AMD HD7*** series info

[3DVagabond]

Nothing less than 100% faster than the previous dual gpu card is acceptable in my opinion.
5870 is a bit faster than a 4870x2 and the gtx480 was a bit faster than a gtx295.

With a full process node smaller, both Nvidia and AMD should beat there gtx590 and 6990 cards with a single gpu card.

I will expect the next gen to beat, or come close to, what the 590/6990 would be if they didn't blow the pci-e spec out of the water. I'm not expecting them to double the performance of ~450W designs though. If these specs are true, it looks like AMD is again concerned with perf/W. 28nm @ 300W I would expect could beat 590/6990. Looks like they aren't going to go there, though. At least AMD. We haven't seen anything at all from nVidia on this node. Likely won't for quite a while yet, either.

The 5870 was a bit slower than the 4870x2. I don't recall the 480, to be honest.

I can't get the search function to work for me at the moment (sometimes the latency trying to access stuff from NZ is horrendous), but I recall you saying performance improvement of +30% this gen then +40% next gen. I thought +70%-100% (actually leaning towards ~6870x2 performance for Tahiti XT). Are you changing your position now? You're expecting greater ~100% improvement?

 

[RussianSensation]

How so?

When the GTX 460 was released it was obviously a better buy than the Radeon HD 5830, but not so much when the Radeon HD 6850 was released. It'd be a good statement to say the GTX 460 won against the HD 5830, but not that it won against the 6850. (I'm talking in terms of price/performance.)

I have to agree with the comments regarding GTX460 made by other posters. It was superior to HD6850 and HD6870 for many many months. Most of those GTX460 768/1GB cards were available for $120-180 and lower with rebates. Add far superior overclocking headroom and HD6850/6870 weren't a slam dunk. HD6870 became superior only in the last 3 months or so when its price started to dip to $150-155. Not to mention HD6870 initially was a disappointing card since it launched at around $240 at the time when GTX460 FTW was about $200 and GTX470 was $240.

 

[RussianSensation]

Since we have this wall, squeezing as much performance under a power envelope is not really undermining performance gains at all.

Sure you are if you lower the power envelope to 150W from 200W for 7950 over 6950 and to 190W from 250W for 7970 over 6970.

Also, they are releasing mid-range parts at 120-130W? Of course they are artificially limiting performance by focusing more on power consumption. What's wrong with a mid-range card that pushes 180-190W of performance, and high end card pushing 250W? Based on those specs, it looks like they put way more emphasis on power consumption for this generation, rather than performance which they should have done on the desktop given 28nm transition.

My personal opinion is that delivering cards with similar performance at lower power envelope is not the direction I want on the desktop; but I would rather see higher performance at the same power envelope. If they can't deliver higher performance at lower power envelope, I am perfectly fine with 250W high-end GPUs that offer 75-100% more performance. I would much rather have that than a card with 50-60% more performance at 190W. Desktop = highest performance possible, power consumption secondary. If Kepler is a 250W part and brings 75-100% more performance over GTX580, it would be far more attractive than a 190W HD7970 with only 50-60% more performance -- we are talking high-end here. In my view, for highest end GPUs, performance should be first and foremost, at all costs. You have low-end and mid-range if you want lower power consumption. 250-275W isn't a big deal for a single GPU imo.

 

[DeathReborn]

If they go XDR2 who would manufacture them?

Samsung & Elpida both made the XDR for the PS3 so they'd be likely candidates.

 

[LOL_Wut_Axel]

I have to agree with the comments regarding GTX460 made by other posters. It was superior to HD6850 and HD6870 for many many months. Most of those GTX460 768/1GB cards were available for $120-180 and lower with rebates. Add far superior overclocking headroom and HD6850/6870 weren't a slam dunk. HD6870 became superior only in the last 3 months or so when its price started to dip to $150-155. Not to mention HD6870 initially was a disappointing card since it launched at around $240 at the time when GTX460 FTW was about $200 and GTX470 was $240.

Wat.

The Radeon HD 6850 and 6870 reach 1000MHz if you have a decent sample. The GTX 460 can reach 900MHz with a decent sample, so no. At those speeds, the GTX 460 matches the 6850 and loses to the 6870.

The EVGA GTX 460 FTW wasn't decent at all if you considered overclocking headroom, either. You can only get around 50MHz more from it. The GTX 470, even though it consumes a lot more power, had tons more overclocking headroom.

Comparing factory over-clocked models to reference ones is apples-to-oranges. The GTX 460 was an amazing deal until the HD 6850 arrived.

 

[happy medium]

Are you changing your position now? You're expecting greater ~100% improvement?

No, I still think its gonna be about 40% but I see no reason why they can't have a part much faster at around 250 watts. They went from 55nm to 40nm and doubled performance, now they are going from 40nm to 28nm and going to give us under 50%?

The only thing I can think of is mabe the CPU/GPU things are taking up room and making the chips hotter and more power hungry mabe.

Mabe Russian is right, with all the console ports being released, mabe the power is just not needed?

 

[happy medium]

The Radeon HD 6850 and 6870 reach 1000MHz if you have a decent sample. The GTX 460 can reach 900MHz with a decent sample, so no. At those speeds, the GTX 460 matches the 6850 and loses to the 6870.

wrong.....
A stock 675 core gtx460 at 900 core (33% overclock) will beat a 10% overclocked 6870 at 1000 core.
A stock 675 core gtx460 @ 730 core = a 6850.
My gtx460 rules all.

 

[RussianSensation]

Wat.

The Radeon HD 6850 and 6870 reach 1000MHz if you have a decent sample. The GTX 460 can reach 900MHz with a decent sample, so no. At those speeds, the GTX 460 matches the 6850 and loses to the 6870.

Very few HD6850/6870s reach 1000mhz, while most GTX460s reach 900mhz. In fact, many GTX460s reach 925-950mhz. But that's besides the point since HD6850/6870 cards launched way too late after GTX460 and at too high prices. Performance wise, at 1920x1200 4AA/16AF, an HD5870 is 13% faster than an HD6870.

A GTX460 @ 900mhz competes with an HD5870.

So even if we assume you have a good HD6870 @ 1000mhz, that's only an 11% overclock, which still puts it below HD5870 in performance. So no an overclocked GTX460 is not slower than an overclocked HD6870. But GTX460 has sold below the price of an HD6870 for a while (except last 3 months). The HD6850 wasn't even in the picture really since GTX460 768mb sold for prices far below the HD6850 and a 5 min MSI afterburner overclock on the GTX460 1GB made it better than the HD6870 for most of its life since you could have it for $130-160 AR.

Not to mention most GTX460s shipped with free games like Metro 2033, Just Cause 2, Mafia 2. AMD really butchered the entire HD6850/6870 launch. Those cards were pretty underwhelming until HD6870 came into the picture in the last quarter with $150 pricing + a bunch of games. HD6750/6770 were horrendous rebadges, HD6850 was too slow and priced too high most of its life and HD6870 launched at too high of a price in the beginning. HD6970 became pretty much irrelevant since you could unlock an HD6950. Overall, HD68xx/69xx generation was pretty weak since they gave up the performance lead that HD5850/5870 had to refresh GTX4xx line. This entire generation, AMD's best card was HD6950 2GB without a doubt. The rest was nothing special since NV had an answer at every other price point.

 

[3DVagabond]

No, I still think its gonna be about 40% but I see no reason why they can't have a part much faster at around 250 watts. They went from 55nm to 40nm and doubled performance, now they are going from 40nm to 28nm and going to give us under 50%?

The only thing I can think of is mabe the CPU/GPU things are taking up room and making the chips hotter and more power hungry mabe.

Mabe Russian is right, with all the console ports being released, mabe the power is just not needed?

I don't think AMD has the desire to make 250W chips. I think they only did it because they were stuck on 40nm with a 32nm design. New CEO now though. They might go in a completely different direction. We'll have to wait and see about that.

Working from the info posted, which is just as likely to be totally made up as not, GCN isn't power hungry or inefficient.

 

[blackened23]

Very few HD6850/6870s reach 1000mhz, while most GTX460s reach 900mhz. In fact, many GTX460s reach 925-950mhz. But that's besides the point. At 1920x1200 4AA/16AF, an HD5870 is 13% faster than an HD6870.

A GTX460 @ 900mhz competes with an HD5870.

So even if we assume you have a good HD6870 @ 1000mhz, that's only an 11% overclock, which still puts it below HD5870 in performance. So no an overclocked GTX460 is not slower than an overclocked HD6870. But GTX460 has sold below the price of an HD6870 for a while (except last 3 months). The HD6850 wasn't even in the picture really since GTX460 768mb sold for prices far below the HD6850 and an overclocked GTX460 1GB spanked HD6850 for similar price ($130-150 AR) price.

I thought 5870 and 6870 were basically the same, with the latter having lower power consumption / improved tesselation / xfire scaling. The cayman GPU is only used in the 69xx series, 68xx is the same as 58xx with minor improvements and better power consumption

Or thats what I seem to remember reading, need a memory refresh here

 

[blackened23]

AMD Radeon HD 5870
Stream Processors 1600
Texture Units 80
ROPs 32
Core Clock 850MHz
Memory Clock 1.2GHz (4.8GHz effective) GDDR5
Memory Bus Width 256-bit
Frame Buffer 1GB
FP64 1/5
Transistor Count 2.15B
Manufacturing Process TSMC 40nm
Price Point ~$349

AMD Radeon HD 6870
Stream Processors 1120
Texture Units 56
ROPs 32
Core Clock 900MHz
Memory Clock 1.05GHz (4.2GHz effective) GDDR5
Memory Bus Width 256-bit
Frame Buffer 1GB
FP64 N/A
Transistor Count 1.7B
Manufacturing Process TSMC 40nm
Price Point $239

Well then , nevermind

 

[happy medium]

I thought 5870 and 6870 were basically the same, with the latter having lower power consumption / improved tesselation / xfire scaling. The cayman GPU is only used in the 69xx series, 68xx is the same as 58xx with minor improvements and better power consumption

Or thats what I seem to remember reading, need a memory refresh here

No, I think it goes ,gtx460, 6850, 5850, 6870, gtx470/gtx560ti, 5870, 6950, (gtx480/gtx570/6970), gtx580

In that order but very close in performance.

 

[LOL_Wut_Axel]

wrong.....
A stock 675 core gtx460 at 900 core (33% overclock) will beat a 10% overclocked 6870 at 1000 core.
A stock 675 core gtx460 @ 730 core = a 6850.
My gtx460 rules all.

Nope. GTX 460 clocked 100MHz lower than Radeon HD 6850=equivalent speed (see stock HD 6850 @775MHz vs GTX 460@675MHz). With the HD 6870 that goes up as it delivers ~10% better performance at the same clock speed as the 6850.

900MHz GTX 460=1000MHz HD 6850
900MHz GTX 460<1000MHz HD 6870

Assuming these max over-clocks, the GTX 460 would have 4% higher over-clocking headroom than the HD 6850, but then the 6850 stock is 3% faster than the GTX 460. I'm not really here to argue over a 1% difference, nor a 3% difference. The GTX 460 and HD 6850 are for all intents and purposes equivalent stock and over-clocked.

 

[3DVagabond]

Very few HD6850/6870s reach 1000mhz, while most GTX460s reach 900mhz. In fact, many GTX460s reach 925-950mhz.

We had a poll done here that showed that "most 460's will do 850Mhz+" is not true. I do agree though that most 6800 cards won't do 1GHz. "Very few" though is probably an exaggeration.

 

[happy medium]

Nope. GTX 460 clocked 100MHz lower than Radeon HD 6850=equivalent speed (see stock HD 6850 @775MHz vs GTX 460@675MHz). With the HD 6870 that goes up as it delivers ~10% better performance at the same clock speed as the 6850.

900MHz GTX 460=1000MHz HD 6850
900MHz GTX 460<1000MHz HD 6870

Assuming these max over-clocks, the GTX 460 would have 4% higher over-clocking headroom than the HD 6850, but then the 6850 stock is 3% faster than the GTX 460. I'm not really here to argue over a 1% difference, nor a 3% difference. The GTX 460 and HD 6850 are for all intents and purposes equivalent stock and over-clocked.

See post #158

 

[saratoga172]

*facepalm*

They really need to be more original with videocard nomenclatures.

I imagine it must drive Best Buy employees nuts trying to explain to customers why an 8800GT was a better card than a 9400.



Technically is an HD 7950 or whatever, it's not exactly like BMW releasing a C240.

I wonder if most Best Buy employees actually know that an 8800GT was a better card than a 9400?

 

[cusideabelincoln]

They are releasing mid-range parts at 120-130W, so of course they are artificially limiting performance. What's wrong with a mid-range card that pushes 180-190W of performance? Sorry, this is 28nm process. I expect higher performance for mid-range and lower power consumption. Based on those specs, it looks like they put way more emphasis on power consumption for this generation.

The HD6850 has a 127W TDP. So the current generation mid range card from AMD fits into that 120-130W market you just said. Judging from the specs, the next generation mid-range HD 7850, at the 120-130W power envelope, will be about 40&#37; faster.

I don't follow why you expect higher performance and lower power consumption. You throw too many variables into play because "higher" and "lower" don't account for "how much" higher or lower. The fact is, at the same power consumption levels, you are getting higher performance, in the mid range, by what looks like the count of 40%. If you want more performance at lower power consumption, you aren't going to get as high of performance as you would at the same power consumption. There's really just no way around that, unless they do something magical. Just look at recent history and compare cards.

4850->5850: Big performance jump, but a small power jump.
4850->5750: Small performance jump, lower power.
4670->5750: Big performance jump, more power.

Do you see the correlation? You can't exactly have your cake and eat it too. Well, it's really more like you can't have your cake, ice cream, and candy bars. Because you can get more performance at lower power. You just can't get extra more cake at lower power. But extra more cake at the same power is to be expected. Now for extra, extra more cake you will need more power.

I also don't see why you expect the mid range cards to push 180-190W. The 5000 and 6000 cards did not do that. Even the GTX 460 and 560 don't do that. Hell the previous gen high end card, HD 4890, was rated at 190W.

My personal opinion is that delivering cards with similar performance at lower power envelope is not the direction I want on the desktop. If they can't deliver higher performance at lower power envelope, I am perfectly fine with 250W high-end GPUs that offer 75-100% more performance. I would much rather have that than a card with 40-50% more performance at 190W. Desktop = highest performance possible, power consumption secondary. If Kepler is a 250W part and brings 75-100% more performance over GTX580, it would be far more attractive -- we are talking high-end here.

Highest performance possible? POSSIBLE? That is a loaded statement. Because what you're asking for has far more variables than just power and performance. Time for development and price also come into play.

 

[LOL_Wut_Axel]

long rant with no real meaning

Not all GTX 460's reach 900MHz, but most samples should. Same for the HD 6850 at 1000MHz. Some 6850s reach 1050MHz yet you don't see me arguing over them as there's too little of them; same as GTX 460s reaching 950MHz.

Your games, rebates and what not argument is meaningless. Like I've said a million times, you still have to pay upfront price, and whether a game comes or not depends on the manufacturer.

 

[notty22]

My take on the 6850/70 launch is similar to Russians, I debated , debating history earlier. The faster 6870's was equal or faster in half the games in comparison to the 850mhz gtx 460's. They were also close to 240.00, while Nvidia lobbed another 20 dollars of a big lineup of custom 460's.
That was also when the discussion about the AMD driver that lowered quality slightly was brought up. It was the launch driver for 6850/70. So things were that much closer, what raised original questions around this driver, was it made the 5850/70 also suddenly almost 10&#37; faster.

 

[happy medium]

My take on the 6850/70 launch is similar to Russians, I debated , debating history earlier. The faster 6870's was equal or faster in half the games in comparison to the 850mhz gtx 460's. They were also close to 240.00, while Nvidia lobbed another 20 dollars of a big lineup of custom 460's.
That was also when the discussion about the AMD driver that lowered quality slightly was brought up. It was the launch driver for 6850/70. So things were that much closer, what raised original questions around this driver, was it made the 5850/70 also suddenly almost 10% faster.

Yea I remember those debates and saved a few slides for old time sake.
The gtx460 and 6870 traded blows while the 6850 is left in the dust.

 

[blackened23]

No, I think it goes ,gtx460, 6850, 5850, 6870, gtx470/gtx560ti, 5870, 6950, (gtx480/gtx570/6970), gtx580

In that order but very close in performance.

My point was that 6870 wasn't cayman based and AMD changed their naming conventions.

6950/6970 were the true successors to the 5870, the 6870 wasn't. In fact in most scenarios, the 5870 should be about 10&#37; faster than the 6870....6870 just has better power consumption / better tesselation. As the 6870 wasn't cayman based. It stands to reason that 7970 will be the enthusiast card to get, and 7950 if you're short on cash.

 

[LOL_Wut_Axel]

Yea I remember those debates and saved a few slides for old time sake.
The gtx460 and 6870 traded blows while the 6850 is left in the dust.

I like the part where you only show the games where the GTX 460 either matched or won by a slight amount, and only the ones which showed the GTX 460 having a lesser performance hit from AA.

As you can see, overall the GTX 460 at 900MHz is slower than the Radeon HD 6870 at 1000MHz. It's also either the same speed or a bit quicker than the HD 6850.

 

[RussianSensation]

I thought 5870 and 6870 were basically the same, with the latter having lower power consumption / improved tesselation / xfire scaling. The cayman GPU is only used in the 69xx series, 68xx is the same as 58xx with minor improvements and better power consumption

Or thats what I seem to remember reading, need a memory refresh here

Ya, pretty much with some minor changes (HDMI 1.4, UVD3.0, some reduction in unneeded shaders / TMUs). But my main concern is that AMD is simply looking to shift VLIW-4 HD6950/6970 chips to 28nm HD7850 / 7870 chips. Sure they might get more than 50% reduction in power consumption but that means almost no performance improvement. Is that what the consumer wants for discrete desktop GPUs? I think 20-30% more performance in the mid-range at 170-180W power envelope would have been far more preferable imo. It's good to keep increasing mid-range performance since a lot of buyers want these cards.

 

[cusideabelincoln]

Very few HD6850/6870s reach 1000mhz, while most GTX460s reach 900mhz. In fact, many GTX460s reach 925-950mhz. But that's besides the point since HD6850/6870 cards launched way too late after GTX460 and at too high prices. Performance wise, at 1920x1200 4AA/16AF, an HD5870 is 13&#37; faster than an HD6870.

A GTX460 @ 900mhz competes with an HD5870.

So even if we assume you have a good HD6870 @ 1000mhz, that's only an 11% overclock, which still puts it below HD5870 in performance. So no an overclocked GTX460 is not slower than an overclocked HD6870. But GTX460 has sold below the price of an HD6870 for a while (except last 3 months).

wrong.....
A stock 675 core gtx460 at 900 core (33% overclock) will beat a 10% overclocked 6870 at 1000 core.
A stock 675 core gtx460 @ 730 core = a 6850.
My gtx460 rules all.

http://www.anandtech.com/show/4137/...-6950-1gb-xfxs-radeon-hd-6870-black-edition/5

A 6870 @ 940 MHz core speed is awfully close to the 5870, and trades blows with the 560 Ti depending on the game.

http://www.techpowerup.com/reviews/Powercolor/HD_6870_PCS_Plus_Plus/23.html

A 6870 @ 975 MHz core is again trading blows with the 560 Ti, and has a very small edge over the 5870.

The HD6850 wasn't even in the picture really since GTX460 768mb sold for prices far below the HD6850 and a 5 min MSI afterburner overclock on the GTX460 1GB made it better than the HD6870 for most of its life since you could have it for $130-160 AR.

I still believe those 768MB cards are going to really feel the pain soon, and they already are in a few games.

And most 460s reach 900 MHz? As in more than half? I really doubt that.

Not to mention most GTX460s shipped with free games like Metro 2033, Just Cause 2, Mafia 2. AMD really butchered the entire HD6850/6870 launch. Those cards were pretty underwhelming until HD6870 came into the picture in the last quarter with $150 pricing + a bunch of games. HD6750/6770 were horrendous rebadges, HD6850 was too slow and priced too high most of its life and HD6870 launched at too high of a price in the beginning. HD6970 became pretty much irrelevant since you could unlock an HD6950. Overall, HD68xx/69xx generation was pretty weak since they gave up the performance lead that HD5850/5870 had to refresh GTX4xx line. This entire generation, AMD's best card was HD6950 2GB without a doubt. The rest was nothing special since NV had an answer at every other price point.

The 6850 competed in price with the GTX 460 most of the time. Most 460s didn't undercut the 6850 significantly until the 560 came around.

And you have it half-backwards. AMD had an answer at every other price point as well. It's what you call competition.

 

[blackened23]

Ya, pretty much with some minor changes (HDMI 1.4, UVD3.0, some reduction in unneeded shaders / TMUs). But my main concern is that AMD is simply looking to shift VLIW-4 HD6950/6970 chips to 28nm HD7850 / 7870 chips. Sure they might get more than 50% reduction in power consumption but that means almost no performance improvement. Is that what the consumer wants for discrete desktop GPUs? I think 20-30% more performance in the mid-range at 170-180W power envelope would have been far more preferable imo. It's good to keep increasing mid-range performance since a lot of buyers want these cards.

Totally agree with you on that point.