Alt+Esc Review: HD6970 vs. GTX680/770 vs. HD7970GE vs. GTX780

Page 4 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

2is

Diamond Member
Apr 8, 2012
4,281
131
106
I liked the 7970 and used it for several months, but in retrospect - the big problem was that the launch of the 7970 had several mistakes. The biggest one being that the launch drivers had comparatively poor performance and weren't WHQL certified for a couple of months.

I don't think they're stupid enough to replicate these problems though, AMD has apparently expressed that they learned hard lessons from the 7970 launch and won't make the same mistakes again. If they have solid launch drivers with workable crossfire, I expect good things from the R9xxx.

Not to mention, it was more expensive and slower than a 680 at launch. That's why I went with a 680. In retrospect, I'm glad I did because I've been running SLI for a while now and AMD only recently started improving their multi-gpu performance.
 

Elfear

Diamond Member
May 30, 2004
7,096
640
126
Why are you comparing a stock 780 against 7970 GHz? 780s come in much faster flavors now so lets compare top vs top. I'm still waiting on someone with a 7970 ghz edition to produce a stable OC that "nips at the heels" of a 780 classified at stock. Don't forget the 780 (esp classified) can OC like mad as well. Even Titan can run at 1.3v now and hit 1300 mhz+ core on air which absolutely destroys the 7970 ghz.

You can make any comparison you want to but the original statement by SlowSpyder was:
I bet a 7970 with a healthy overclock would nip at the heels of a GTX780 (at factory clocks).

I took that to mean a regular 780. No one is debating that a overvolted/overclocked 780 isn't a very fast card, one a 7970 will never catch up to. But a 7970 can certainly match a stock 780 (i.e. non-uber edition) with a healthy overclock as I showed in my previous post.

Regarding the 780 Classified, from the only review I could find, it appears to be ~5% faster than the regular 780. So adding 5% to the 7970 clocks I previously laid out would give us ~1260Mhz for a 780 with factory settings and 1310-1365Mhz for a 780 with manual adjustments. 1365Mhz is getting into the upper limit of a 7970's capability on software voltage mods but is still within the realm of reality.

Overclocking the 780 wasn't part of the original discussion but people are getting some pretty amazing overclocks out of the aftermarket versions.
1410Mhz on Air
1300Mhz+
 

5150Joker

Diamond Member
Feb 6, 2002
5,559
0
71
www.techinferno.com
@Elfear - I'm not knocking HardOCP and others but the problem is that despite claiming to be "Hard" all their overclocking and testing is really limited to what the OEM gives them so there's nothing special about their reviews. Same holds true for Anandtech and most other websites. They don't represent the true potential of any of these cards in any of their tests/reviews. I can understand that they're catering to a wide audience but what's the point in calling your site "Hard" if you're gonna do run of the mill tests like Engadget?

My point above is that referencing most of these reviews when talking about max performance is totally worthless because they never venture into voltage unlocks and custom vbios usage which everyone has access to (see the Tech|Inferno link in my sig).

Now this isn't like a 780 but here's a run I just did with my Titans with tons of stuff running in the background and this OC isn't even the highest I can go (my Titans are voltage unlocked with custom vbios btw): http://www.3dmark.com/3dm11/7061789

That's why I never really put any weight behind these reviews that people often reference because they don't hold any credibility for a real enthusiast and I'd like to believe that people who drop $700+ on video cards are categorized as enthusiasts and not your average noob.
 

2is

Diamond Member
Apr 8, 2012
4,281
131
106
If I bought a Titan, I personally wouldn't bother with custom bios' and voltage unlocks. The Risk:Reward isn't worth it to me, but more importantly, I don't have the patience for it. Some people like running benchmarks and getting the highest score possible, me, I like to actually use the cards to play games with. Based on my past experiences, the performance differences in-game between a card that's OCed within what the stock BIOS allows for, and one with a modded BIOS, is rather slim more often than not.

I can easily see why review sites review the way they do. Reviewing PC gaming hardware is a niche market to begin with, narrowing it even further would be relevant to almost no one.
 

5150Joker

Diamond Member
Feb 6, 2002
5,559
0
71
www.techinferno.com
If I bought a Titan, I personally wouldn't bother with custom bios' and voltage unlocks. The Risk:Reward isn't worth it to me, but more importantly, I don't have the patience for it. Some people like running benchmarks and getting the highest score possible, me, I like to actually use the cards to play games with. Based on my past experiences, the performance differences in-game between a card that's OCed within what the stock BIOS allows for, and one with a modded BIOS, is rather slim more often than not.

I can easily see why review sites review the way they do. Reviewing PC gaming hardware is a niche market to begin with, narrowing it even further would be relevant to almost no one.

I've seen very nice gains in game with overclocking, especially in more demanding games where the min fps drops.
 

Elfear

Diamond Member
May 30, 2004
7,096
640
126
@Elfear - I'm not knocking HardOCP and others but the problem is that despite claiming to be "Hard" all their overclocking and testing is really limited to what the OEM gives them so there's nothing special about their reviews. Same holds true for Anandtech and most other websites. They don't represent the true potential of any of these cards in any of their tests/reviews. I can understand that they're catering to a wide audience but what's the point in calling your site "Hard" if you're gonna do run of the mill tests like Engadget?

My point above is that referencing most of these reviews when talking about max performance is totally worthless because they never venture into voltage unlocks and custom vbios usage which everyone has access to (see the Tech|Inferno link in my sig).

Now this isn't like a 780 but here's a run I just did with my Titans with tons of stuff running in the background and this OC isn't even the highest I can go (my Titans are voltage unlocked with custom vbios btw): http://www.3dmark.com/3dm11/7061789

That's why I never really put any weight behind these reviews that people often reference because they don't hold any credibility for a real enthusiast and I'd like to believe that people who drop $700+ on video cards are categorized as enthusiasts and not your average noob.

I get what your saying Joker. GK110 really has a lot of potential once you raise the voltage and max the temp and power consumption sliders. I'm very happy to see Nvidia ease up on that restriction and hope that practice continues in the future.

As an enthusiast you can bet I'm going to try and get as much performance out of a GPU as I can. That's why I think overclocking potential should be taken into consideration when purchasing a card. I felt that way when the 7970 launched and I feel that way about the 780. Free performance is awesome. :wub:

SlowSpyder was just making a statement about an oced 7970 vs a stock 780. To an enthusiast the comparison doesn't make much sense because you'd overclock both cards. I believe he was merely pointing out that ruhtraeel should set his expectations higher with an overclocked 7970, being more comparable to a stock 780 rather than a stock 770.
 

2is

Diamond Member
Apr 8, 2012
4,281
131
106
I've seen very nice gains in game with overclocking, especially in more demanding games where the min fps drops.

"Nice gains" isn't a quantifiable term, and I'm not talking about merely overclocking, but comparing the added OC headroom a modded bios gives you vs a simple OC with the stock bios. The in-game performance advantage of the former is hardly, if ever, worth it for most people. Your in game performance advantage is more often than not, going to be about 5-8%

Not wanting to go through that hassle for 5% doesn't mean you aren't an "enthusiast" it just means you'd rather use that time doing other things.
 

amenx

Diamond Member
Dec 17, 2004
3,848
2,013
136
So much for the higher bit bus, moar vid mem trouncing Nvidias 256 bit performance at high res, multi-monitor. The 770 still beats or is on par with the 7970GE in the higher res charts.
 

wand3r3r

Diamond Member
May 16, 2008
3,180
0
0
I get what your saying Joker. GK110 really has a lot of potential once you raise the voltage and max the temp and power consumption sliders. I'm very happy to see Nvidia ease up on that restriction and hope that practice continues in the future.

Did I miss Nv allowing higher volts now? Do you have a source for this? Did they actually quit neutering the cards because the last I knew that was the direction they were going.
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
So much for the higher bit bus, moar vid mem trouncing Nvidias 256 bit performance at high res, multi-monitor. The 770 still beats or is on par with the 7970GE in the higher res charts.

Are you reading the same review? GTX770 was 3% slower at 2560x1440 overall and tied at multi-monitor. Considering HD7970 1.05Ghz goes for $300 and GTX770 2-4GB cards are $400-450, that's one of the most overpriced cards in NV's line up. The 770 is so badly overpriced that this week you could get 2 after market $200 7950s for the same price. We know that a single 7950 OC > stock 680/7970GE. NV needs to drop the price to $350 for this card to make any sense.
 

Red Hawk

Diamond Member
Jan 1, 2011
3,266
169
106
Are you reading the same review? GTX770 was 3% slower at 2560x1440 overall and tied at multi-monitor. Considering HD7970 1.05Ghz goes for $300 and GTX770 2-4GB cards are $400-450, that's one of the most overpriced cards in NV's line up. The 770 is so badly overpriced that this week you could get 2 after market $200 7950s for the same price. We know that a single 7950 OC > stock 680/7970GE. NV needs to drop the price to $350 for this card to make any sense.

I think you're missing his point. Yes, the 7970 GHz is better performance per dollar, but looking at the underlying components -- for 50% more memory and a 50% larger memory bus, the 7970 only gets that 3% advantage. (And I'd say that a measly 3% advantage counts more as "on par"). The memory bus and memory size are not pulling through as a decisive factor in AMD's favor.
 

bystander36

Diamond Member
Apr 1, 2013
5,154
132
106
@Elfear - I'm not knocking HardOCP and others but the problem is that despite claiming to be "Hard" all their overclocking and testing is really limited to what the OEM gives them so there's nothing special about their reviews. Same holds true for Anandtech and most other websites. They don't represent the true potential of any of these cards in any of their tests/reviews. I can understand that they're catering to a wide audience but what's the point in calling your site "Hard" if you're gonna do run of the mill tests like Engadget?

My point above is that referencing most of these reviews when talking about max performance is totally worthless because they never venture into voltage unlocks and custom vbios usage which everyone has access to (see the Tech|Inferno link in my sig).

Now this isn't like a 780 but here's a run I just did with my Titans with tons of stuff running in the background and this OC isn't even the highest I can go (my Titans are voltage unlocked with custom vbios btw): http://www.3dmark.com/3dm11/7061789

That's why I never really put any weight behind these reviews that people often reference because they don't hold any credibility for a real enthusiast and I'd like to believe that people who drop $700+ on video cards are categorized as enthusiasts and not your average noob.

HardOCP does have something special about their reviews, they test cards at the settings that most people would use them. Rather than having all their cards run at the same settings, they find the best settings for a good playing experience for each individual card. That is quite unique.

The bad part about it is they assume everyone has the same tastes on what is considered smooth. I need 1.5-2 times their playable FPS to feel comfortable and not get nauseated.

Now they may not be extreme enthusiast, as you seem to be, but they do seem to go after the average enthusiast.
 

The Alias

Senior member
Aug 22, 2012
647
58
91
I think you're missing his point. Yes, the 7970 GHz is better performance per dollar, but looking at the underlying components -- for 50% more memory and a 50% larger memory bus, the 7970 only gets that 3% advantage. (And I'd say that a measly 3% advantage counts more as "on par"). The memory bus and memory size are not pulling through as a decisive factor in AMD's favor.

the gtx 770 is automatically overclocking itself really highly so the comparison isn't valid only once both cards are oc'ed to the max can we really compare
 

Red Hawk

Diamond Member
Jan 1, 2011
3,266
169
106
the gtx 770 is automatically overclocking itself really highly so the comparison isn't valid only once both cards are oc'ed to the max can we really compare

The Radeon HD 7970 GHz has Boost, so you could say it's "automatically overclocking" itself too, right? And the problem with OC'ing is that it varies from card to card, from board design to board design. It's not a reliable comparison.
 

The Alias

Senior member
Aug 22, 2012
647
58
91
The Radeon HD 7970 GHz has Boost, so you could say it's "automatically overclocking" itself too, right? And the problem with OC'ing is that it varies from card to card, from board design to board design. It's not a reliable comparison.

the 7970 limits it's boost to 1050mhz, the 770 has no such limitation . Oc'ing both is certainly more meaningful than comparing a card that truly limits it's clocks to one that says it has a max boost clock, but in reality, really doesn't .
 

amenx

Diamond Member
Dec 17, 2004
3,848
2,013
136
Are you reading the same review? GTX770 was 3% slower at 2560x1440 overall and tied at multi-monitor. Considering HD7970 1.05Ghz goes for $300 and GTX770 2-4GB cards are $400-450, that's one of the most overpriced cards in NV's line up. The 770 is so badly overpriced that this week you could get 2 after market $200 7950s for the same price. We know that a single 7950 OC > stock 680/7970GE. NV needs to drop the price to $350 for this card to make any sense.
Missed this post, but basically answered it in another thread:

Not arguing for value of any cards here, my main focus was on the 770's 256-bit performance at high res. In that review, there was an anomaly (in Grid2) where the 770 underperformed at 2560x1440 yet beat the 7970GE at 5760x1200. In the end, they lumped all the FPS together and averaged them out which can give misleading conclusions. Yet, the 770 beat (or equaled) the 7970GE in 4 games vs 2 @ 2560x1440 and in 5 games vs 1 @ 5760x1200. After this I will never look at averaged performance of games in cards, incl that of TPU. But rather look at each individual games performance to arrive at what I think would be a clearer view a cards overall performance.

Look at the individual charts to confirm what I said rather than their misleading averaged figures.
 
Last edited:

Red Hawk

Diamond Member
Jan 1, 2011
3,266
169
106
the 7970 limits it's boost to 1050mhz, the 770 has no such limitation . Oc'ing both is certainly more meaningful than comparing a card that truly limits it's clocks to one that says it has a max boost clock, but in reality, really doesn't .

I'm not an Nvidia user, so I don't know example how Nvidia's GPU turbo works, but it seems clear that the 770 has specific number for how high it turbos. There is certainly value in comparing overclocked GPUs, but the point remains that if those clocks were reliable they would be sold like that. Comparing what you are guaranteed when you buy the product is the most valid comparison there can be.

Anyways, amenx was talking about memory size and memory bus width, which are not directly related to core clock speed. Even if overclocking the 7970 gives it more of an advantage, it is not the memory bus or memory size that grants this advantage, which is what I think amenx's point was.
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
I think you're missing his point. Yes, the 7970 GHz is better performance per dollar, but looking at the underlying components -- for 50% more memory and a 50% larger memory bus, the 7970 only gets that 3% advantage. (And I'd say that a measly 3% advantage counts more as "on par"). The memory bus and memory size are not pulling through as a decisive factor in AMD's favor.

Depends on the metric used. They are on par in gaming workloads. Tahiti's "extra resources" make it hugely faster in others.
 

Red Hawk

Diamond Member
Jan 1, 2011
3,266
169
106
Depends on the metric used. They are on par in gaming workloads. Tahiti's "extra resources" make it hugely faster in others.

That is true, but irrelevant as we are specifically talking about high-res/multi-monitor gaming.
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
That is true, but irrelevant as we are specifically talking about high-res/multi-monitor gaming.

It's not irrelevant, it's the reason Tahiti has the extra resources you mention. You can't justify a more expensive card that doesn't perform any better in gaming and is less capable overall by leaving it out, and call that a reasonable comparison.
 
Last edited:

Red Hawk

Diamond Member
Jan 1, 2011
3,266
169
106
It's not irrelevant, it's the reason Tahiti has the extra resources you mention. You can't justify a more expensive card that doesn't perform any better in gaming and is less capable overall by leaving it out, and call that a reasonable comparison.

It's irrelevant when talking about hi-res gaming. When making an overall purchasing decision, sure it can factor in.
 

3DVagabond

Lifer
Aug 10, 2009
11,951
204
106
It's irrelevant when talking about hi-res gaming. When making an overall purchasing decision, sure it can factor in.

I think that the fact that Tahiti has more transistors is irrelevant. :p

What you are doing is saying that GK104 is more efficient at gaming loads than Tahiti by giving appreciably the same performance with less transistors. That's not news though. It's the one metric that GK104 has beat Tahiti in since day one. It used to beat it in price as well, which made it a good value for gamers. That day is long since gone, though.
 

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
Look at the individual charts to confirm what I said rather than their misleading averaged figures.

When did anyone make a conclusion that 384-bit bus cards are automatically faster than 256-bit bus cards? That in itself is an erroneous statement that has been made in this thread. We know for instance that GTX680 is faster than 384-bit bus HD7950.

Yet, the 770 beat (or equaled) the 7970GE in 4 games vs 2 @ 2560x1440 and in 5 games vs 1 @ 5760x1200. After this I will never look at averaged performance of games in cards, incl that of TPU. But rather look at each individual games performance to arrive at what I think would be a clearer view a cards overall performance.

Did you actually analyze the significance of those "wins"?

770 vs. 7970GE - 2560x1440
BF3 = +1 fps
Bioshock = +0.6 fps
Crysis 3 = +0.8 fps
Far Cry 3 = +4.5 fps
GRID 2 = -14.7 fps
Hitman Absolution = - 7.2 fps

770 vs. 7970GE - 3 monitors
BF3 = tie
Bioshock = +0.5 fps
Crysis 3 = +1.1 fps
Far Cry 3 = +1.0 fps
GRID 2 = +4.2 fps
Hitman Absolution = - 4.1 fps

Those wins 770 has don't mean anything as they are so minor. When 7970GE won, it won by much larger margins which makes it a more consistent card in those games. At triple monitor, both cards are basically tied. You keep ignoring the $100-150 price difference but I have no idea why that even makes sense when 770 cannot provide a higher level of real world playability in these titles.

If you want to ignore averages in GPU reviews on the net, that's up to you. We can't possibly know what games Gamer ABCD plays. We always use averages for that reason. If all you do is play games where NV is faster (i.e., Secret World, Lost Planet 2, Assassin's Creed IV), by all means get an NV card. That is how GPU buying has always worked for each individual consumer. The whole point of looking at averages is to include as many games as possible as it gives a better representation of the card's overall performance in different games.

Also, you say that GTX770 beat HD7970GE but you completely ignored that the wins are small. If you choose to exclude GRID 2 then why shouldn't we exclude Far Cry 3? It doesn't work that way. With 770 you pay $100-150 more for a card that's merely ties or per TPU/Computerbase, loses to the 7970GE at high resolution gaming.

Finally, anyone who is dead serious on playing on triple monitor gaming understands that a single 770/7970GE isn't enough for those titles to begin with.

There is also TechReport which found that 7970GE actually beat 770 in frame rate smoothness.

"Meanwhile, the GTX 770 is in a tougher spot. When it was introduced a couple of weeks ago, its $399.99 price tag undercut the Radeon HD 7970 GHz Edition. The price advantage was especially welcome since the 7970 GHz is apparently still the faster card."

We have then at least 4 reviews (TPU, ComputerBase, Alt+Esc, Tech Report) that all universally showed that 770 cannot beat 7970GE at high resolution gaming on average despite an unjustifiable price increase. If you want to ignore those reviews, it's totally up to you. No matter how you slice it, 770 is hugely overpriced right now and nearly every recent review site confirms this fact.

Comparison of 256-bit vs. 384-bit is really irrelevant since gamers look at features, price and performance. Specs on paper are meaningless in this case.

Anyways, amenx was talking about memory size and memory bus width, which are not directly related to core clock speed. Even if overclocking the 7970 gives it more of an advantage, it is not the memory bus or memory size that grants this advantage, which is what I think amenx's point was.

Ok but this comparison came out of nowhere. Who has claimed in this thread or otherwise that a 384-bit bus card needs to be faster than a 256-bit card? GTX680 vs. 7950 already proved otherwise.

If you have a car with a twin-turbo 3.8L 6-cylinder engine vs. a car with a 6.2L naturally aspirated V8, both approaches can produce a very fast car. We already know you can't directly compare NV and AMD GPUs based on memory bandwidth. Not sure why this is even mentioned but the terrible price/performance of 770 is completely ignored? Sounds like a strawman to me. How AMD or NV derive their performance makes no difference to me as a gamer. If NV releases a Volta card with 2560 CUDA cores that beat's AMD's card with 6000 Shaders, I could care less. I don't buy CUDA cores or shaders --> I buy performance.
 
Last edited:

amenx

Diamond Member
Dec 17, 2004
3,848
2,013
136
RS, I'm with you on analyzing individual game performance to see where cards stack up, but not when they are averaged. There was clearly an anomaly in Grid 2 with 770 sharp performance drop @2560x1440 while it beat the 7970GE at 5760x1200 in same game. That huge double digit FPS anomaly is included in the averaged percentage in the end, which throws off any reasonable assessment of one card vs another.

Some FPS are more valuable than other FPS. Games where FPS are generally low (ie, 30's) vs games where they are high (70-80) cannot be averaged! You can afford to be 10 FPS down in the latter but not in the former! Thats where averaging loses its value as a measuring criteria. As an extreme, how about throwing Quake 3 in there and see one card having 100 FPS more than the other and include that in the averages?

Finally, with ref to the value of the cards, I am not arguing on that. People may have certain preferences in hardware that go beyond value. Yes, I could have gotten a 7970GE cheaper than my 770 but I didnt. And I doubt value as an exclusive factor will figure very high in my future hardware purchases. Just saying.