• 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.

AMD's 6870 coming in November - Early silicon doing very well

Clarkson said:
Doubt they will be under £150. I doubt they will be under £200...and thats possibly being conservative. The 6770 looks (from my quick look at that table) a bit of an upgrade from the 5850, which is £200 at best at the mo.
Click to expand...

So they're going to replace the 5770 £90-120 part with a £200 part, yeah?

Do my wages at work get replacements at the same rate when this happens? If not, I doubt AMD are going to go with this if it's not world wide on everything... ¬_¬

And 6870 will be £400, yeah? And what about when NVidia release something faster? £500? These economics are boom graphics!
 
Rroff said:
Don't forget tho the 5850 has extra SPs its not fully utilising them to the best performance - for 9/10 scenarios they haven't made any difference other than cutting out deadwood. As I've said for awhile you might see some big performance increases in some scenarios but overall they will only be a few percent faster. Its been about making the architecture more efficent, more cost effective, etc. the real performance gains will come next round.
Click to expand...

Well, this is performance per transistor efficiency we're talking about here, so if they're going to be increasing the number of shaders and including these enhancements, we can expect a greater performance increase than if they were simply adding more shaders. Bear with me here:

Most vector data types in high level shader languages have 4 elements or less (float4 and vec4 in HLSL and GLSL being the widest of such types respectively). Ignoring driver optimisations that attempt to work around this (by, for example, packing 5 scalar additions into a 5-way vector addition - let's not forget this has its own overhead in terms of CPU time), each 5-way group of shaders can only work on up to 4 pieces of data at any one time largely because of this. That collection of 'fifth' shaders goes to waste most of the time. Say on the 5870, you were to remove each of those 'fifth' shaders, you'd be left with 1280 shaders, right? But most of the time (let's ignore that we've just lost our ability to perform transcendental operations for the sake of simplicity) there shouldn't be a performance hit. If we replace those lost shaders to bring our total back up to 1600 shaders, we now have an equivalent of 25% more shader throughput over the 5870.

Now what about our '6870' with 1920 shaders of the 4-way type? Well as we can see, there are 20% more shaders than in the 5870, and they should in most cases have 25% more throughput. The maths here isn't hard, 1.25*1.2 = 1.5. This should work yet again with the 6770 in regards to the earlier performance claims relating to the 5850. (960/800)= 1.2, so 20% more shaders, multiplied by our efficiency increase, and that again gives us 50%. This ties in with the claims of the 6770 being about the speed of, or slightly faster than the 5850 (a quick glance at techpowerup's 5770 review shows the 5850 to be about 43% faster than the 5770).

I'm unconvinced this update is just about making things cheaper and more cost effective - if they were going for that, the 6770 wouldn't have a costlier 256-bit memory interface. In my opinion what they're trying to do is cement their dominance over Nvidia for the Christmas season.
 
Lightnix said:
I'm unconvinced this update is just about making things cheaper and more cost effective - if they were going for that, the 6770 wouldn't have a costlier 256-bit memory interface. In my opinion what they're trying to do is cement their dominance over Nvidia for the Christmas season.
Click to expand...

Its about both efficency and cost effectiveness not just pure cost effectiveness.
 
Lightnix said:
Well, this is performance per transistor efficiency we're talking about here, so if they're going to be increasing the number of shaders and including these enhancements, we can expect a greater performance increase than if they were simply adding more shaders. Bear with me here:

Most vector data types in high level shader languages have 4 elements or less (float4 and vec4 in HLSL and GLSL being the widest of such types respectively). Ignoring driver optimisations that attempt to work around this (by, for example, packing 5 scalar additions into a 5-way vector addition - let's not forget this has its own overhead in terms of CPU time), each 5-way group of shaders can only work on up to 4 pieces of data at any one time largely because of this. That collection of 'fifth' shaders goes to waste most of the time. Say on the 5870, you were to remove each of those 'fifth' shaders, you'd be left with 1280 shaders, right? But most of the time (let's ignore that we've just lost our ability to perform transcendental operations for the sake of simplicity) there shouldn't be a performance hit. If we replace those lost shaders to bring our total back up to 1600 shaders, we now have an equivalent of 25% more shader throughput over the 5870.

Now what about our '6870' with 1920 shaders of the 4-way type? Well as we can see, there are 20% more shaders than in the 5870, and they should in most cases have 25% more throughput. The maths here isn't hard, 1.25*1.2 = 1.5. This should work yet again with the 6770 in regards to the earlier performance claims relating to the 5850. (960/800)= 1.2, so 20% more shaders, multiplied by our efficiency increase, and that again gives us 50%. This ties in with the claims of the 6770 being about the speed of, or slightly faster than the 5850 (a quick glance at techpowerup's 5770 review shows the 5850 to be about 43% faster than the 5770).

I'm unconvinced this update is just about making things cheaper and more cost effective - if they were going for that, the 6770 wouldn't have a costlier 256-bit memory interface. In my opinion what they're trying to do is cement their dominance over Nvidia for the Christmas season.
Click to expand...

Great post, and good summary. I agree with your final statement as well...
 
so, given what little we know about the 6xxx cards, and my complete lack of insight and tech blurb, is the 6870 looking to be a nice card ?
 
Very nice indeed.

Im hopeing no ones buys them, if no ones buyes them they will have enough 6870 chips to make a 6970 and the only card im looking to buy is the 6970 or 7970 in the future.
 
Phoenixdancer said:
Very nice indeed.

Im hopeing no ones buys them, if no ones buyes them they will have enough 6870 chips to make a 6970 and the only card im looking to buy is the 6970 or 7970 in the future.
Click to expand...

LOL Greed pure Greed! 7970 I'll take 2 of those the name sounds like a plane & hopefully the cooler won't.
 
Posted this on another thread:

Doing a quick research this puts the 6770 at around 470GTX speeds. So i would expect it to come in at £175-£199, No doubt Nvidia will quickly undercut them.

After the 6xxx series comes out i expect the following

460GTX 768Mb = ~£100
460GTX 1Gb = ~£130-£150
470GTX = ~£170 - £200 (depending on where the 6770 comes in)
480GTX = ~ £280 - £300

This is based purely on an educated guess
 
Excellent. I read a while back that cards with non-reference coolers should be out quicker this time (can't remember why), any idea on how soon they could be out?

Also, I have a Vapor-X cooler at the moment and am very happy with it, but I heard that the Powercolour custom cooler was supposed to be even quieter and I see HIS have their own iCooler too, what do people think are the best custom coolers for ATI cards?
 
You might alreaady know about this info, but anyway, some snippets I picked up from vr-zone with more about the flagship 6xxx cards.


"Barts will launch in October and Cayman in November.
It is likely that Cayman is HD 6800 and Antilles is dual-Cayman - HD 6900. "

"The Antilles will be the flagship GPU and is expected to be called the ATI Radeon HD 6970. If we are to believe the rumor, then this will be a dual-GPU Cayman card and will be released in December – just in time for the holiday season then. The list of codenames for future ATI cards is as follows: Cayman, Antilles, Blackcomb, Barts, Whistler, Onega, Turks, Seymour, Caicos and Caspian."

"The flagship will be Antilles, and branded as the ATI Radeon HD 6970. Antilles, as expected, will be a dual-GPU Cayman. While the HD 5970 lowers clock speeds from the HD 5870, HD 6970 is expected to feature the same clock speeds as the HD 6870 - basically a HD 6870 in CF. This will be much like the HD 4870 X2. The Radeon HD 6970 is scheduled for December."

"An Antilles PRO is listed in the codenames - and we could have a Radeon HD 6950 GPU. However, this depends largely on the requirements. A Radeon HD 5950 - Hemlock Pro - was rumoured as well, but was never released due to a lack of competition. The suggestion that the HD 6970 will feature the same clock speeds as the HD 6870 has two implications. First, a larger performance gap between the HD 6870 and HD 6970 for the HD 6950 to fit in. More importantly, the HD 6800 series could have a lower TDP than the HD 5800 series, in which case Efficiency could be one of the key objectives of the HD 6000 family."
 
Well I hope the 6870 is same size as 5870 otherwise I wouldn't be able to fit it in my case :( might be going for 6850 instead.
(Got Antec 300, from what I've seen the 5870 just about fits.)
 
Hopefully the x2 varient has at least 1.5 gigs of USABLE ram for each gpu. be nice to have 2 gig for each gpu but that would cost quite a bit id imagine.
 
You must log in or register to reply here.
Back
Top Bottom