• 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.

** ALL HAIL THE NEW HEAVEN KING - ATI 7970 3GB!! (TRI-FIRE TESTED)**

I am assumed to be an NVidia fanboy because I am disappointed with 7970 perfromance, even though I own and have owned multiple 48xx, 58xx and 6xx cards? Does that mean that all AMD fanboys are delighted with the performance?

I tend to buy whichever card is top dog, or offers close to top dog performance for reasonable money. I play down the cards which I believe offer poor performance or poor value, whichever colour they be. I have not seen too many reviewers jumping up and down and proclaiming how great these cards are either.
 
Last edited:
SamLB said:
Just found this, sorry if it's already been posted:

http://i1130.photobucket.com/albums/m529/malmental/Untitled-5.png
Click to expand...

As others have said, those figures can't be accurate: A 448-bit memory bus would not operate with 2Gb of RAM (it would need to have 1792Mb, or 3584Mb). Also, the core would scale downwards in proportion to the memory bus (implying 672 cores, rather than 704).

Besides, I would be disappointed if those shader-counts were accurate. Rumour suggests that the first "high-end" Kepler part will be the GK104, with 768 cores on the full part. On 28nm, assuming relatively minor architectural changes, this should be possible using a significantly smaller die than was used for the GTX480 / 580.

555BUK said:
8800/9800 (128 cores / 754M transistors) > GTX280 (240 cores / 1.4bn transistors) > GTX580 (512 cores / 3bn transistors)

1024 cores would continue the trend for GK100, but transistor count would be 5-6bn. NVidia do have history with huge cores on therir top cards, but perhaps they will be conservative this time around and concentrate on performance per watt. NVidia should be able to fit approx 5.5bn 28nm transistors on the same die size as the GTX 580.
Click to expand...

If Nvidia follow the historical trend for transistor density, then they should be looking at fitting just over 6Bn transistors into a GTX580-sized die (increase of +105%). If they improve by the ~75% margin that AMD have managed on 28nm, then you're looking at around 5.3Bn. Either way, a 1024 shader card at around 6Bn transistors should be feasible, but it would be another beast in terms of die size.

If, as expected, we only see the 1024 shader card towards the end of 2012, it will give Nvidia more time to refine the more complex design (and for the TSMC 28nm process to mature). If the 7970 is the highest-spec single-GPU card that AMD are planning to release this year, they certainly don't need to rush to market with the full part.
 
Last edited:
Duff-Man said:
If Nvidia follow the historical trend for transistor density, then they should be looking at fitting just over 6Bn transistors into a GTX580-sized die (increase of +105%). If they improve by the ~75% margin that AMD have managed on 28nm, then you're looking at around 5.3Bn. Either way, a 1024 shader card at around 6Bn transistors should be feasible, but it would be another beast in terms of die size.

If, as expected, we only see the 1024 shader card towards the end of 2012, it will give Nvidia more time to refine the more complex design (and for the TSMC 28nm process to mature). If the 7970 is the highest-spec single-GPU card that AMD are planning to release this year, they certainly don't need to rush to market with the full part.
Click to expand...
Don't forget that transistor count can be "reduced" by not doubling the memory bandwidth. Going from 384bit to 512bit (instead of 768bit) could save a lot of die space. Just doubling the shaders, ROP's etc does not mean overall transistor count will doucble, but I think 5-6bn will be a reasonable estimate. Look at the way that GF104 die was much smaller than GF100, even though the number of cores were pretty close. Much of the saving was made bu using a reduced memory bus, with minimal affect to performance.

GF104 384 cores / 256bit mem / 1.9bn transistors
GF100 512 cores / 384bit mem / 3bn transistors

GF104 had 25% fewer cores but GF100 needed more than 50% more transistors.
 
Last edited:
555BUK said:
Don't forget that transistor count can be "saved" by not doubling the memory bandwidth. Going from 384bit to 512bit (instead of 768bit) could save a lot of die space.
Click to expand...

Yes - good point :)

Considering that, I suppose sub-6Bn transistors is more likely for the 1024-shader part. Of course, all this assumes that Kepler does indeed resemble a scaled-up Fermi with relatively minor architectural changes!
 
Yamahahahahaha said:
Personally, I love the platitudes all the Nvidia fanboys are trotting out.
Click to expand...

This kind of garbage deserves a ban tbh.

Like most on the forum we buy the best bang for the buck and were expecting AMD to do a repeat of the 5870 (50% faster than 4890 for £300-£350), they didn't they got 33% faster than 6970 for ~£450-£500 hence the disappointment. Best card I've owned in recent years was the 5870
 
Hi there

Struggling for time today guys, but here is a couple of 3D Mark benchmarks with Tri-Crossfire in 3D Mark 11 & Vantage:-


3dm11trioc.png


vantagetrioc.png




Any good guys? :D
 
straxusii said:
This kind of garbage deserves a ban tbh.

Like most on the forum we buy the best bang for the buck and were expecting AMD to do a repeat of the 5870 (50% faster than 4890 for £300-£350), they didn't they got 33% faster than 6970 for ~£450-£500 hence the disappointment. Best card I've owned in recent years was the 5870
Click to expand...
Talking nonsense is just as worthless.
UHHeT.gif
http://hardocp.com/article/2011/12/22/amd_radeon_hd_7970_video_card_review/14
The card is at least 30% faster in every situation, for the same power envelope, has much improved GPGPU capabilities and scales better in multi-card configurations. Stop bellyaching.

Yes, it is expensive. Only because there is no competition for it at the moment.
The GTX 580's die is truly enormous (520mm^2) which makes it easy for AMD to milk this situation until Nvidia get their rears in gear.
 
555BUK said:
What does this spreadsheet and Jordan's boobs have in common?

GK100 is widely rumoured to have 1024 cores and 512bit memory.
GK104 is widely rumoured to have 512 to 768 cores and 256 to 384bit memory.

...but we will only know for certain once the reviews hit in 3-6 months time. One thing that is for sure is that NVidia will have not set clock speeds yet, so the above spreadheet is fake.
Click to expand...

Dunno - most likely information I've heard so far is that the design is based on 1024 cores but the high end card (non-ultra) is aimed at having 60% more cores and 20% higher clockspeed than the GTX480 which about matches that spreadsheet - aside from those stats are there for the "690" card.
 
Last edited:
Rroff said:
Dunno - most likely information I've heard so far is that the design is based on 1024 cores but the high end card (non-ultra) is aimed at having 60% more cores and 20% higher clockspeed than the GTX480 which about matches that spreadsheet - aside from those stats are there for the "690" card.
Click to expand...

It still doesn't explain the obvious discrepancies in shader-count, memory bus and memory size though. The chart is fake.

If your numbers are correct, then we can realistically expect a 768-shader part at around 850Mhz. This falls in line nicely with the rumoured specs for the GK104 part. Assuming the architecture behaves similarly to Fermi, we should be looking at something like a 50% performance hike over the GTX580 (~60% increase in shader power, but probably less of an improvement in memory bandwidth).
 
Yamahahahahaha said:
The card is at least 30% faster in every situation, for the same power envelope, has much improved GPGPU capabilities and scales better in multi-card configurations. Stop bellyaching.
Click to expand...

We can all cherry pick reviews, hardocp was the kindest toward the 7970. I put more faith in Anand and TPU which show it around the 30-35% faster than 6970, cut your garbage about fanboys.
 
@ DamoTourettes

Interesting signature.

You may want to trim it a little.

You may use a signature (commonly used in email messages) in your posts. You can specify in your profile a signature to use at the bottom of your posts. When you post, there is an option to use your signature. A signature may contain one or more images but they must fit within a box with maximum dimensions of 400 x 75 pixels (width x height) and with a maximum (combined) file size of 20k. Images containing mostly very large words or animated .gifs will not be permitted. You may also add 1 consecutive lines of normal or smaller text directly above or underneath the image with no spaces inbetween image and text. If no images are used then up to 4 consecutive lines of normal or smaller text is permitted. Do not pad out your text lines with blanks lines. A line of text is defined as having a maximum length of 400 pixels. If you want to turn the signature off in the future, you can come back and edit your post. If you have specified a signature in your profile, the box will be automatically ticked for you.
The forum Dons/Admins reserve the right to judge individual signatures, particularly those deemed to contain material of an adult or offensive nature. If the Dons/Admins think the signature is unsuitable for a family orientated forum, then it will be removed. If the user then replaces it without making changes, they will be suspended.
Click to expand...
 
Duff-Man said:
It still doesn't explain the obvious discrepancies in shader-count, memory bus and memory size though. The chart is fake.

If your numbers are correct, then we can realistically expect a 768-shader part at around 850Mhz. This falls in line nicely with the rumoured specs for the GK104 part. Assuming the architecture behaves similarly to Fermi, we should be looking at something like a 50% performance hike over the GTX580 (~60% increase in shader power, but probably less of an improvement in memory bandwidth).
Click to expand...

I haven't heard of anyone referring to them at the 6xx series either - tho with nVidia these days you never know from one day to the next what they are gonna end up calling them - and definitely not heard anything about a "GTX655" part which AFAIK was a one off for the 4xx series due to yield issues.
 
Looking good but a bit out of my price range, hopefully when released other cards on the market will drop in price a little.
 
You must log in or register to reply here.
Back
Top Bottom