it rather than see what the rest of the Internet is up to. Apparently HardOCP is the only place you can check out Fury scores, tests and overclocking. Fury can't do 4K, can't overclock, can't unlock memory and apparently there were only about 100 cards for all of North America. The green defense team would rather distort reality and continue the hyperbole. This forum is becoming a joke to other forums too point and laugh at. If AMD fan boys were doing the same thing to NV I'd be calling that crap as well.
 
Last edited:
razor1 said:
And that was stupid marketing on AMD's part, saying FuryX will overclock like a dream and well nothing to show for it......

That's the type of hype that kills AMD. What if when they finally unlock the voltages for the GPU and memory we find out that you have to over volt Fury X to overclock? Will that be good? Its going to suck so much more power, it might actually be ridiculous amounts of power.

Edit:

What would be interesting is when we see the air cooled version of the cut down Fury and it uses around the same power as the Fury X, then we will know that Fury X's water cooler wasn't a nice add on, it will be at the point noted as a necessity to remain somewhat competitive still is lacking.
Click to expand...

Air cooled version is a cut back chip. Think 780ti / 780. So no comparison can really be made.
 
Some here could do some testing on this, with Nvidia cards. Heard a few saying this is the case with default driver settings and is noticeable. Nvidia in the past stated their driver optimizations would not noticeably degrade IQ. How about some real user testing.

Since BF4 is a very popular game and benchmark game - has Nvidia profile this game to do this for better performance? How about some testing of Nvidia performance with different quality settings in Nvidia Control panel? To see if there is really any significant difference.

In the past AMD cheated on AF giving non variant angle AF during screen shots but in motion it was degraded :D. I believe this was a real cheat.

Nvidia had some drivers with AA not doing what was selected (driver bug?)

Anyways the BF4 images looks to be a more positive lod setting pushing up the mipmap textures (smaller textures) closer to view. If the case this would alleviate memory bandwidth required on texture processing. This would have a bigger impact over AF.

As for Fury CFX performance at 4K the initial testing is looking rather good - good for AMD.
 
Bowman15 said:
And the ignorance continues. That second part is ironic.

https://www.google.com/search?q=can+fury+hbm+be+overclocked&ie=utf-8&oe=utf-8
Click to expand...

Your description of yourself is amusing.

http://wccftech.com/amd-radeon-r9-fury-memory-oveclocked-20/
UPDATE : We’ve confirmed with Robert Hallock, technical PR lead at AMD, that while the GPU-Z tool is reporting an increase in memory frequency in reality the frequency did not change.
Click to expand...

Would you like to post some specific information about how it increases performance in anything other than a benchmark which may have been fiddled?
 
cageymaru said:
Air cooled version is a cut back chip. Think 780ti / 780. So no comparison can really be made.
Click to expand...


If its using the a similar power envelope to a full Fury chip, yes it can, if you understand how leakage, power and heat work, its easy to see it, because its the same chip, just cut parts.

And the 780 ti to the 780, you can't because they are different chips, even in this case you can still extrapolate some things based on the size of the silicon, like leakage, but much harder to do power draw based on the silicon size.

Sorry no you can its the same chip, with cut down parts, I was mistaken, so yeah you can still get a good idea of power usage.
 
Last edited:
cageymaru said:
Air cooled version is a cut back chip. Think 780ti / 780. So no comparison can really be made.
Click to expand...

has this been confirmed?

I assume the air cooled version will just be clocked lower, or not be able to maintain it's clock speed to due leakeage / heat and throttle more.
 
Yep it looks like Fury will be the same chip as Fury X because they have the same power rating.
 
WorldExclusive said:
Because 10bit gaming is so popular right now. :rolleyes:

If true 10bit color is needed, it's more likely I will have a workstation card paired with a true 10bit monitor. Both of which are awful for gaming.

If you're judging consumer cards based on workstation features, I can see why you're POV is terribly skewed.
Click to expand...

All POVs can be considered skewed by perception.

Here there are more than a few users of 10bit panels who also game. Sure there are better gaming panels in terms of response, but not all of us want multiple, different monitors, or rigs, or have space or coin for them.
Many take advantage of some workstation features offered on a consumer card, which therefore give a buying advantage over competing products.

As a productivity user, with colour often being very important, plus occasional to semi regular gaming, with an AMD gpu, I can achieve both quality gaming, along with limited workstation functions (10 bit outside of gaming chiefly). Can't do that with a single Nvidia card. Leaves more budget towards things that benefit my workload greater, e.g. more ram, good compatible screen etc.

Along with the availability of 10bit panels, I presumed they'd reversed this for Maxwell, thus was considering a 980Ti over AMD offerings, until seeing this is not the case and this current potential crapstorm..


noko said:
Anyways the BF4 images looks to be a more positive lod setting pushing up the mipmap textures (smaller textures) closer to view. If the case this would alleviate memory bandwidth required on texture processing. This would have a bigger impact over AF.

As for Fury CFX performance at 4K the initial testing is looking rather good - good for AMD.
Click to expand...

Agreed, It does look more like differences in lod.

4K CFX results are rather impressive for early days, especially for a 4gb card. Waiting for a good 390x CFX review too..
 
durquavian said:
You must not read anything outside of [H]forum. There are a number of posters everywhere talking about the OC mem bug in CCC, where ever so often when you boot it is available. With those that have tried, a 9%core 20%mem resulted in 20% firestrike score increases.
Click to expand...
Yep, random people that provide no proof that can be replicated can be trusted.
 
This really is not just about Fury X .. as how long has Nvidia been cheating the IQ settings for reviews is the question because we have to go back and think about 290x as AMD has worked there ass off making that card faster since 780GTX /Ti and now 980/Ti for reviews that are not on a level playing field and they did not drop off the IQ in doing so.

The root of this is money and what a Company will do to get it and if they (NVIDIA) can keep there products at the top of benchmarks by FPS to sell them then it says a lot about why the IQ is missing from the products reviewed under there driver..

Remember how they used Pissy X cheat to make there 3DMark scores look so much better then AMD products..Well the BOYZ are back to sell you some more products and I hate to see review sites being used as a tool because we buy based off what you tell us and someone has been telling some lies.
 
Last edited:
Nenu said:
Your description of yourself is amusing.

http://wccftech.com/amd-radeon-r9-fury-memory-oveclocked-20/


Would you like to post some specific information about how it increases performance in anything other than a benchmark which may have been fiddled?
Click to expand...

They've got another article (http://wccftech.com/unlock-memory-overclocking-amd-r9-fury/) posted yesterday, indicating that the memory can be overclocked with some third party software, and they've got links to third parties with benches and details.
 
Nenu said:
Your description of yourself is amusing.

http://wccftech.com/amd-radeon-r9-fury-memory-oveclocked-20/


Would you like to post some specific information about how it increases performance in anything other than a benchmark which may have been fiddled?
Click to expand...
AMD Fury X Memory Overclocking Unlocked In MSI Afterburner – HBM Overclocked And Benchmarked, GPU Voltage Control Coming
Click to expand...


http://wccftech.com/unlock-memory-overclocking-amd-r9-fury/

In Damient’s Fury X review at Hardware.fr the Fury X was tested with and without the memory overclock. So we can easily ascertain if the memory frequency did in fact change and it’s not just a software glitch. This also gives us the opportunity to find out how much of a performance boost we can expect from overclocking HBM. You can see the results of an 8% core and memory overclock on Fury X below.
Click to expand...
 
Status
Not open for further replies.
Back
Top