First GTX 670 review(s) up (tt & oc.net) * TT OC review added*

[aaksheytalwar]

I have my doubt whether this was really a 670. They never mentioned the clocks or brand or anything. Prolly just a 680 and scaling can't be that bad that 10% lower clocks doesn't affect performance at all.

 

[blastingcap]

How much credibility does an index have if it places a 285 below a 5770? Please stop embarrassing yourself. http://www.techpowerup.com/reviews/HIS/HD_5770/30.html

Check these indexes, the enthu index. 5850 70+ while 285 didn't score 40 and was below 5770 as well.

http://www.tomshardware.com/charts/...compare,2674.html?prod[4823]=on&prod[5254]=on
Enthusiast Index
Tom's Hardware Index
(Index of all "Enthusiast" Results)
Score in percentGo 70.76 38.97

Gamer Index
Tom's Hardware Index
(Index of all "Gamer" Results)
Score in percentGo 104.66 67.27

Index means average of all games and benches, not just one benchmark

 

[hdfxst]

why would he use such old drivers?and does 290.53 drivers support a 670?I thought only the 300 something drivers supported the 600 series

 

[aaksheytalwar]

This is too old. Check the games Tom and at used. Use proper games, not games which aren't intensive at all. Check modern indexes of 2011 and you will know. Shouldn't be so difficult to understand. If you see reviews which were released in 2009, both are nearly equal. If you check games of 2010 late or 2011, the 5850 pwns the 285.

 

[joshhedge]

I wish AMD wasn't so damn conservative with their clock speeds for this generation, I hope they pull something amazing out of the bag for the 8xxx series cards.

 

[Mad_dawg]

I think the point is weak and already explained why.

Agreed... ()

 

[blastingcap]

This is too old. Check the games Tom and at used. Use proper games, not games which aren't intensive at all. Check modern indexes of 2011 and you will know. Shouldn't be so difficult to understand. If you see reviews which were released in 2009, both are nearly equal. If you check games of 2010 late or 2011, the 5850 pwns the 285.

Look, I bought a 5850 about a week or so after it launched. I know how fast it was. Yes it was faster than 285 but by only ~8% stock vs stock. When both are overclocked the 5850 pulls away, especially in some of the heavier games like you said such as Crysis.

But you said it "blew away" the 285 which isn't that accurate unless you think ~20-25% faster oc vs oc (even with updated drivers and benching heavier games) is "blew away." Maybe you do, but for me a card has to be something like >40% faster oc vs oc to "blow away" another card. To me, 20-25% faster is a solid beating, but not a "blow away" level of beatdown. I think maybe we just disagree on wording there.

But you are wrong to say the 5770 is faster than the 285; that is just plain laughable unless you mean the 5770 is better at DirectX 11 in which case that isn't really a fair comparison, is it?

 

[aaksheytalwar]

Techpowerup only tested ancient games unlike toms

 

[aaksheytalwar]

I can't help you if you don't read the links carefully.

Mafia 2 - Gamer
Medium Settings
(1680x1050, no AA, 4xAF)
Score in fpsGo 109.40 78.10


Call of Juarez - Gamer
Medium Settings
(1680x1050, no AA)
Score in fpsGo 74.70 45.30



Metro 2033 - Enthusiast
Very High Settings
(1920x1080, 4xMSAA, 16xAF)
Score in fpsGo 21.20 17.50

Call of Pripyat - Enthusiast
Ultra Settings, Tessellation on
(1920x1080, 4xMSAA, SSAO)
Score in fpsGo 31.21 19.00


Mafia 2 - Enthusiast
Maximum Settings
(1920x1080, MSAA, 16xAF)
Score in fpsGo 52.10 41.60

Metro 2033 - Gamer
Medium Settings
(1680x1050, AAA, 4xAF)
Score in fpsGo 69.60 54.10


Call of Juarez - Enthusiast
Maximum Settings
(1920x1080, 4xMSAA)
Score in fpsGo 56.00 36.10

Enthusiast Index
Tom's Hardware Index
(Index of all "Enthusiast" Results)
Score in percentGo 70.76 38.97


In each of the modern games including crysis 5850 blows it away. In games of that time 5850 was only 10% faster. But being newer tech it showed its mettle with newer games With old games both are nearly equal but in 2011 it all changed. I have proved this. I hope you understand else I can't help you. And I too had a 5850 and it was like twice as fast as a 4850.

 

[Silverforce11]

Look, I bought a 5850 about a week or so after it launched. I know how fast it was.

How much faster is a gtx480 vs 5870 (850mhz core)?

Because a 5850 @ 1ghz is like a 5870 +10% (not perfect scaling).

 

[blastingcap]

@ash:

Are you on a mission to increase your postcount as fast as possible or something?

I'm still laughing at your saying the 5770 was faster than the 285. Seriously?

You talk about ancient games but you include some real whoppers yourself. Here's a modern comparison: http://www.xbitlabs.com/articles/graphics/display/battlefield-bad-company-2_5.html#sect1

BFBC2 uses Frostbite 1.5.

Other comparisons: http://www.xbitlabs.com/articles/graphics/display/radeon-hd5850_6.html#sect0

Please don't cherrypick. Lots of good games were built on older engines anyway like Unreal 2.5/3 so it's a little biased to simply exclude everything that doesn't use a more modern engine. Besides, if you were serious about excluding older engines, you would exclude Crysis. Or if you wanted to include older-but-stressful games like Crysis, you should include Far Cry 2 as well: http://www.xbitlabs.com/articles/graphics/display/radeon-hd5850_7.html#sect0

Remember to look at results with AA and AF turned on, too, not bare benches. I have no idea why you are posting benches of games on Medium setting, either. Try High.

How much faster is a gtx480 vs 5870 (850mhz core)?

Because a 5850 @ 1ghz is like a 5870 +10% (not perfect scaling).

The problem with comparing oc's is each chip oc's differently... so you can wind up with mismatches if you compare a dud NV chip to a stud AMD chip. ludicrous to say the typical reference 5850 could hit 1GHz. Mine sure could not; it crapped out before 900MHz IIRC.

 

[aaksheytalwar]

I rest my case. From 12x10 to 19x12, the minimum fps of the 5850 are a good 20% more. That is the real performance. And it is anyway playable only till 16x10 so going further makes no sense. And till19x12 min fps of 5850 are a good 20% more making tolerable gameplay to comfortable gameplay class.


1. Far cry 2 is old and not even remotely in the same league as crysis.

2. With aa off, the 5850 blows the 285 away. With aa on both are close but neither is playable so it is irrevelant, besides far cry 2 isn't a modern game while crysis can be counted as one.

3. In bad company 2, 5850 stills blows it away due to min fps

 

[aaksheytalwar]

And no, for most games a 5770 may not be faster but for many games it is

 

[aaksheytalwar]

And toms did an index specially for 2011 only, no other online source can come close to that

Please merge post

 

[Maximilian]

Would it have killed them to bench a few decent games? BF3 and civ V is are the two most modern non console ports available, every GPU review should bench these games. Crysis wouldnt hurt either

 

[LOL_Wut_Axel]

This card is too powerful for NVIDIA's own good. There's pretty much no reason to get a GTX 680 now, considering its limited overclocking and the fact they use exactly the same RAM chips. Remember what I said about clock scaling not being that great with GK104? That's why the lower clock speed doesn't mean much in terms of performance, and given the fact the card will probably clock nearly as well as the 680 that means the performance difference clock-for-clock will be 2-3%, if not lower.

I don't like NVIDIA, but this is the same speed as the HD 7970 and if the price lowers to $350 as we get to the end of the year it's gonna be very tough to get a 7950 over this. Also, the power consumption test they did is wrong. The HD 7970 most definitely consumes more power stock.

If this thing ever comes to $350 or under this year I think I'll have to get this instead of the 7950. For the 7950 to catch up it'd need to be overclocked to 950-1000MHz, and at that clock speed you won't be able to undervolt. This also seems to consume a decent amount less power than the GTX 680. Yeah, the compute performance from the HD 7900 series is nice if you're one of the very few that will take advantage of it and yes, AMD is a better/nicer/less shifty company than NVIDIA, and yes AMD isn't restricting us when it comes to tweaking our cards, but stock this card is too good.

 

[f1sherman]

although this might look that it substantiates your Kepler-bad-scaling theory,
dont forget that the card is not only clocked lower but it has 1/8 less cuda cores,

so we could just as well claim Kepler scaled badly with CC.

 

[aaksheytalwar]

At did crysis and the 5850 blew the 285 there as well. Even in bf bc2 5850 was significantly faster

And can't do bf3 because 285 doesn't do dx11, again pwnd

Acc to AT, owned in the following

Crysis
Metro
Starcraft
At every setting
Won = zero

Diff is 20-50% through the board

http://www.anandtech.com/bench/Product/512?vs=520


Now let us compare 285 to 5770
Mafia 2, 5770 pawned
Call of Juarez, 5770 10% faster
Metro, 285 10% faster
Gamer index, tie
Call of Pripyat, tie with 5770 winner
Mafia 2 extreme settings both are unplayable but285 is better but irrelevant as both are unplayable
Metro max playable, 285 slightly better
Call of Juarez, 5770 much better but both unplayable
Enthusiast index, 5770 pwns 285

So both are very comparable cards and no point going for higher settings as these are close to max playable already or often unplayable so no point going to sub 5fps range

285 trades blows with 5770 in modern games, like 10% better without dx11 etc.

5850 trades blows with 460 and 470 both of which pwn 285

http://www.tomshardware.com/charts/...tml?prod[4823]=on&prod[4822]=on&prod[5254]=on

 

[LOL_Wut_Axel]

although this might look that it substantiates your Kepler-bad-scaling theory,
dont forget that the card is not only clocked lower but it has 1/8 less cuda cores,

so we could just as well claim Kepler scaled badly with CC.

It has decent scaling, just not as "great" like the HD 7950/70. Again, if Kepler scaled better, the performance spread between both would be over 10% constantly given the loss of compute units.

In this case, I think that's good. Not that great clock scaling=very little performance lost, which means the GTX 670 effectively makes the 680 meaningless. It just needs to be priced right. Now, this is not anywhere near the same situation as the GTX 570 and 580, because the 580 overclocked a lot better because of the superior power delivery system and it scaled very well with that additional frequency. Here the clocks don't scale as well, and the loss of compute units means little, so what we get is a very small performance difference. Clock-for-clock, again, the difference should be less than 2-3%.

 

[bryanW1995]

1. It is too close, perhaps not a 670 in reality.

2. Point 1
Else

3. An oc so high that nearing its oc potential and prolly with a msrp of $460-480 or so with a 10-15% oc over stock with not much headroom.

Next

1. 5850 competed with 285 and blew it away, not 4890. 5870 wasn't even needed and competed close to 295, tho much slower

2. 6970 competed against 480 and won over. 580 released a lot later so that is still just half a victory due to the delay. For half the time amd had won anyway before the 580 launched

3. Once 580 launched, for quite a while nvidia won, but after that 7970 was the winner for a good 3 months or so

6. 680 oc and 7970 oc are so close that nobody wins or loosely in absolute true sense, they are head on, like 570 and 6970 had 580 not been released ever for example sake

7 and ya, 260 competed with 4870 and amd didn't have anything for 280, by the time 5xxx was out only then was 285 released.

8 nvidia typically lasts 2 years, first 6 months first gen and next 18 months the refresh. But approx 2 years cycle. Amd lasts 15 months approx. so due to diff cycles you can't exactly compete all the time as diff winners at diff times.

9 loosing for 4 months and then winning for 6 months is actually just tieing and not really winning

You're wrong on this point. gtx 580 released a few weeks BEFORE 6970.

 

[Silverforce11]

Dang, you guys sure know how to take a thread about bacon and end up talking about wine.

 

[RussianSensation]

1. 5850 competed with 285 and blew it away, not 4890. 5870 wasn't even needed and competed close to 295, tho much slower

GTX285 was from HD4890 generation. Fermi was delayed by 6 months but a real competitor to the HD5850 was GTX470, not GTX285. Although HD5850 cost less so it really didn't have a direct competitor until GTX470 fell in price later in its life.

2. 6970 competed against 480 and won over. 580 released a lot later so that is still just half a victory due to the delay.

No. HD6970 competed against GTX570 not 580. You can say all you want that HD6970 competed against 580 and it's not correct since $370 videocards don't compete against $500 cards. Also, the performance of the 6970 was similar to the 570 and the price was as well.

GTX570/580 launched 1 month before HD6950/6970 series (November 9 for NV, December 14 for AMD).

GTX480's competitor was HD5870. Fermi got delayed by 6 months though. See above.

7 and ya, 260 competed with 4870 and amd didn't have anything for 280, by the time 5xxx was out only then was 285 released.

No.

GTX285 Launched January 15, 2009
HD5850 Launched September 23, 2009

nvidia typically lasts 2 years, first 6 months first gen and next 18 months the refresh. But approx 2 years cycle. Amd lasts 15 months approx. so due to diff cycles you can't exactly compete all the time as diff winners at diff times.

Totally not true. I have no idea where you pulled this estimate from, especially if we go back to GeForce 2. You can't make such a blanket statement. Also, even in recent history your estimates are way off:

8800GTX 768 = November 8, 2006
9800GTX (refresh) = March 31, 2008 (that's way longer than 6 months for a first gen part)

9800GTX refresh only lasted a quarter before next generation appeared, not 18 months as you state.

GTX280 = June 16, 2008 (less than 2 years after 8800GTX)
GTX285 = January 15, 2009 (6 months after 280)

GTX480 = March 23, 2010 (that's 14 months after 285, not 18 months)
GTX580 = November 9, 2010 (that's about 8 months for a refresh not 6 months as you state).

loosing for 4 months and then winning for 6 months is actually just tieing and not really winning

Winning means 63% desktop discrete market share despite launching 6 months late with Fermi and in the process making more $ in any quarter in 2011 than AMD made in the entire 2011. Any company in the world can launch a product, undercut the competitor and lose $ quarter after quarter. Also, having single fastest GPU with 8800GTX, 280/285, 480/580 can also be considered winning. Although I think having amazing price/performance is also winning (4850/4870/GTX460/5850/6950 did that).

Overall, NV has been winning the desktop discrete based on profits and market share while AMD has been doing better on the mobile/laptop side. AMD was in the game with 4800/5800/6900 series because of lower power consumption and better price/performance. HD7800/7900 series offer none of those advantages and it shows as now a $400 GTX670 will trade blows with a stock 7970. Not looking good for AMD.

GCN architecture is overhyped. The main reason it looks good is because of the 28nm shrink. People kept comparing 40nm 580 to the 7970 and highlightning how much more advanced the 7970 was. In fact, what they were highlighting was how much superior 28nm node is compared to 40nm.

Now that Kepler has been manufactured on 28nm, it's clear that GCN is still just a 1st gen compute architecture for AMD while Kepler is a 2nd generation. As an architecture GCN 1.0 trails Kepler by miles in FP16 texture and tessellation performance as well as power consumption, performance/watt, performance/transistor, you name it. In terms of tessellation and texture performance, GCN is about a generation behind Kepler. NV is able to keep up despite just 192 GB/sec bandwidth and 1344 SPs. Ouch. I hope AMD delivers with GCN 2.0 i.e., HD8000 series or NV will sell us GTX760 for $400 next round.

 

[Don Karnage]

No BF3 tests? This would be amazing if this pushes amd card prices even lower.

BTW this card at 399.99 smokes the 7970 at 479.99

 

[blastingcap]

although this might look that it substantiates your Kepler-bad-scaling theory,
dont forget that the card is not only clocked lower but it has 1/8 less cuda cores,

so we could just as well claim Kepler scaled badly with CC.

Hey fish what is the fastest graphics card you own btw?

 

[blastingcap]

Dang, you guys sure know how to take a thread about bacon and end up talking about wine.

I'm not responding to that guy anymore; he is a man on a mission, even claiming that "for most games a 5770 may not be faster but for many games it is" (compared to gtx285).