Radeon 7970 Specifications & Release Date slide leak?

[Skurge]

The photo of the card looks legit, so it could be AMD messing with us again.

 

[bunnyfubbles]

again, 64 ROPs never fit into the rumored 384 bit bus for how we understand past hardware to work, so that old number never made sense, but not only would 32 ROPs be just as incompatible, its no more than what the 6900s have, hell, the 5800s had 32 ROPs. So yeah 32 ROPs makes even less sense than 64. I'm really betting we see at least 48 ROPs

so yeah, I'm very skeptical of these numbers

 

[Mistwalker]

It really makes no sense for AMD to release fake slides though, especially with a bunch of erroneous information (unless they're not confident they have a winning part).

If they ARE confident, why not talk about the real performance--even just at a high level? They're already missing the holidays and could stand to show people something worth waiting for, especially if these parts perform and are priced above the 6900 cards (so they're not risking lost sales at lower price points anyway).

So again, I'm not convinced AMD is behind these leaks. Either way I personally dislike the total lack of good, verifiable information right until the hardware launches. Fear of Nvidia knowing what your offerings can do doesn't seem like a big threat given the hardware is shipping already, they can't change anything, and they're guaranteed at least a few weeks alone at market until Kepler shows up regardless.

 

[Anarchist420]

I don't know why they would've gone from 6 zixels/clock to 4 when nvidia does 8 so I'm skeptical of the number of ROPs as well. Also, I don't see why they'd need that much bandwidth unless they aren't going to have much on die cache. GDDR5 modules capable of 5.5 GHz is going to create a huge ass shortage.

I think the slide is fake.

 

[Idontcare]

It really makes no sense for AMD to release fake slides though, especially with a bunch of erroneous information (unless they're not confident they have a winning part).

If they ARE confident, why not talk about the real performance--even just at a high level? They're already missing the holidays and could stand to show people something worth waiting for, especially if these parts perform and are priced above the 6900 cards (so they're not risking lost sales at lower price points anyway).

So again, I'm not convinced AMD is behind these leaks. Either way I personally dislike the total lack of good, verifiable information right until the hardware launches. Fear of Nvidia knowing what your offerings can do doesn't seem like a big threat given the hardware is shipping already, they can't change anything, and they're guaranteed at least a few weeks alone at market until Kepler shows up regardless.

You say that like you are completely unaware of AMD admitting they did this before. You do remember the intentional misinformation/secrecy campaign waged internally and externally surrounding eyefinity, right?

 

[Idontcare]

I'm not sure what is wrong with the other 7xxx thread.

Could have sworn we had a thread for this :hmm:

I'm going to make a AMD HD7970 officially official AMD endorsed, 100% accurate, the real deal thread. lol

Until the cards come out or we get real information nothing is Offical yet me thinks

For the folks who are wondering about the official moderator position on the creation of multiple threads about "pre-release info/sepculation on upcoming GPU XYZ", I refer you to the following community poll in which the community directed the moderators to implement a policy of allowing for multiple parallel threads co-existing on the same topic at the same time.

If the community wishes to revote on this matter, please post in Moderator Discussions and petition me and the other VC&G mods to create a poll to revisit the topic.

Until then we will continue to uphold the community's wishes as previously expressed and captured in the community poll linked above.

Administrator Idontcare

 

[Grooveriding]

I don't know why they would've gone from 6 zixels/clock to 4 when nvidia does 8 so I'm skeptical of the number of ROPs as well. Also, I don't see why they'd need that much bandwidth unless they aren't going to have much on die cache. GDDR5 modules capable of 5.5 GHz is going to create a huge ass shortage.

I think the slide is fake.

I think the ROP count is bogus, but I have a hard time with thinking the actual picture of the cooling shroud is fake. Haven't seen it elsewhere, unless it is a masterful photoshop doctoring.

Of course obr-hardware is pretty much trash and seems to post whatever fart in a windstorm they catch or let-go themselves.

 

[Mistwalker]

You say that like you are completely unaware of AMD admitting they did this before. You do remember the intentional misinformation/secrecy campaign waged internally and externally surrounding eyefinity, right?

Apparently I missed the extent of it...was it that terrible? I remember them doing some crazy things to keep Eyefinity's unveiling a secret until the last minute (identifying it by difference codenames with the various board partners to identify leaks and such), but intentionally leaking false specifications themselves?

Even then, wasn't that some years ago now? I don't recall AMD doing anything of the sort for the 6000 series (the slides that showed up for that had an Nvidia watermark or some such)...I just don't follow the reasoning IF they have a solid product in the channel already.

 

[bullbearish]

I'm currently running CF 4850s from a build 2 years ago. I want performance. If the refresh offers only Eyefinity as the major upgrade, then I'll be buying 6970s or 580s.

 

[Diceman2037]

theres absolutely no way that this gpu can have 2048 GCN style shaders.

sounds like amd is decieving buyers by multiplying the true shader count with something else connected to the shaders, the last gen only had up to 400 or so physical shaders

 

[badb0y]

theres absolutely no way that this gpu can have 2048 GCN style shaders.

Have you been living under a rock for the past 2 hours?

 

[palladium]

OP: very nice leak.

 

[Diceman2037]

Have you been living under a rock for the past 2 hours?

its clear that you still are.

it does not perform as a card with 2048 shaders, i would guess 600-700 physical shaders at the best.

don't know where amd gets its 2048 shaders from, but that is definitely not a physical count.

and the truth is found,

[FONT=verdana,geneva]Now if we take 4 of these SIMD Units that will be the basis of one Compute Unit (CU). Each SIMD unit is 16 wide, times four per compute unit means that each CU unit has 64 shader processors. The GPU has 32 Compute units meaning 64SIMDs x 32 CUs = 2048 Shader processors (for the R7970).[/FONT]

so without the marketting bullcrap, it has 512 shaders when compared to nvidia's cuda core calculation scheme.

 

[palladium]

its clear that you still are.

it does not perform as a card with 2048 shaders, i would guess 600-700 physical shaders at the best.

don't know where amd gets its 2048 shaders from, but that is definitely not a physical count.

and the truth is found,

so without the marketting bullcrap, it has 512 shaders when compared to nvidia's cuda core calculation scheme.

Each "shader" means an ALU that is capable of processing a shader instruction. Tahiti has 2048 ALUs, that is how AMD gets is 2048 shader (SIMD) count. Fermi (GTX580) has 512 ALUs, hence 512 "shaders" or "cores". The architectural differences are huge, each CUDA core does not directly map to one Tahiti SIMD. There are other factors (granted not as great) that influence performance too e.g. texture performance, ROP, tessellation, memory, drivers..... At the end of the day the "core count" doesn't really matter - the performance of the card as a whole does (case in point: Bulldozer with 8 cores can't compete with a quad core SB).

 

[Diceman2037]

do the math, the transistor count and alleged shader number do not add up.

it took nvidia more then doubling the transistors of fermi to get to 512 over the G200b's 240, and amd leads us to believe that its design is more advanced then nvidia's per shader, which would indicate more transistors per sp AS WELL AS including a deeper memory bus.

 

[Idontcare]

As we have historically done with all pre-release threads and speculation threads that remain active past the review release time, locking this thread now that the conversation has migrated to the official stickied review thread.

Administrator Idontcare