Polaris performance numbers perf/$

[Daniel_Chang]

I'm just not as excited for this generation as I previously was. Maybe because it's that I used to buy cards for $150-$200 (last two were 9600GT and GTX 560). Buying lower mid-range cards means drooling over the performance of higher end cards. But this time I went with a GTX 970. I kept telling myself that the 980ti was nearly twice as fast, or at least a solid 60% faster. Not even close. I looked at Techpowerup's GTX 980ti review (I like how they lay it out). Using the GTX 970 as the base, the 980ti was only this much faster at a given resolution (on average over numerous games):

1080p - 35%
1440p - 41%
2160p - 45%

Assuming that AMD gets Polaris at or near 980ti performance for $300, that would be $300 for a small, yet appreciable performance boost. Reality is sinking it. I'm likely not upgrading this generation. Now, had I gone with the GTX 960, as originally planned, the GTX 980ti (and the rumored Polaris 10) is/are 2x as fast on average. That would have seemed more impressive to me.

EDIT: Adjusted numbers from 30-41% to 35-45%. Corrected math.

 

[TaintedSquirrel]

"Only" 40% faster.
At the $300 price point that's a game changer.

 

[flatty]

The performance/$ isn't unfortunately linear, at the top ure in position to pay 15-20-25% of price for 5-10% of performance

 

[razor1]

AMD Polaris 11 and Polaris 10 GPUs pictured? | VideoCardz.com

Polaris 11

Polaris 10 die picture

Polaris 10 die comparison to tonga

 

[Alkaladur]

AMD Polaris 11 and Polaris 10 GPUs pictured? | VideoCardz.com

Polaris 11

Polaris 10 die picture

Polaris 10 die comparison to tonga

Nice picture of the Polaris 11 die, must be a joint venture between Canon and AMD.

AMD allowing Canon to photograph their upcoming GPU die with Canon's state-of-the-art 128k pixel CCD sensor

 

[Daniel_Chang]

"Only" 40% faster.
At the $300 price point that's a game changer.

My point was that I'm not going to upgrade from a $300 GPU one year later for another $300 for only a 30-40% boost. I want double for that price.

GTX 960/ R9 380 users would do well to upgrade to this, as with owners of older generation cards. But I'm going to stand pat for one more year, most likely. Nothing wrong with that.

Now, if Nvidia/AMD have a card for $400 that doubles the performance of my GTX 970, I could be swayed.

 

[Alkaladur]

My point was that I'm not going to upgrade from a $300 GPU one year later for another $300 for only a 30-40% boost. I want double for that price.

GTX 960/ R9 380 users would do well to upgrade to this, as with owners of older generation cards. But I'm going to stand pat for one more year, most likely. Nothing wrong with that.

Now, if Nvidia/AMD have a card for $400 that doubles the performance of my GTX 970, I could be swayed.

I expect them to offer 980ti performance around that price, that's roughly as well performing as 970sli is

 

[Maddness]

I expect them to offer 980ti performance around that price, that's roughly as well performing as 970sli is

That's highly unlikely as that would mean Fury and Fury X would have to be slashed by 50% or they would go EOL. Considering how expensive Fury is to make and how long they have been on the market. I can't really see this happening.

 

[Daniel_Chang]

I expect them to offer 980ti performance around that price, that's roughly as well performing as 970sli is

Ok, but as I showed (with actually numbers), the GTX 980ti is only 35-45% faster than a single GTX 970 over a plethora of actual games. If the R9 480/490 is this fast for $300, that would be an excellent achievement, but not enough for me to drop $300 on an upgrade. I was previously more excited because, like you, I had falsely assumed that the 980ti was closer to double the 970. But after looking at the numbers I realized just how wrong I was.

This is potentially a great value and a great card, just not a great upgrade for those already at or near that performance.

EDIT: Adjusted numbers from 30-41% to 35-45%. Corrected math.

 

[Alkaladur]

Ok, but as I showed (with actually numbers), the GTX 980ti is only 30-41% faster than a single GTX 970 over a plethora of actual games. If the R9 480/490 is this fast for $300, that would be an excellent achievement, but not enough for me to drop $300 on an upgrade. I was previously more excited because, like you, I had falsely assumed that the 980ti was closer to double the 970. But after looking at the numbers I realized just how wrong I was.

This is potentially a great value and a great card, just not a great upgrade for those already at or near that performance.

depends on the game, but no obvously not double the performance, but a comfortable 60% advantage clock for clock isn't bad

 

[skline00]

Up the resolution to 1440 and see how close the 970 is to the 980ti

 

[Maddness]

If the Polaris 10 is 30%-40% faster than the 980 Ti at US $300, I will get 2 and Crossfire them. Or whatever it is called in DX12.

 

[NKD]

If the Polaris 10 is 30%-40% faster than the 980 Ti at US $300, I will get 2 and Crossfire them. Or whatever it is called in DX12.

That's just wishful thinking. You really mean actually faster than 980ti? or 970?

They are not even expecting the gtx 1080 to be that much faster over 980ti. lol. I hope that was a typo, lol.

 

[Alkaladur]

That's just wishful thinking. You really mean actually faster than 980ti? or 970?

They are not even expecting the gtx 1080 to be that much faster over 980ti. lol. I hope that was a typo, lol.

If the GTX 1080 is 150-200% faster than the 980ti at 500$ i'll buy six of them

 

[Maddness]

That's just wishful thinking. You really mean actually faster than 980ti? or 970?

They are not even expecting the gtx 1080 to be that much faster over 980ti. lol. I hope that was a typo, lol.

I was making a joke. I can't see it being that fast. Some people here have far to higher expectations. If it is $300 I don't think it will be as fast as a 980 Ti, except in a very few DX12 titles.

 

[cageymaru]

I was hoping for 30% - 40% faster than 980ti for $100.

 

[NKD]

I was making a joke. I can't see it being that fast. Some people here have far to higher expectations. If it is $300 I don't think it will be as fast as a 980 Ti, except in a very few DX12 titles.

I honestly expect to be around GTX 980 speeds. Or trade blows with 1070. I think Fury cards are just for people that want a fancy card with HBM and watercooling, there will be no other reason to buy those cards if this card is as fast as even a 980. So to sum it up if its priced 350 or below it will sell all day. I don't expect it to be as fast as 980ti

 

[KazeoHin]

I was hoping for 30% - 40% faster than 980ti for $100.

I am fully expecting 300% faster than a TitanX for $35, and passive cooling, with the ability to achieve a 250% overclock without voltage adjustments.

 

[Maddness]

Hahaha. Some of you guys are a Crack up.

 

[Araxie]

I'm just not as excited for this generation as I previously was. Maybe because it's that I used to buy cards for $150-$200 (last two were 9600GT and GTX 560). Buying lower mid-range cards means drooling over the performance of higher end cards. But this time I went with a GTX 970. I kept telling myself that the 980ti was nearly twice as fast, or at least a solid 60% faster. Not even close. I looked at Techpowerup's GTX 980ti review (I like how they lay it out). Using the GTX 970 as the base, the 980ti was only this much faster at a given resolution (on average over numerous games):

1080p - 41%
1440p - 35%
2160p - 30%

What shocked me was the resolution scaling. So much for that 3.5GB VRAM issue. Assuming that AMD gets Polaris at or near 980ti performance for $300, that would be $300 for a small, yet appreciable performance boost. Reality is sinking it. I'm likely not upgrading this generation. Now, had I gone with the GTX 960, as originally planned, the GTX 980ti (and the rumored Polaris 10) is/are 2x as fast on average. That would have seemed more impressive to me.

any aftermarket cooled 980TI those that achieve ~1400mhz out of the box, will perform at similar GTX 970 SLI setup where SLI actually works good, and will destroy it in games where SLI doesn't work.. as you like Techpowerup numbers here ya go:

Spoiler: Relative Performance 1920*1080

Spoiler: Relative Performance 2560*1440

Battlefield 4 (SLI Works Flawless)

Spoiler: BF4 1080

Spoiler: BF4 1440

The witcher 3 (SLI work)

Spoiler: The witcher 3 1080

Spoiler: The witcher 3 1440

Just cause 3 (SLI Doesn't Work)

Spoiler: Just Cause 3 1080

Spoiler: Just Cause 3 1440

Fallout 4 (SLI works/Also CPU bottleneck here)

Spoiler: Fallout 4 1080

Spoiler: Fallout 4 1440

Assassins creed Syndicate (SLI barely works)

Spoiler: AC Syndicate 1080

Spoiler: AC Syndicate 1440

So that's it.. conclusions where SLI works flawlessly 970SLI Edge out the performance of a high clocked 980TI, where SLI work good trade blows, where SLI just barely Work the 980TI is better, where SLI doesn't Work the 980TI destroy 970SLI setup. Honestly, a Single powerful card Solution will always going to be better than a matching SLI Solution, due to performance consistency, no stuttering, no issues with Profiles, etc etc.. most games on launch date doesn't have any profile, some games have weeks or months without a working SLI profile issue that nobody with a single powerful card have.

 

[Alkaladur]

any aftermarket cooled 980TI those that achieve ~1400mhz out of the box, will perform at similar GTX 970 SLI setup where SLI actually works good, and will destroy it in games where SLI doesn't work.. as you like Techpowerup numbers here ya go:

Spoiler

Battlefield 4 (SLI Works Flawless)

The witcher 3 (SLI work)

Just cause 3 (SLI Doesn't Work)

Fallout 4 (SLI works/Also CPU bottleneck here)

Assassins creed Syndicate (SLI barely works)

So that's it.. conclusions where SLI works flawlessly 970SLI Edge out the performance of a high clocked 980TI, where SLI work good trade blows, where SLI just barely Work the 980TI is better, where SLI doesn't Work the 980TI destroy 970SLI setup. Honestly, a Single powerful card Solution will always going to be better than a matching SLI Solution, due to performance consistency, no stuttering, no issues with Profiles, etc etc.. most games on launch date doesn't have any profile, some games have weeks or months without a working SLI profile issue that nobody with a single powerful card have.

For the love of all the gods you may or may not believe in, use [ spoiler][ /spoiler] tags - no spaces!

We are lazy people and do not like having to scroll so much

 

[TaintedSquirrel]

970 SLi was pretty sweet before the 980 Ti came out, though.
If it's all you've got then it's an appealing option.

Kinda explains why AMD is focused on multi-GPU solutions for upcoming series.

 

[Alkaladur]

Honestly the 980Ti shines when properly watercooled, full block. With a decent non-reference board with good power delivery you're looking at 1500-1600mhz 8ghz oin the memory, all day every day

there's a man (assume it's a man) on overclock3d with watercooled 980Ti G1s

1640mhz on both

meow

Spoiler

 

[Daniel_Chang]

depends on the game, but no obvously not double the performance, but a comfortable 60% advantage clock for clock isn't bad

That would be fabulous if the number wasn't fabricated. It's 35-45%, not 60%. (SOURCE) I know that you stated clock for clock, but the stock/vendor clocks aren't that different to account for and additional 15-25% performance.

Up the resolution to 1440 and see how close the 970 is to the 980ti

I already stated that, but I'll do it again just for you. It's 41% (SOURCE)

In this performance summary (which averages out the differences across 21 popular titles), the GTX 980ti is 100%, and the GTX 970 is 71%. But you use the GTX 970 as the base, so it's (100-71) / 71 = 40.8%. At 1080p it was 35.1%. At 4k it was 44.9%.

 

[Araxie]

For the love of all the gods you may or may not believe in, use [ spoiler][ /spoiler] tags - no spaces!

We are lazy people and do not like having to scroll so much

Hey, Im even more Lazier yet so, have to do all that things on phone is really crappy. anyway did for you =)

That would be fabulous if the number wasn't fabricated. It's 35-45%, not 60%. (SOURCE) I know that you stated clock for clock, but the stock/vendor clocks aren't that different to account for and additional 15-25% performance.



I already stated that, but I'll do it again just for you. It's 41% (SOURCE)

In this performance summary (which averages out the differences across 21 popular titles), the GTX 980ti is 100%, and the GTX 970 is 71%. But you use the GTX 970 as the base, so it's (100-71) / 71 = 40.8%. At 1080p it was 35.1%. At 4k it was 44.9%.

read my post above and try again.. =) you are using Reference board with the crappy reference cooler. try again with any customized board with a very good cooler like the card i've used which isn't the best at all.. but is one of the newest in the market so its a recent review...

 

[Alkaladur]

That would be fabulous if the number wasn't fabricated. It's 35-45%, not 60%. (SOURCE) I know that you stated clock for clock, but the stock/vendor clocks aren't that different to account for and additional 15-25% performance.



I already stated that, but I'll do it again just for you. It's 41% (SOURCE)

In this performance summary (which averages out the differences across 21 popular titles), the GTX 980ti is 100%, and the GTX 970 is 71%. But you use the GTX 970 as the base, so it's (100-71) / 71 = 40.8%. At 1080p it was 35.1%. At 4k it was 44.9%.

2816/1644

Hey, Im even more Lazier yet so, have to do all that things on phone is really crappy. anyway did for you =)



read my post above and try again.. =) you are using Reference board with the crappy reference cooler. try again with any customized board with a very good cooler like the card i've used which isn't the best at all.. but is one of the newest in the market so its a recent review...

980ti has more than 50% more shaders, 50% more memory bandwidth more rops more tmus more geometry performance

 

[Araxie]

2816/1644



980ti has more than 50% more shaders, 50% more memory bandwidth more rops more tmus more geometry performance

Man I know that, but 50% more everything also isn't directly translated into 50% performance increase, not in a not-directly comparison as Daniel_Chang is doing..

Clocks have a heavy weight which most tend to not consider, 90% of the GTX 970s are custom (and even a "reference" model doesn't exist beyond the BestBuy Edition with reference 980 Cooler) those non reference card with non-reference coolers GTX 970 tend to go above 1400mhz out of the box in heavy contrast with the GTX 980TI which average 1150mhz and tend to max at 1200mhz unless you crank the fan to 100% and still the boost will not be that high.. so yes at first instance it may appear that the GTX 970 isn't as far to the 980TI as numbers in papers may show, but isn't the complete and true story of the chapter.

That's of course in real world non direct performance comparisons, now when we add a GTX 980TI with non-reference cooler and board, then we know easily how those 50% more shaders, more bandwidth, Rops, TMUs, involve a better scaling with the increase of clocks and that's the main reason why we see every Custom GTX 980TI performing a good margin better than the GTX Titan X, not precisely by the clock for clock performance but because the high clock they can achieve out of the box in comparison to the GTX Titan X. the truly thing is at the end, there's no way someone can have in his mind the idea that the GTX 980TI doesn't worth the performance over a GTX 970 when the truly fact is is a night/day difference even in 1080P usage scenarios. I have over 8 GTX 980TI and 4 GTX 980 and 4 GTX 970, different brand and models and here's no way I can consider a GTX 970 Beyond 1080P and still it lack of performance for the kind of usage Im used and like to, is really half the card of what a 980TI can offer of course this is with heavy overclocking in mind =)

 

[razor1]

yeah bottlenecks shifts so can't do a direct relationship with units and increase of those units.

 

[Daniel_Chang]

2816/1644



980ti has more than 50% more shaders, 50% more memory bandwidth more rops more tmus more geometry performance

Irrelevant. My numbers are actual performance figures as tested across 21 popular titles. Your numbers show theoretical performance, not actual.

In real world usage, the GTX 980ti is roughly 40% faster than the GTX 970, and that is undisputable fact.

 

[Alkaladur]

Irrelevant. My numbers are actual performance figures as tested across 21 popular titles. Your numbers show theoretical performance, not actual.

In real world usage, the GTX 980ti is roughly 40% faster than the GTX 970, and that is undisputable fact.

No that's an average over many games, there could be bottlenecks thinning the delta.

Araxie mentioned average clocks of a 970 vs a 980ti but again, clock for clock theres 50%+ more everything really

Spoiler

970 = 57%

(100-57)/57 = 75% faster

even at 1080p in the link you posted

Spoiler

970 = 74%

26/74 = 35% faster despite lower clocks

At 4k reference vs reference

Spoiler

980ti is 45% faster, again running at lower clocks

50% is not far fetched

50% faster

 

[Araxie]

Irrelevant. My numbers are actual performance figures as tested across 21 popular titles. Your numbers show theoretical performance, not actual.

In real world usage, the GTX 980ti is roughly 40% faster than the GTX 970, and that is undisputable fact.

And here I'm, proving you are wrong, even explaining why you are wrong, even with facts from your favorite site with a recent review. And still you are there ignoring those facts and stuck with your "indisputable fact"..

 

[Daniel_Chang]

No that's an average over many games, there could be bottlenecks thinning the delta.

Please, re-read what you just posted. It's an average...across many...games. How else would you measure performance? Cherry pick one game that supports your specific point of view? That's now how I do it. The larger the sample, the better. On average, the GTX 980ti is 40% faster than the GTX 970. There's no way around that unless you cherry pick or manipulate the date (which you did and I'll explain below).

970 = 57%

(100-57)/57 = 75% faster

Nope. A STOCK CLOCKED 970 is 57%. A STOCK CLOCKED 980ti is 81%. A GROSSLY OVERCLOCKED 980ti is 100%. That's not even apples to oranges. That's apples to hand grenades.

In that example, the stock 980ti is 42.1% faster than the stock 970. That's not far of the 40.8% that I showed before at 1440p.

even at 1080p in the link you posted

970 = 74%

26/74 = 35% faster despite lower clocks

The "lower clocks" is about 50mhz, or 5% of the base clock rate. That's not a huge difference. It's not like one card was 1000mhz and the other was 1500mhz. And 35% at 1080p is the same as what I got before. So this doesn't disagree with my findings.

At 4k reference vs reference

980ti is 45% faster, again running at lower clocks

50% is not far fetched

Ok, so, three things here.

1) 45% is what I got, so no disagreement there.

2)You're arguing that if you overclock the 980ti, it gains more performance. Uh...no shit? The GTX 970 overclocks too. Just an FYI.

3) You moved the goalposts. You said in your last post that the 980ti was "a comfortable 60% faster clock for clock," but even now, you're settling for less than that.

So, let's summarize:

I've shown that based on a plethora of games benchmarked, the 980ti is 35-45% faster than the 970. You've disagreed with me. You've posted numbers that agree with my numbers. You've then argued that you can overclock the 980ti to make it a comfortable 60% faster. You've revised that down to "50% is not far fetched." In all of this you're ignoring the fact that you're comparing an overclocked 980ti to a reference clocked 970...which can also be overclocked. I really don't think that I have anything else to prove here. A 980ti is 35%-45% faster that a 970. Both your data and mine proves that.

And here I'm, proving you are wrong, even explaining why you are wrong, even with facts from your favorite site with a recent review. And still you are there ignoring those facts and stuck with your "indisputable fact"..

I'm talking about a single GTX 970 vs. a single GTX 980ti, both at reference. And if you're going to overclock one, you should be fair and overclock the other.

In your post (which was excellent, btw), you're talking about 970 SLI vs a single 980ti. That has not a darn thing to do with my post. Excellent data, just not relevant to my part of the discussion.

 

[Alkaladur]

Please, re-read what you just posted. It's an average...across many...games. How else would you measure performance? Cherry pick one game that supports your specific point of view? That's now how I do it. The larger the sample, the better. On average, the GTX 980ti is 40% faster than the GTX 970. There's no way around that unless you cherry pick or manipulate the date (which you did and I'll explain below).



Nope. A STOCK CLOCKED 970 is 57%. A STOCK CLOCKED 980ti is 81%. A GROSSLY OVERCLOCKED 980ti is 100%. That's not even apples to oranges. That's apples to hand grenades.

In that example, the stock 980ti is 42.1% faster than the stock 970. That's not far of the 40.8% that I showed before at 1440p.



The "lower clocks" is about 50mhz, or 5% of the base clock rate. That's not a huge difference. It's not like one card was 1000mhz and the other was 1500mhz. And 35% at 1080p is the same as what I got before. So this doesn't disagree with my findings.



Ok, so, three things here.

1) 45% is what I got, so no disagreement there.

2)You're arguing that if you overclock the 980ti, it gains more performance. Uh...no shit? The GTX 970 overclocks too. Just an FYI.

3) You moved the goalposts. You said in your last post that the 980ti was "a comfortable 60% faster clock for clock," but even now, you're settling for less than that.

So, let's summarize:

I've shown that based on a plethora of games benchmarked, the 980ti is 35-45% faster than the 970. You've disagreed with me. You've posted numbers that agree with my numbers. You've then argued that you can overclock the 980ti to make it a comfortable 60% faster. You've revised that down to "50% is not far fetched." In all of this you're ignoring the fact that you're comparing an overclocked 980ti to a reference clocked 970...which can also be overclocked. I really don't think that I have anything else to prove here. A 980ti is 35%-45% faster that a 970. Both your data and mine proves that.



I'm talking about a single GTX 970 vs. a single GTX 980ti, both at reference. And if you're going to overclock one, you should be fair and overclock the other.

In your post (which was excellent, btw), you're talking about 970 SLI vs a single 980ti. That has not a darn thing to do with my post. Excellent data, just not relevant to my part of the discussion.

Since it is gpu performance you are comparing maybe it is the case to limit it to gpu-bound scenarios hm?

For example
The Division: PC graphics performance benchmark review

 

[Araxie]

Please, re-read what you just posted. It's an average...across many...games. How else would you measure performance? Cherry pick one game that supports your specific point of view? That's now how I do it. The larger the sample, the better. On average, the GTX 980ti is 40% faster than the GTX 970. There's no way around that unless you cherry pick or manipulate the date (which you did and I'll explain below).



Nope. A STOCK CLOCKED 970 is 57%. A STOCK CLOCKED 980ti is 81%. A GROSSLY OVERCLOCKED 980ti is 100%. That's not even apples to oranges. That's apples to hand grenades.

In that example, the stock 980ti is 42.1% faster than the stock 970. That's not far of the 40.8% that I showed before at 1440p.



The "lower clocks" is about 50mhz, or 5% of the base clock rate. That's not a huge difference. It's not like one card was 1000mhz and the other was 1500mhz. And 35% at 1080p is the same as what I got before. So this doesn't disagree with my findings.



Ok, so, three things here.

1) 45% is what I got, so no disagreement there.

2)You're arguing that if you overclock the 980ti, it gains more performance. Uh...no shit? The GTX 970 overclocks too. Just an FYI.

3) You moved the goalposts. You said in your last post that the 980ti was "a comfortable 60% faster clock for clock," but even now, you're settling for less than that.

So, let's summarize:

I've shown that based on a plethora of games benchmarked, the 980ti is 35-45% faster than the 970. You've disagreed with me. You've posted numbers that agree with my numbers. You've then argued that you can overclock the 980ti to make it a comfortable 60% faster. You've revised that down to "50% is not far fetched." In all of this you're ignoring the fact that you're comparing an overclocked 980ti to a reference clocked 970...which can also be overclocked. I really don't think that I have anything else to prove here. A 980ti is 35%-45% faster that a 970. Both your data and mine proves that.



I'm talking about a single GTX 970 vs. a single GTX 980ti, both at reference. And if you're going to overclock one, you should be fair and overclock the other.

In your post (which was excellent, btw), you're talking about 970 SLI vs a single 980ti. That has not a darn thing to do with my post. Excellent data, just not relevant to my part of the discussion.

but that's exactly the main thing bud, "reference" or "Stock"clocks doesn't exist anymore with nivida cards since kepler, due the way GPU boost works you can have 10 cards of the same brand and model and all of those 10 cards will behave different and all those 10 cards even being the same brand and model will have a different out of the box boost clock, that's why I use Out of the box clock and no stock or reference clock, because stock doesn't exist anymore,

As I said before even a "Reference" GTX 970 doesn't truly exist, so 99% of the GTX 970 are AIB custom boards with custom coolers that clock over 1400mhz out of the box, some of the highest end customized models go above 1500mhz out of the box, so as you say how we compare that to a GTX 980TI? easy.. utilize a non-reference customized 980TI that can go to a high clock and that's what exactly what Alkaladur did with the Asus card posted and even better, you can take the gigabyte Xtreme 980TI i've posted, then we have cards that can be directly compared in a more "clock for clock" basis because they achieve similar boost clock out of the box WITHOUT manual overclock which is where you failed by saying "GROSSLY OVERCLOCKED" above, because isn't overclocked its just going to a high boost clock, the same way as every other GTX 970 do.. so yes you have some truly points and facts, but you are using it incorrectly by not take in consideration the clocks of the cards, can we take the average of 1400mhz boost clock of the GTX 970 as Grossly Overclocked? the same way you're calling any other high boost clocked 980TI?. because that would be the perfect way to do an apple to apple comparison to really understand how architecture scale with clocks...

 

[Alkaladur]

but that's exactly the main thing bud, "reference" or "Stock"clocks doesn't exist anymore with nivida cards since kepler, due the way GPU boost works you can have 10 cards of the same brand and model and all of those 10 cards will behave different and all those 10 cards even being the same brand and model will have a different out of the box boost clock, that's why I use Out of the box clock and no stock or reference clock, because stock doesn't exist anymore,

As I said before even a "Reference" GTX 970 doesn't truly exist, so 99% of the GTX 970 are AIB custom boards with custom coolers that clock over 1400mhz out of the box, some of the highest end customized models go above 1500mhz out of the box, so as you say how we compare that to a GTX 980TI? easy.. utilize a non-reference customized 980TI that can go to a high clock and that's what exactly what Alkaladur did with the Asus card posted and even better, you can take the gigabyte Xtreme 980TI i've posted, then we have cards that can be directly compared in a more "clock for clock" basis because they achieve similar boost clock out of the box WITHOUT manual overclock which is where you failed by saying "GROSSLY OVERCLOCKED" above, because isn't overclocked its just going to a high boost clock, the same way as every other GTX 970 do.. so yes you have some truly points and facts, but you are using it incorrectly by not take in consideration the clocks of the cards, can we take the average of 1400mhz boost clock of the GTX 970 as Grossly Overclocked? the same way you're calling any other high boost clocked 980TI?. because that would be the perfect way to do an apple to apple comparison to really understand how architecture scale with clocks...

Honestly I didn't understand what you're saying.

A fair comparison would be at same clocks

However there is also to consider that maxwell has a memory bandwidth advantage because of wider bus and no funky memory config.

It's 50% faster in most cases, in some cases more (performance scaling slows down at high clocks for 970)

 

[Daniel_Chang]

Ok, so let's do one game, The Division. Let's see how this pans out.

1080p - 970 is 57fps, 980ti is 81, that's 42.1%.

1440p - 47.5%

2160p - 94.4%

Aside from 4k, this STILL doesn't support your viewpoint, and it's a cherry picked game. You literally chose one game to support your point of view, and it still didn't quite reach the 50% (let alone 60%) at 1440p that you preached. So what would happen if I did that?

Metro Last Light:

1080p - 30.1%

1440p - 33.5%

2160p - 37.8%

And that is why I took a loom at 21 games. It's not biased, or cherry picked. It gives you and idea of just how the two platforms perform in a broad spectrum. I know that you want to be right. You desperately want to be right. That's why you keep cherry picking numbers to suit your theory. But by now, you should begin to realize two things:

1. Even your cherry picked numbers don't fully suit your side.

2. You're not going to convince me that you're right unless you use another large batch of games that supports your stance that a GTX 980ti consistently outperforms the GTX 970 (apples to apples, not pitting an overclocked card against a reference one) by more than 50% across the board (you originally said 60%, and I'm not holding you to that).

 

[Araxie]

Honestly I didn't understand what you're saying.

A fair comparison would be at same clocks

However there is also to consider that maxwell has a memory bandwidth advantage because of wider bus and no funky memory config.

It's 50% faster in most cases, in some cases more (performance scaling slows down at high clocks for 970)

that would close to impossible to really test in real world, every game stress every card in a different way, triggering the TDP limits faster than others, fair comparisons will be with cards that behave with some similarities with similar boost clocks and so on, I don't know why you even said maxwell there, Yes I guess you meant GM204(GTX 980/GTX970) versus GM200 (Titan X/GTX980TI) which, then yes to all you said.

 

[Alkaladur]

Ok, so let's do one game, The Division. Let's see how this pans out.

1080p - 970 is 57fps, 980ti is 81, that's 42.1%.

1440p - 47.5%

2160p - 94.4%

Aside from 4k, this STILL doesn't support your viewpoint, and it's a cherry picked game. You literally chose one game to support your point of view, and it still didn't quite reach the 50% (let alone 60%) at 1440p that you preached. So what would happen if I did that?

Metro Last Light:

1080p - 30.1%

1440p - 33.5%

2160p - 37.8%

And that is why I took a loom at 21 games. It's not biased, or cherry picked. It gives you and idea of just how the two platforms perform in a broad spectrum. I know that you want to be right. You desperately want to be right. That's why you keep cherry picking numbers to suit your theory. But by now, you should begin to realize two things:

1. Even your cherry picked numbers don't fully suit your side.

2. You're not going to convince me that you're right unless you use another large batch of games that supports your stance that a GTX 980ti consistently outperforms the GTX 970 (apples to apples, not pitting an overclocked card against a reference one) by more than 50% across the board (you originally said 60%, and I'm not holding you to that).

Alright last try for me

Those averages over many games are t good because they could include non gpu bound situations which skew the results.

The 980Ti has a 50%h hardware advantage from every point of view, it does translate into a performance advantage, if it doesn't at 1080p it's because 1080p isn't pushing hard enough for it to show.

Why would you ignore 4k, 4k is exactly where these differences are most obvious

 

[Daniel_Chang]

but that's exactly the main thing bud, "reference" or "Stock"clocks doesn't exist anymore with nivida cards since kepler, due the way GPU boost works you can have 10 cards of the same brand and model and all of those 10 cards will behave different and all those 10 cards even being the same brand and model will have a different out of the box boost clock, that's why I use Out of the box clock and no stock or reference clock, because stock doesn't exist anymore,

As I said before even a "Reference" GTX 970 doesn't truly exist, so 99% of the GTX 970 are AIB custom boards with custom coolers that clock over 1400mhz out of the box, some of the highest end customized models go above 1500mhz out of the box, so as you say how we compare that to a GTX 980TI? easy.. utilize a non-reference customized 980TI that can go to a high clock and that's what exactly what Alkaladur did with the Asus card posted and even better, you can take the gigabyte Xtreme 980TI i've posted, then we have cards that can be directly compared in a more "clock for clock" basis because they achieve similar boost clock out of the box WITHOUT manual overclock which is where you failed by saying "GROSSLY OVERCLOCKED" above, because isn't overclocked its just going to a high boost clock, the same way as every other GTX 970 do.. so yes you have some truly points and facts, but you are using it incorrectly by not take in consideration the clocks of the cards, can we take the average of 1400mhz boost clock of the GTX 970 as Grossly Overclocked? the same way you're calling any other high boost clocked 980TI?. because that would be the perfect way to do an apple to apple comparison to really understand how architecture scale with clocks...

But no, he didn't.

You're contradicting yourself. First, you state that there is no reference GTX 970 (you're correct, it's only a reference spec). So, you state that we can only compare overclocked cards (mostly correct, there are some reference clocked cards of each, though they are the minority). But then you cite Alkaladur's post, which actually compared a reference clocked GTX 970 to an ASUS Matrix, which has a 21.6% base overlock of the core, and a factory memory overclock (which not all cards do). This isn't the comparison that you think it is, and now I have to ask:

Are you deliberately cherry picking to fit a theory, or are you genuinely unaware that it was not an apples to apples comparison?

 

[Daniel_Chang]

Alright last try for me

Those averages over many games are t good because they could include non gpu bound situations which skew the results.

The 980Ti has a 50%h hardware advantage from every point of view, it doesn't translate into a performance advantage, if it doesn't at 1080p it's because 1080p isn't pushing hard enough for it to show.

Why would you ignore 4k, 4k is exactly where these differences are most obvious

You're right in that 4k is where it's most obvious. And your cherry picked game showed a 94% gain. And my cherry picked game showed a 38% gain. Again, the reason that you test multiple games, and not just one game is to get an overall picture of how a card performs. And the CPU limitation does matter, just not the way that you intend it to.

Let me clarify. My games run at a certain speed, let's say, 100 miles per hour. In order for me to justify spending $300 on a new GPU, I want to go 200mph. Are you with me? Going from a 970 (100mph) to a 980ti (135-145mph) is NOT 200mph, even if I were to change my CPU (not changing the CPU, so the CPU limitation stays, regardless of what it is). So, if AMD comes out with a new card that runs as fast as the 980ti...will it get me to 200mph? Nope.

I cannot explain it any simpler than that. My entire point was that I don't upgrade until I get 2x benefit. I showed a 1.35x-1.45x benefit, so I'm staying pat. You argued a 1.6x benefit with no data to back it up (still not 2x). You then reduced your argument to 1.5x (data doesn't back this up). You then showed one particular game with one specific scenario where it ALMOST eeked out 2.x. So unfortunately, despite your best efforts to cherry pick data in your favor, you couldn't demonstrate 1) that my 1.35x-1.45x data was incorrect, and 2) that the card would be a 2x upgrade for me.

But, you said that was your last attempt, so I'll make this mine. I bear no animosity towards you. And I'm not going to claim to be a GPU expert. All I'm going to say is that the data before me convinced me that an upgrade at this time isn't worthwhile. Like you, I thought it was 1.6x at LEAST, but then the data changed my mind.