Radeon HD 6xxx

[dhanson865]

look at it another way

Code: Select all

The TDP is reference design TDP values from AMD.
Different non-reference board designs from vendors may lead to slight variations in actual TDP.

Card Load Idle GFLOPS  GFLOPS/Watt
5670  64W 15W    620       ~9.7
5750  86W 16W   1008      ~11.7
5770 108W 18W   1360      ~12.6

6850 127W 15W   1488      ~11.7

The 6850 is about 47% faster but uses about 47% more power than the 5750. That is not a gain in any sense for SPCR points of view. The 6850 vs the 5770 is even worse, it's over the TDP limit for a quiet retail card and on top of that you get a 9% gain in performance for a cost of 18% more power. That's really going backwards.

[Modo]

Comparing GFLOPS values for different architectures is a great way to not get anywhere. Look at some benchmarks—the actual rendering performance increase is many times that of the power draw rise.

[dhanson865]

Comparing GFLOPS values for different architectures is a great way to not get anywhere. Look at some benchmarks—the actual rendering performance increase is many times that of the power draw rise.

GFLOPS is just another benchmark, every benchmark you choose will give a different percentage increase. You have to test with the games you care about at the resolution you care about with the CPU you use, and the SSD/HDD, and so on to get specific comparisons. I don't play Crysis: Warhead only in the Frost zone so I say Don't miss the forest for the trees (look at the big picture).

In addition I'm not a fan of those charts as they left the 5750 out and that is my target for performance/TDP comparisons.

Here is the longer list updated with the proper numbers for the 68xx cards and the new numbers for the 69xx cards.

Code: Select all

Card    GFLOPS TDP
HD 5670  620   64W
   6670 1152   63W  (numbers/name could be wrong, unreleased card)
HD 5750 1008   86W
HD 5770 1360  108W
HD 6850 1488  127W (new)
HD 5830 1792  175W
HD 5850 2088  151W
HD 6870 2016  151W (new)
HD 5870 2720  188W
   6950 2720  198W (numbers could be wrong, unreleased card)
   6970 3264  220W (numbers could be wrong, unreleased card)
   6990 5440  285W (numbers could be wrong, unreleased card)

assuming those TDP numbers hold the 6990 should be pretty loud.

Give me a better way of comparing the cards GPU power vs TDP and I'll do it. I'm sure a few months down the road we'll have enough benchmarks to make that comparison in different ways but I'm not happy with Anandtechs first shot at it.

[Modo]

GFLOPS is just another benchmark, every benchmark you choose will give a different percentage increase.

The difference is quite big in every game benchmark on every site. It's a lot more than the modest GFLOPS increase. You'd know that if you'd bothered to check instead of repeating your assumptions. Have a nice life.

[dhanson865]

GFLOPS is just another benchmark, every benchmark you choose will give a different percentage increase.

The difference is quite big in every game benchmark on every site. It's a lot more than the modest GFLOPS increase. You'd know that if you'd bothered to check instead of repeating your assumptions. Have a nice life.

I actually did read the entire anandtech article before coming to SPCR today. You just seem to be stuck on assuming that I don't understand the difference between FPS and GFLOPS.

I do understand the difference but I'm also comparing numbers for unreleased cards which forces my hand since I don't have FPS benchmark numbers to use for some of the cards I'd like to discuss.

I'm not sure why you are so upset. I chose to discuss cards that will be released in a month or two instead of just focusing on what got released today. I also chose to look at TDP and heatsink designs as more important factors that performance alone.

I'd feel better about the comparison if Anand had used more cards and more resolutions. Take http://www.techpowerup.com/reviews/Powe ... en/28.html for example. There we have about 20 benchmarks each run at 5 different resolutions each with about 25 different cards. Admittedly the lowest resolution is too low but with that amount of data you can make some serious comparisons. Anand did some of his benches with only 5 cards, those 5 got 3 different resolutions, the other 5 or so cards only got tested at 2 resolutions (total of just over 10 cards) and in all there were about 10 benchmarks. Now if Anand chose the exact 3 resolutions I care about with the exact 5 cards I want to see in the exact 5 benchmarks I care about it'd be OK but as a matter of fact he didn't so I'm still waiting for other reviews to roll in.

The noise tests on Anandtech are useless and their idea of a power test is total system power draw which is usable enough but inconvenient to compare vs other sites or vs manufacturer specs of a single component. http://www.techpowerup.com/reviews/Powe ... en/27.html is more what I'd want to be able to point to for power consumption.

Now techpowerup did review a few 6xxx cards and yes the performance per watt numbers do look better for the 6xxx than 5xxx cards. But that doesn't fix the TDP issue nor will it change the tone of your replies. It'd be more constructive if you would submit something a little longer than a one liner and a link to an article I've already read if you take issue with something I've written.

http://www.techpowerup.com/reviews/HIS/ ... 50/29.html shows the HIS 6850 being about 5% more efficient than the 5750 and 10% more efficient than the 5770 when taken on a wider range of benchmarks and resolutions. It'll take some time to see how all the cards average out as power efficiency and performance will vary as new cards get added to the mix.

Pick and choose your favorite benchmarks, post any you like, just don't dismiss my comparisons without pause just because I use a different benchmark.

[dhanson865]

I should also quote this

As a result our card ran with 1120 shaders, when it should have 960 instead. This would have resulted in an unfair performance increase that the regular customer would not get. Since neither HIS nor AMD could help us get our card fixed we had to take things into our own hands and manually reconfigured the card to run at the proper configuration. If you see HD 6850 reviews that show surprisingly high performance numbers, ask the reviewer to double-check using GPU-Z or fillrate tests.

[micksh]

Just look at idle noise level of Sapphire HD 6850. I think at idle it's the quietest active cooled card that HT4U tested.
As of right now the card is still available at newegg for $180.

http://ht4u.net/reviews/2010/amd_radeon ... ndex17.php

[Cryoburner]

I have to agree that comparing GFLOPS between the HD 5770 and 6850 is rather useless for just about any practical application, due to the other architectural changes not being taken into account. While the GFLOP increase is only around 9%, the actual performance difference in games is upwards of 35%, and in some cases much higher. This is all without drawing significantly more power under load, and with a good stock cooler, making the card relatively quiet out of the box. I'd say the HD 6850 looks great from a performance-oriented quiet pc perspective, and certainly provides better performance per watt than the 5770. At the very least, it's a good alternative to the GTX 460.

[MamiyaOtaru]

I had been interested in a 5850. Now the 6870 is out, and has about the same performance, with the same power draw. Yet in most of the tests I have seen it runs louder and hotter.

So correct me if I am wrong on this: given two cards with the same power draw, the one with the smaller die size will run hotter. Am I wrong in thinking that dissipating the same amount of power from a smaller area won't work as well?

If this is the case (and test results seem to bear it out) a 5850 would be preferable. Same-ish performance and power draw, but runs cooler or quieter. Smaller die in the 6870 is an advantage in terms of cost (cheaper to manufacture and thus to buy) but it will be hotter than the larger die of the 5850.

Anything off in this reasoning?

[Modo]

The new GPU runs hotter because AMD used less efficient heatsinks on the reference cards. The performance per watt is better, but if you want coll & quiet, you need to look for non-reference heatsinks.

[KayDat]

The 6870 sits between the 5870 and 5850, so it actually performs better than the 5850. Power draw is similar to the 5850, and by the way, die shrinks actually result is lower power consumption and therefore less heat, so I'd agree with Modo about the coolers.

[MamiyaOtaru]

The 6870 sits between the 5870 and 5850, so it actually performs better than the 5850.

The edge is there but slight, and according to most reviews goes hand in hand with an increase in power used. They are quite comparable. But how can you possibly think it is relevant to say "die shrinks actually result is lower power consumption and therefore less heat" right after saying " Power draw is similar to the 5850"?? I know that die shrinks generally lower power, but like you just said in the same bloody sentence, the 6870's power draw is similar to the 5850's. in this case it did not result in lower power, so it's irrelevant to the question at hand to say that it usually does. I am asking about a particular, atypical situation where a die shrink does not result in lower power, as with the 6870 vs 5850.

Power draw is basically the same (though most reviews show the 6870 actually draws more). The 6870, which is smaller, nevertheless uses the same amount of power (due to higher frequencies or whatever). Therefore it uses more power per square millimeter. It is using the same amount of power (crammed into a smaller area), and has a smaller area for heat transfer. Is there any reason to think it won't get hotter, given an equivalent cooling solution?

The performance per watt is better

That's not actually the case: see http://www.techpowerup.com/reviews/HIS/ ... 70/30.html

The new GPU runs hotter because AMD used less efficient heatsinks on the reference cards ... if you want coll & quiet, you need to look for non-reference heatsinks.

That really doesn't have anything to do with what I asked. Yes, better cooler will help, but that goes for both the 5850 and 6870. My question has nothing to do with coolers, I am talking about a fundamental property of the chip. In short, from what I can see, the same cooler on both means the 5850 will not run as hot since it is dissipating its heat through a larger surface area. Is there any reason to think this would not be the case?

[Modo]

That really doesn't have anything to do with what I asked. Yes, better cooler will help, but that goes for both the 5850 and 6870.My question has nothing to do with coolers, I am talking about a fundamental property of the chip.

You seem to be confusing the chip with the card as a whole. I already wrote about the performance per watt of the new GPU. The new chip generates less heat than the previous one when providing the same performance. For example, the 6850 is close in power consumption to the 5770, but closer in performance to the 5850. This is a fact, easily checked by taking time to read some reviews.

In short, from what I can see, the same cooler on both means the 5850 will not run as hot since it is dissipating its heat through a larger surface area. Is there any reason to think this would not be the case?

Even if it's the exact same cooler, the new PCB layout (VRMs on the other side of the GPU) means different thermals.

[MamiyaOtaru]

That really doesn't have anything to do with what I asked. Yes, better cooler will help, but that goes for both the 5850 and 6870.My question has nothing to do with coolers, I am talking about a fundamental property of the chip.

I already wrote about the performance per watt of the new GPU. The new chip generates less heat than the previous one when providing the same performance.

No it doesn't. Performance per watt is worse: http://www.techpowerup.com/reviews/HIS/ ... 70/30.html (see 6870 and 5850 in those charts)

For the purposes of my inquiry, I DO NOT CARE about 5770 or 6850, I am talking about the 6870 vs the 5850. Not any other chips. Those two, specifically.

They compare very closely in benchmarks. The 6870 is slightly faster, but uses more power. That is something *you* can verify by checking reviews. (*) Any performance gains it has are matched by an increase in power used. They are closely comparable, but the 6870 runs hotter and louder and I suspect that this is due at least in part to the dimensions of the chip, which you cannot change by putting a new cooler on it. That is to say, sure a new cooler would help its temperatures, but would help the 5850 too. I am looking at this independent of whatever cooler happens to be on it.

You have yet to explain to me why the 6870 should run cooler, since it uses the same amount of power but has less surface area through which to dissipate heat.

In fact, forget model numbers. Tell me why a smaller chip that uses the same power as a larger chip should run cooler, even though it has less area for heat dissipation.

____
ps: Yeah *maybe* the 6870 is mounted on a board that uses a lot more power than the board the 5850 is on and the chip itself uses less power. I find it more likely that is not the case. it is hard to measure the power for the GPU alone so that remains speculative of course, and gives wiggle room. From all appearances though complete 6870 cards use approximately the same amount of power as 5850 cards and odds are the GPUs have a similar power draw. But one is bigger, and can shed heat faster. It's not the 6870.

[Modo]

In fact, forget model numbers. Tell me why a smaller chip that uses the same power as a larger chip should run cooler, even though it has less area for heat dissipation.

It won't. Funny thing is, AnandTech shows the 6870 as having virtually the same power consumption as the 5850 while providing added performance. I don't see how both can be right, so somebody has a bad testing setup or is simply lying. I guess it would be "fun" to compare 20 different reviews to see what's what, but I have better things to do.

[quest_for_silence]

They compare very closely in benchmarks. The 6870 is slightly faster, but uses more power. That is something *you* can verify by checking reviews. (*)

However, just to confirm the performance difference is very often indeed small.

the 6870 runs hotter and louder and I suspect that this is due at least in part to the dimensions of the chip, which you cannot change by putting a new cooler on it.

Well, I guess that louder isn't definitely so: talking about reference design, the 5850 seems to have a small advantage, but not at idle.

But, due to the AMD Radeon heat resilience, I guess it is always possibile a tradeoff which gives more heat for more perfomance, giving the same baseline noise, or even better, as it's maybe already seen with the reference HD6850 and the relevant Sapphire design.

[Tzupy]

Heat removal from the GPU does indeed depend on the chip size, but also chip orientation.
And not at least it depends on the quality of the cooler, with respect to the first step: removing heat from the chip.
A tilted chip, like the one in the 5850 or 6870, will be better at transferring heat to 4 heatpipes, than a 'straight' chip like the 5770.
IMO the 6870 having a smaller area than the 5850 could be a bit more difficult for heat removal, but if you get a good aftermarket cooler, it won't matter much.
What I would like to find out is if there is any aftermarket VRM cooling for the 6870, I think this is the real problem.
Maybe the existing Thermalright solutions for 5850 can be used on the 6870?

[quest_for_silence]

AnandTech shows the 6870 as having virtually the same power consumption as the 5850 while providing added performance. I don't see how both can be right

As AnandTech is (AFAIK) the only reviewer to say so (among the others there currently are: Hardware Canucks, TechReport, TechPowerUp, HT4U, et c.), more probably that not it is wrong. But I don't see any scandal in such a situation.

[MamiyaOtaru]

They compare very closely in benchmarks. The 6870 is slightly faster, but uses more power. That is something *you* can verify by checking reviews. (*)
Perhaps you have to check your post (and rip off the () around the quoted URL, maybe).

However, just to confirm the performance difference is very often indeed small.

the 6870 runs hotter and louder and I suspect that this is due at least in part to the dimensions of the chip, which you cannot change by putting a new cooler on it.

Well, I guess that louder isn't definitely so: talking about reference design, the 5850 seems to have a small advantage, but not at idle.

But, due to the AMD Radeon heat resilience, I guess it is always possibile a tradeoff which gives more heat for more perfomance, giving the same baseline noise, or even better, as it's maybe already seen with the reference HD6850 and the relevant Sapphire design.

It is true, various benchmarks and reviews have varying reports. Anandtech's review shows the 6870 as using the same load power at draw as the 5850. Others show it using more, though almost always less at idle (I have seen some where idle power draw was the same) Performance is generally higher, but it is, as you say, indeed small. That page you linked shows the 6870's performance to be average 4% better than 5850 (Performance-Index – Gesamt), while using 12% more power (Leistungsaufnahme - Last), which fits with what i said about performance per watt. Comparing reference cards of course.

That is a good point about heat resilience. If the 6870 can take it, why worry about it? It is a tradeoff worth making for AMD, as smaller die means it is cheaper to produce. If that means a little more heat (or more accurately more heat in a smaller area leading to higher temperatures) that is fine as long as the chip can take it. I don't mean to say the 6870 is inferior and no one should get it, I am looking at it from the perspective of someone who is irrationally obsessed with temperatures and is not above paying extra for a larger more expensive to produce GPU if it runs a little cooler. I don't expect everyone to do that.

I just think it should not be ignored that the 6870 will probably run hotter than the 5850 due to its smaller size not being matched by a smaller power draw, in addition to its lower performance per watt. Thus, to keep it sufficiently cool it is conceivable that it may take more noise than it would for a 5850, which is relevant to this forum *for someone comparing 5850 and 6870*. As always, check benchmarks and see if that doesn't play out.

It really is too bad that TSMC's fumble meant AMD couldn't fit in a process shrink as well, as that would have negated my entire line of inquiry. Oh well, next time.