Nothing else matters, performance is all at this end of the market.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
More 6970 slides
- Thread starter straxusii
- Start date
Nothing else matters, performance is all at this end of the market.
Maybe 5970 was a typo and he meant 5870..? I mean, didn't you notice the bit where he said 6970 has 20% more shaders
Soldato
- Joined
- 4 Jan 2004
- Posts
- 20,802
- Location
- ¯\_(ツ)_/¯
Thanks, I've always thought single GPU was better, it will be interesting to see if AMD ever make a 6970x2 but it all depends on how much power it's going to use...
Exactly. Performance is all that matters in every section of the GPU market. It's the price that dictates whether the performance is any good or not.
Caporegime
- Joined
- 18 Oct 2002
- Posts
- 33,188
Oh boo hoo, I said 5970 instead of 5870, impossible to tell what with suggesting 1920 shaders is a 20% increase.
As for the 5970, its 40% ahead of a 580gtx in some games, 50-60% ahead in Call of Juarez, it can be both slower than a 5970 and faster than a 580gtx, either way, where in my post did I mention 580gtx performance, and since when has "it only been about performance at this end of the market", it never has been, and never will be, its about price.
If a 6970 is 10% slower than a 580gtx, and costs 30% less, it wins, if its 10% faster, and costs 10% less, it wins, if its 50% slower, and costs 70% less, it wins, etc, etc, etc.
If the 6970 is 10% slower and gives them pretty much the same profit per card selling at £300, as a 580gtx makes sold at £450, and has more availability, then it will sell 10 times as many cards purely because a heck of a lot more people can afford a £300 card than a £450 card. Thats how life works, the 4870 wasn't very close to a 280gtx in performance, but it was less than half the price, and outsold the 280gtx with absolute ease.
The 5870 isn't faster than the 480gtx, yet it outsold it with absolute ease, value, and price are the key factors, performance only matters if it increases value or is completely unavailable elsewhere, there are lots of combinations that beat a 580gtx, and for a lot cheaper.
What I liked is the tessellation slide 
Not sure if it's because they are tuning the existing tessellation unit (which they have adapted to distance check for applying of varying levels of tessellation), or whether it's actually more powerful. Good to see none the less, as it's one of the things that people are wary of, and faster than I expected, games are using tessellation more heavily (hawx 2, metro 2033 etc). Stil doubt you need a mega tess unit but it certainly won't hurt to have a faster unit in there.
Not sure if it's because they are tuning the existing tessellation unit (which they have adapted to distance check for applying of varying levels of tessellation), or whether it's actually more powerful. Good to see none the less, as it's one of the things that people are wary of, and faster than I expected, games are using tessellation more heavily (hawx 2, metro 2033 etc). Stil doubt you need a mega tess unit but it certainly won't hurt to have a faster unit in there.
What I liked is the tessellation slide
They talk about dual tesselators plus some buffer somewhere so I guess it maybe just 2 of the old one that is on the 5870 plus some caching/buffering tom foolery.
Soldato
- Joined
- 4 Jan 2004
- Posts
- 20,802
- Location
- ¯\_(ツ)_/¯
What's the 6990? Two 6950's on the same PCB?
Yup, did read about a buffer issue (i.e. it was too small or something), with AMD's tessellator. Didn't mean it was bad per se, just that in benchmarks, and anything else that very heavily used tessellation it would be maxed fairly quickly compared to the Nvidia equivalent.
Good to see some slides even if people do just want to see the cards.
Good to see some slides even if people do just want to see the cards.
so AMD go down the power management route as well, it would seem that both companies have decided that this is the best way to proceed, the theoretical power graph is quite a funny one with the unrestricted power being nearly twice as high as the constrained power, of course there is no scale and it does quite clearly say its theoretical 
.
Power is listed as <300w for the 6970 and <225w for the 6950, now with the GTX580 listed at 244w things are certainly looking interesting. now we know that if you disable the GTX580 power management it can use a lot more, ive yet to see any performance tests to show if it actually improves performance by disabling it.
with the 6990 being listed as using 300w things are getting very interesting indeed.
oh yes and seeing as they haven't been released today i guess we can now say they are officially delayed.......
.Power is listed as <300w for the 6970 and <225w for the 6950, now with the GTX580 listed at 244w things are certainly looking interesting. now we know that if you disable the GTX580 power management it can use a lot more, ive yet to see any performance tests to show if it actually improves performance by disabling it.
with the 6990 being listed as using 300w things are getting very interesting indeed.
oh yes and seeing as they haven't been released today i guess we can now say they are officially delayed.......
These slides have October on the bottom while the "older" one has 13th December, so the "older" slide seems to be the newer one.
So in October the sides said TBD on a few points but the current one shows 1920 SP, etc. which makes sense timeline wise.
Nice to have some hard information "from the horses mouth" as it were
There still aren't a whole lot of details, but there are a number of interesting things to take from those slides. First up, the "power containment". We saw a version of this with the GTX580, where furmark runs that draw more than 300W would see the clockspeed reduced. It looks like AMD is implementing a similar version. I think that this is something we will see more of in the future; power restriction is a very real issue with modern GPUs (and will only continue to be as GPU power increases, for reasons of entropy if nothing else!). By hard-capping the maximum power draw you allow yourself to push the GPU closer to its limit in the majority of applications, without having to restrict GPU performance in order to take account of the few scenarios where power draw could increase to unsafe levels (like Furmark or similar applications).
Of course, this does have serious implications for overclocking though... We will have to wait to see exactly how this is implemented, but clearly those who increase the clockspeeds and/or voltages of their cards will hit this power-draw limit more frequently than those who do not. We may start to see more firm limits to how much performance we can gain from overclocking... Still, at least this is being done based on very real, physical constraints (heat/power draw) rather than for marketing reasons (c.f. locking multipliers on CPUs), so we can't really complain. I think both AMD and Nvidia will be continue to implement very similar power-draw limitation mechanisms in future generations, so there is nothing to choose between the two. If anything it will reward improved "performance per Watt" engineering, which AMD have shown themselves to be superior at over the past two generations.
I'm a little surprised at the stated "<300W" TDP of the 6970 - that's a little higher than I was expecting. Of course "<300W" can mean anything, and I suspect that this is just the threshold at which the power management device kicks in. The GTX580 is rated at 244W, but in reality can draw up to 300W before the power restriction device lowers clockspeeds. It was always going to be the case that this new and more scalable architecture comes with a price in terms of significantly increased power draw (since it's on the same 40nm process), but if it really can approach 300W in real-world applications I will be surprised.
edit:
lol wut?
If you're referring to the slide second from the bottom, then it says that the new VLIW4 design ("4D shader") offers 'similar performance' to the old VLIW5 design ("5D shader"), yet uses a 10% smaller area. So, a 20% reduction in shaders with a 10% reduction in area? Clearly there is proportionally more area (and so transistors) given over to core logic than to shaders. This is the precisely the opposite of what you're implying...
There still aren't a whole lot of details, but there are a number of interesting things to take from those slides. First up, the "power containment". We saw a version of this with the GTX580, where furmark runs that draw more than 300W would see the clockspeed reduced. It looks like AMD is implementing a similar version. I think that this is something we will see more of in the future; power restriction is a very real issue with modern GPUs (and will only continue to be as GPU power increases, for reasons of entropy if nothing else!). By hard-capping the maximum power draw you allow yourself to push the GPU closer to its limit in the majority of applications, without having to restrict GPU performance in order to take account of the few scenarios where power draw could increase to unsafe levels (like Furmark or similar applications).
Of course, this does have serious implications for overclocking though... We will have to wait to see exactly how this is implemented, but clearly those who increase the clockspeeds and/or voltages of their cards will hit this power-draw limit more frequently than those who do not. We may start to see more firm limits to how much performance we can gain from overclocking... Still, at least this is being done based on very real, physical constraints (heat/power draw) rather than for marketing reasons (c.f. locking multipliers on CPUs), so we can't really complain. I think both AMD and Nvidia will be continue to implement very similar power-draw limitation mechanisms in future generations, so there is nothing to choose between the two. If anything it will reward improved "performance per Watt" engineering, which AMD have shown themselves to be superior at over the past two generations.
I'm a little surprised at the stated "<300W" TDP of the 6970 - that's a little higher than I was expecting. Of course "<300W" can mean anything, and I suspect that this is just the threshold at which the power management device kicks in. The GTX580 is rated at 244W, but in reality can draw up to 300W before the power restriction device lowers clockspeeds. It was always going to be the case that this new and more scalable architecture comes with a price in terms of significantly increased power draw (since it's on the same 40nm process), but if it really can approach 300W in real-world applications I will be surprised.
edit:
lol wut?
If you're referring to the slide second from the bottom, then it says that the new VLIW4 design ("4D shader") offers 'similar performance' to the old VLIW5 design ("5D shader"), yet uses a 10% smaller area. So, a 20% reduction in shaders with a 10% reduction in area? Clearly there is proportionally more area (and so transistors) given over to core logic than to shaders. This is the precisely the opposite of what you're implying...
Last edited:
Its not conclusive but the 1920 SP slide (the image is lower quality) appears to have smudging around the "edited" numbers, the 0s don't match the scale of the number to the left of them on the fields that would have been edited but on the fields that match the other slide the 0s do match up. Couple of other minor pointers that would seem to indicate editing but nothing conclusive due to the poor image quality in the first place.