Official AMD Radeon 6000 Series Discussion Thread

[Thatguy]

Don't know about the rest, but sorry to point out but the 570 beats the 6870 for performance per dollar XD

Have to hand it to NV this time, 570 is a good card.


but not in the UK, 6970=570 for price so the graph would have them equal here.

hmmm, you sure about that ? what becomes obvious is that at higher gaming resolutions. the 6970 scores better and better vrs the nvidia offerings. I expect AMD to lower card prices soon. I think they might spread them like this

5770 $130-140
6850 $160-180
6870 $190-220
6950 $240-275
6970 $290-325
6990 $385-400

based on their currnet product tiers that would be the sensiable move and it respectively puts each card into a good price/performance catagory.

I expect to see prices go down.

I also reserve judgement on these cards until they get at least one more driver revision. That will tell us if its drivers or the acrhitecture just wasn't as good as everyone hoped.

what I find odd is the report of 1900sp's initially from a few very reliable sources. Makes me wonder if there is another larger chip in the works maybe a 6980 ?

 

[EastCoasthandle]

@Eastcoasthandle, refresh? naww whole new architechture next year, end of q2/start of q3 bulldozer and the 7000 series are due for release.

Nah, it's very possible as it's happened before. Like I said IMO, it depends on TSMC.

 

[Thatguy]

I was thinking the same thing. It might be another x1800 to x1900 or 2900 to 3800. But that might depend on how soon TSMC can get their act together. If TSMC isn't ready until mid to late 2012 as predicted then I believe we might see a gpu refresh next year.

I wonder if TSMC didn't screw the pooch here. If TSMC doesn't get it together we might see a refresh at glofo at 32nm. This of course beg the question will nvidia go to intel fabs since they decided to open them up ?? given the bitch slapping going on between nvidia and intel as of late, I don't know if intel would ever get around to giving them a qoute.

 

[EastCoasthandle]

hmmm, you sure about that ? what becomes obvious is that at higher gaming resolutions. the 6970 scores better and better vrs the nvidia offerings. I expect AMD to lower card prices soon. I think they might spread them like this

5770 $130-140
6850 $160-180
6870 $190-220
6950 $240-275
6970 $290-325
6990 $385-400

based on their currnet product tiers that would be the sensiable move and it respectively puts each card into a good price/performance catagory.

I expect to see prices go down.

I also reserve judgement on these cards until they get at least one more driver revision. That will tell us if its drivers or the acrhitecture just wasn't as good as everyone hoped.

what I find odd is the report of 1900sp's initially from a few very reliable sources. Makes me wonder if there is another larger chip in the works maybe a 6980 ?

What I found interesting is that the difference between the 6970 and the 5870 2gig at highest resolution. It's night and day as far as performance is concerned. So there is more involved then just having 2gigs of memory. Heck even at 925MHz for the 5870 Toxic it's not even close.

So why isn't this conconsistent across the board? That's the question. We know drivers are a factor but their could be other factors at play as well.

 

[Thatguy]

at the moment it would be rampant speculation, I think its a safe assement that the drivers just aren't mature yet. I found that trend interesting myself and I think the optimizations are going to be what really makes this card shine. I don't think I'd expect it to beat the 580 but I expect the gap to close up a bit more.

Its a bit early and the issues with tsmc may have cuased a avalanche of problem that AMD was not prepared for. When you working silicon gets pushed back driver developement takes a hit as well. For all we know the reason for the daul bios might be a very real need. I wonder what bios improvements may bring.

Thats why I siad i would wait for a bit before jugding these cards.

What I found interesting is that the difference between the 6970 and the 5870 2gig at highest resolution. It's night and day as far as performance is concerned. So there is more involved then just having 2gigs of memory. Heck even at 925MHz for the 5870 Toxic it's not even close.
http://i11.photobucket.com/albums/a154/eastcoasthandle/6900 Series/HD6900AVP.jpg

http://i11.photobucket.com/albums/a154/eastcoasthandle/6900 Series/HD5870-TOXICAVP.jpg





http://i11.photobucket.com/albums/a154/eastcoasthandle/6900 Series/HD6900.jpg

http://i11.photobucket.com/albums/a154/eastcoasthandle/6900 Series/HD5870-TOXIC-78.jpg

So why isn't it doing this good at resolutions lower then 2560? That's the question we know drivers are a factor but their could be other factors at play as well.

 

[EastCoasthandle]

at the moment it would be rampant speculation, I think its a safe assement that the drivers just aren't mature yet. I found that trend interesting myself and I think the optimizations are going to be what really makes this card shine. I don't think I'd expect it to beat the 580 but I expect the gap to close up a bit more.

Its a bit early and the issues with tsmc may have cuased a avalanche of problem that AMD was not prepared for. When you working silicon gets pushed back driver developement takes a hit as well. For all we know the reason for the daul bios might be a very real need. I wonder what bios improvements may bring.

Thats why I siad i would wait for a bit before jugding these cards.

No doubt that drivers will improve frame rates overall. But the question will be how long will it take them to do it? Another thing is the overall improvement in CF scaling. That is something, IMO, they seem to have worked on the most. As that is the best aspect for the 6900 series.

 

[crazyeyesreaper]

exactly i dont care that it cant beat a 580 its that in terms of dual cards it compares evenly with the 580s i had dual 5850s best i saw was 40-50% scaling this time the average in the same games should be around 80-85% thats enough to warrant my upgrade as in dual card alone i should see 30% and in terms of performance increase theres already nearly a 40% gain when its 1 card vs 1 card of what i have now so the performance boost for me will be rather massive no matter if its single or dual card depending on the game

 

[Thatguy]

No doubt that drivers will improve frame rates overall. But the question will be how long will it take them to do it? Another thing is the overall improvement in CF. That is something, IMO, they seem to have worked on the most. As that is the most positive attribute for the 6900 series at this time.

Absolutely and if they can really hammer with a single card like the dual gpu 6990 at the right price point, CF is a very workable option as a selling point. I think AMD has some good product and it is interesting to note the differences between the 6800 and 6900 series in terms of architecture. In the back of my mind I wonder if they are working on a bios that unlocks something on the card. I wonder if these cards are meant to be more like engineering work horses for the ongoing 7000 series cards on the horizon ?? I just can't see why else you'd put a dual bios on the card. They must be expecting to do a good bit of updating. Maybe they are refining things as we speak. The dual bios is interesting however.It might be very telling to. Putting a dual bios on a card is easy. Use double the memory size you need and flip the switch to pull a leg on the flash up or down and you can page half a memory very very easily. A trick I have used in automotive applications for a long time.

So it might have been a very late implementation on that switch in the pcb process but a needed one based on the fact that they got short changed on the process and developement time.

 

[EastCoasthandle]

In the last few posts, it was shown that a 6970 beats a 5870 2gig OC'd at 925MHz (Toxic) at higher resolution(s). However, as reviews have shown (albeit with different drivers) that performance improvements are not consistent across the board against the 5870. This maybe the result of drivers. It could be that a new bios is needed or perhaps a respin or something else. Which at this time is our speculation. However, it's clear that when 6970 is paired together they offer far better competition for the 580 sli in price/performance ratio requiring far less power.

So it's clear something is amiss. What it is we really don't know. At this time all we can do is what a month or 2 to see what the drivers do. Since it's a new arch that's a reasonable assumption to make. But one thing is clear, if the cards paired together perform that well it would make sense that a single card should also perform just as well within it's performance bracket.

Time will tell...

 

[crazyeyesreaper]

correct something does seem off as noticed in w1zzards 6950 review the cards had greater then 100% scaling in a few titles

 

[Benetanegia]

do you work for nvidia ? if you don't like the cards, don't buy one. seems pretty simple to me. I didn;t sit here and shit on nvidia for the fermi reactor they launch mid last year. If they don;t have what you want, move on.


If you want to call wizard a lier, feel free. You could call AMD a lier to.but thats not in the white papers I have read.

But even all that bullshit aside.

2 things are obvious.

you a a pronvidia poster which is stupid.

AMD wins the most important benchmark of this generation

http://www.techpowerup.com/reviews/HIS/Radeon_HD_6970/31.html

Thanks for posting that. GTX570 clearly better than HD6970 there, which is what I've been saying like how many times? :shadedshu

On overall perf per dollar the GTX570 is 16% better. Performance is the same, power consumption similar, so which is the better card to buy right now? Who is not being objective and is doing as much as it can to avoid the truth?

And regarding the VLIW4 shaders... can you even read? It's in the slide. AMD's own slide, with AMD's own words. I'll transcribe it to you just in case you can't read a simple slide:

All stream processing units have equal capabilities (no more "T" unit)

Special functions (transcendentals) occupy 3 out of 4 issue slots.

http://www.techpowerup.com/reviews/HIS/Radeon_HD_6970/images/slide3.jpg

My God... Thanks for giving me so much patience.

 

[EastCoasthandle]

correct something does seem off as noticed in w1zzards 6950 review the cards had greater then 100% scaling in a few titles

Yup, CF scaling is much better then before. So they did get that right. The 6900 series isn't bad at all when you factor in the price and feature set. There will be obvious gains once the drivers mature but how soon that will be is something to ponder upon.

 

[purefun65]

My guess is there is some hidden special sauce with these cards.

 

[EastCoasthandle]

Drives are getting updated. But it's not clear if it's the hotfix driver or the regular drivers.
source

I believe the fixed function hardware will take care of itself, but I don't know that Tessmark is setup bound in the first place. That could be tested with an OpenGL vertex or triangle test (there should be one somewhere, I'm sure).

BTW - the driver has been updated and should bring improvements in a few of the DX SDK tests. Download it again and give it a go.

Well improvements have started:

3dmark 11

driver that came with the disc
p4881

this driver
p5034

think thats a fair improvement, wish afterburner supported voltage controll on my card..

 

[Thatguy]

No the 570 is not better then the 6970, certainly not at high resolutions. the 570 and the 6970 trade blows and are similarly priced.

Also on the price perf dollar, AMD can down price these cards heavily.

Also the part you keep refusing to read is that they made 2 of the SP's wider in the 4 groupings giving them each nearly the capability of the older 5th SP.

I have no idea why you have trouble with basic reading comprehension.

But you should really just give me a hug unless you have something constructive to add. I am begining to think you work for nvidia.

Thanks for posting that. GTX570 clearly better than HD6970 there, which is what I've been saying like how many times? :shadedshu

On overall perf per dollar the GTX570 is 16% better. Performance is the same, power consumption similar, so which is the better card to buy right now? Who is not being objective and is doing as much as it can to avoid the truth?

And regarding the VLIW4 shaders... can you even read? It's in the slide. AMD's own slide, with AMD's own words. I'll transcribe it to you just in case you can't read a simple slide:



http://www.techpowerup.com/reviews/HIS/Radeon_HD_6970/images/slide3.jpg

My God... Thanks for giving me so much patience.

 

[CDdude55]

@Thatguy
The 6970 only excels better at a higher res due to having more memory and by higher res, i mean mostly 30'' monitors which uses a very uncommon resolution that only .1% of people game on, actual performance between the two at more mainstream resolutions are similar, but the GTX 570 is in fact cheaper, granted, it's not by much. On newegg you can get a 570 for $350 at the lowest, while the 6970, which is AMD's current highest end single GPU solution, is around $370 at it's lowest and it trades blows with only the cheaper mid-range performance GPU the GTX 570, and doesn't come close in most tests against the 580.

 

[Thatguy]

@Thatguy
The 6970 only excels better at a higher res due to having more memory and by higher res, i mean mostly 30'' monitors which uses a very uncommon resolution that only .1% of people game on, actual performance between the two at more mainstream resolutions are similar, but the GTX 570 is in fact cheaper, granted, it's not by much. On newegg you can get a 570 for $350 at the lowest, while the 6970, which is AMD's current highest end single GPU solution, is around $370 at it's lowest and it trades blows with only the cheaper mid-range performance GPU the GTX 570, and doesn't come close in most tests against the 580.

actually thats not the case when compared to the 2gb 5870 which it handily smashs. also you should reread the charts again. Only at lower resolution does the 570 beat it up. At higher resolution the 6970 and 580 close up alot.

I also expect a price drop from AMD on these cards unless they find a bunch of magic in the drivers and currently they appear to be broken. The performance scalling vrs resolution currently makes no sense. The cards are acting like they are being starved by the compiler at low resolution.

If the cards aren't polling frequently enough off the data buss, it could cuase slow performance and thats a bios issue.

 

[CDdude55]

actually thats not the case when compared to the 2gb 5870 which it handily smashs. also you should reread the charts again. Only at lower resolution does the 570 beat it up. At higher resolution the 6970 and 580 close up alot.

I also expect a price drop from AMD on these cards unless they find a bunch of magic in the drivers and currently they appear to be broken. The performance scalling vrs resolution currently makes no sense. The cards are acting like they are being starved by the compiler at low resolution.

Not true,

At 1920x1200 the 570 beat the 6970 multiple times some with some significant frame differences and some with a very little percentage differences, at higher resolutions the 6970 does come within almost the range of a 580, but never actually beats in almost anything, i suggest you look over the charts again and overall resolution performance. The point is that they failed overall in terms of expected performance. In the future i may in fact grab one, but right now it's a pretty big fail in my mind.

Also the drivers can only improve a cards so much, as someone mentioned before, eventually in the next long months ahead you will see performance increases through updated drivers, like the 580 and 570 will too, but by that time, we will be moving on to the next line of cards and these will be EOL.

 

[erocker]

Keep posts from being personal, post in a respectable and civil manner. I will not say this again.

 

[Benetanegia]

Thatguy you have a serious problem with reading comprehension. Where in AMD's slide can you see that they widened two ALUs? It especifies the absolute opposite and the architecture has been explained firther by many people who are far more intelligent than you and me. Did you even read the RealWorldTech article I linked, even just a bit? I think not.

I'll repeat it one last time for you:

ALL 4 ALUs are IDENTICAL and have some of the abilities the T unit on VLIW5 has. Not all of them though and here's the catch, they cannot do trascendental and some other complex operations. For those 3 out of the 4 ALUs in the VLIW4 array work together, so there's one remaining free. 1 complex + 1 simple.

Hell even Charlie understood this (I hope CD is red enough for you):

VLIW4 doesn't lose any of the flexibility, it simply lets you gang shaders together to look like a single complex unit. For example, transcendentals, a type of complex operation, are sent to 3 or 4 shaders at once, and they act as one big virtual complex shader. If the particular op only uses three shaders, an additional simple op can be scheduled in the fourth slot.

http://www.semiaccurate.com/2010/12/...-architecture/

Tom's Hardware:

So, removing the special function transcendental unit and distributing its functionality across the other four units was actually a good performance per area optimization that promised to keep the GPU running within the observed operating rate. There are situations where performance could take a hit (when the VLIW utilization spikes above four), but AMD says that's unlikely.

Hardware Canucks:

Cayman’s design now centers upon a VLIW4 core architecture whereby the fifth special function ALU has been eliminated and its functions have been spread among the remaining four processors.

While the move to VLIW4 architecture netted an approximate 10% performance increase per square millimeter, it could also have a negative impact upon certain rendering scenarios that require high special function utilization.

David Kanter from RealWorldTech.com

However, the biggest difference is the execution of transcendental instructions, which previously used the T-unit in the VLIW5. The T-unit included a dedicated look up table for computing common transcendental functions used in graphics such as logarithms, square roots, reciprocals and trigonometric functions. The Cayman VLIW4 distributes several smaller lookup tables to each of the XYZW pipelines to replace the T-unit. Transcendental instructions are microcoded across three of the four pipelines, using a 3 term Lagrange polynomial interpolation. Thus Cayman VLIWs have comparable transcendental performance (one instruction per clock) and increase the ALU utilization in order to save area. In theory, this could decrease performance for a workload that packed 1 transcendental and 4 ALU ops into every VLIW bundle, since Cayman can only pack 1 additional ALU op. However, the reality is that most workloads do not come close to fully utilizing every slot in the VLIW bundles, so in some sense this approach is ‘free’.

The erratic performance can very well be due to certain games making use of 1 complex + 4 ALU and the reason I think that is the case is, because in most of the games in which HD6970 "underperforms", are the same ones in which the HD5870 does better against Nvidia cards. That IMO suggests a higher VLIW utilization and if that is the case VLIW4 suffers a lot.

 

[Steevo]

Im more interested in what they are planning with the multi-application access, and folding at home, GPGPU acceleration in media and other content.

Unless they pull a ace out of their sleeve with drivers or manage to make good on the crap they promised on many other series I'm going green on my next card.

 

[Benetanegia]

Im more interested in what they are planning with the multi-application access, and folding at home, GPGPU acceleration in media and other content.

Unless they pull a ace out of their sleeve with drivers or manage to make good on the crap they promised on many other series I'm going green on my next card.

GPGPU related stuff should improve significantly over Cypress.

 

[Steevo]

Good, a liquid cooled 3.7ghz quad core 1.1ghz GPU and tight memory running 1200Mhz ddr still suck at editing a paltry 24MBPS HD video.


I needz more speed, it takes longer to edit and put together movies with one or two simple automated effects than it does to shoot them.

 

[Thatguy]

Not true,

At 1920x1200 the 570 beat the 6970 multiple times some with some significant frame differences and some with a very little percentage differences, at higher resolutions the 6970 does come within almost the range of a 580, but never actually beats in almost anything, i suggest you look over the charts again and overall resolution performance. The point is that they failed overall in terms of expected performance. In the future i may in fact grab one, but right now it's a pretty big fail in my mind.

Also the drivers can only improve a cards so much, as someone mentioned before, eventually in the next long months ahead you will see performance increases through updated drivers, like the 580 and 570 will too, but by that time, we will be moving on to the next line of cards and these will be EOL.

look at 2560 and tell me that same story repeats. Again somehow people are overlooking this, repeatadly.

my thought is if the people in this thread don't like AMD/ATI hardware.

Don't !#$%$#% buy it. I just don't care.

We are hear to discuss the card, not compare them to invidia products. If you wish to do so get out of the AMD/ATI subforum and go start a thread and whine about it there.

 

[Thatguy]

And you keep qouting BAD INFO again.

go pull the white papers

Thatguy you have a serious problem with reading comprehension. Where in AMD's slide can you see that they widened two ALUs? It especifies the absolute opposite and the architecture has been explained firther by many people who are far more intelligent than you and me. Did you even read the RealWorldTech article I linked, even just a bit? I think not.

I'll repeat it one last time for you:

ALL 4 ALUs are IDENTICAL and have some of the abilities the T unit on VLIW5 has. Not all of them though and here's the catch, they cannot do trascendental and some other complex operations. For those 3 out of the 4 ALUs in the VLIW4 array work together, so there's one remaining free. 1 complex + 1 simple.

Hell even Charlie understood this (I hope CD is red enough for you):



http://www.semiaccurate.com/2010/12/...-architecture/

Tom's Hardware:



Hardware Canucks:





David Kanter from RealWorldTech.com



The erratic performance can very well be due to certain games making use of 1 complex + 4 ALU and the reason I think that is the case is, because in most of the games in which HD6970 "underperforms", are the same ones in which the HD5870 does better against Nvidia cards. That IMO suggests a higher VLIW utilization and if that is the case VLIW4 suffers a lot.