Wednesday, July 3rd 2013
Radeon HD 9000 Series Arrives This October: Report
When AMD re-branded most of its Radeon HD 7000 series SKUs to HD 8000 series, for OEMs, we saw this coming from a parsec away. AMD's next discrete GPU family for the retail channel will be placed along the Radeon HD 9000 series, and it debuts no later than this October, according to a Guru3D report. Interestingly, the report states that the first parts in the family will be based on existing 28 nanometer silicon fab processes, and will be codenamed "Curacao" and "Hainan."
We've had our run-ins with "Curacao," from time to time. It's been rumored to be an upgrade of existing "Tahiti" silicon, with 2,304 stream processors based on Graphics CoreNext 2.0 architecture, 144 TMUs, 48 ROPs, and a 384-bit wide GDDR5 memory interface. The Guru3D report adds to that with the mention of an improved front-end, which adds four asynchronous computing engines (ACEs), and three independent geometry engines.
"Curacao" will make up most of the top-tier, likely Radeon HD 9900 series; while "Hainan" succeeds "Pitcairn," filling up the performance-segment. "Hainan" reportedly features 1,792 GCN 2.0 stream processors, 112 TMUs, 32 ROPs, a 256-bit wide memory interface, and fewer ACEs and geometry engines than "Curacao."
Source:
Guru3D
We've had our run-ins with "Curacao," from time to time. It's been rumored to be an upgrade of existing "Tahiti" silicon, with 2,304 stream processors based on Graphics CoreNext 2.0 architecture, 144 TMUs, 48 ROPs, and a 384-bit wide GDDR5 memory interface. The Guru3D report adds to that with the mention of an improved front-end, which adds four asynchronous computing engines (ACEs), and three independent geometry engines.
"Curacao" will make up most of the top-tier, likely Radeon HD 9900 series; while "Hainan" succeeds "Pitcairn," filling up the performance-segment. "Hainan" reportedly features 1,792 GCN 2.0 stream processors, 112 TMUs, 32 ROPs, a 256-bit wide memory interface, and fewer ACEs and geometry engines than "Curacao."
81 Comments on Radeon HD 9000 Series Arrives This October: Report
So what this news post is basically telling me is that AMD is skipping the 8xxx naming scheme of the desktop parts and going straight to 9xxx? What about the laptop parts?
(why go with 10k, when they can jump directly to 100k? :p)
Also, I believe their flagship dual-gpu should not be 9990 or 9999. It they should go a bit further, cause an overflow and end up with Radeon HD OVER NINE THOUSAND
ie i had a 4870x2 to460-sli to 6970cfx to 670sli to what would have been 8970cfx....
this sucks:cry:
It seems they are hosing with the names again.
its 15% shader increase with no known clock speeds but considering current designs most likely 1100 is the top clock speed to be seen on reference probably more likely to see 1000-1050
so from current looks about 15-20% if that, coupled with the fact there is still no Crossfire fix yet. Lets just say im not impressed, and I am willing to bet if the trail is followed long enough the source who gave this info is full of it.
This means AMD would release a card thats roughly on par with the GTX 780 but far more power hungry and hot, that said a game bundle like before would be nice,
Regardless this news seems a bit suspect. No disrespect you BTA but it just smells a bit to fishy but thats just my opinion.
in fact many shaders on the AMD cards go unused, so while there is a shader increase its wont impact performance that much
Daves 7950s kinda prove that point they clock as good as my 7970 Direct Cu II did and his 7950s regularly out performed my 7970 in 3Dmark Firestrike and other benchmarks and thats single card vs single card in terms of raw frame rate.
so unless AMD has magically completely redesigned the GCN core for a rehash of 28nm I dont see much of a performance gain.
On average at the same clock speeds a 7950 vs a 7970 is about a 3-6% gap
so if we keep clock speeds the same increase shaders by 300 we get nearly the same margin as between the 7950 and the 7970 as there is between this new chip and the 7970
1792
2048
2304
the increased ROP count could improve performance by how much that remains to be seen
but from 1792 to 2048 there is a 3-6% difference average being around 4%
going from 2048 to 2304 would result most likely in a 5-7% difference so ROP count TMU while increased might offer more performance however I doubt it as the 7870 Tahiti XT can if pushed hard enough via clocks compete with its bigger brothers so we know bandwidth isnt an issue on memory at least.
As such unless they made some major revisions or are getting 15% higher clocks than the GHz edition I dont see performance improving that much. Thus it ends up being slower than the GTX 780 models available now. The dark horse again would be the TMUs and ROPs along with the mysitcal clock speeds if clocks are high enough and some optimizations are done on the back end in a meaningful way they might pull off a decent speed bump like the 4870 vs 4890 however I just dont see it happening.
AMD needs to LEAPFROG Nvidia to make itself relevant in the news not match what they have already.
Also 4 from 2ace engines thats 100% up.
Im not saying it will be epic just you never knoooow , my fingers are crossed , my upgrade
Nerve is twitching.
Tahiti – 2048 stream processors, 128 TMUs, 32 ROPs, 384-bit
12.5% increase Sp and TMU’s while a 50% jump in ROP’s
Hainan - 1792 stream processors, 112 TMUs, 32 ROPs, a 256-bit
Pitcairn – 1280 stream processors, 80 TMUs, 32 ROPs, a 256-bit
40% increase Sp and TMU’s while no bump in ROP’s
Just as reference, while "Bonaire" wasn’t actually GCN 2.0 (completely), it probably entailed some of the "learned" improvements.
7790 - 895 stream processors, 56 TMUs, 16 ROPs, 128-bit
7770 - 640 stream processors, 40 TMUs, 16 ROPs, 128-bit
40% increase Sp and TMU’s while no bump in ROP’s. That see's a 20-25% in performance, while a 20% lower Perf/watt from the 7770.
Given that the GTX760 is 20% faster performance, though better Perf/watt by 18% with a 7870 Pitcairn; if what "Hainan" GCN 2.0 offers as enhancement above what "Bonaire" provided with its' 40% bump, a 9870 clocked at 1Ghz and then with Boost should in theory be above the GTX760, while still improved on perf/watt. Then consider "Hainan" would be about 28% smaller die, could AMD offer more than the normal "one-up’s-mans-ship" at $250?
"Curacao" with 12.5%, but that huge jump it ROP’s might be all AMD needs to un-leash the Tahiti from Nvidias’ moniker of "we expected more". 12.5% might not move the TDP up drastically, offering a "Curacao LE" (HD 9950) enough oomph to bounce the GTX770, while then "Curacao XL" gets some strong jabs on the GTX780, with nowhere near the price, while not much more on power than Tahiti.
Then is Volcanic Islands for Q1 2014... erupts on Titan?
The fact is GK110 was tapped out back when GTX 680 launched it may not have been a viable chip to bring to market untill now or NVIDIA could be sand bagging just as AMD probably is due to how long the change from 28nm to 20nm was going to take.
I expect AMD to be late to the part and match the GTX 780 but still be lower than Titan, which results in Nvidia being able to play the fastest single GPU card, AMD can't use crossfire as a PRO: untill they fix it so thats out currently its a wash nothing big till 20nm in 2014.
Normally at this stage AMD would launch a 20nm part as part of the 7000 series to test the manufacturing process but since its not ready thats a no go. Oh well still dont believe this news. I could be wrong would be nice if I was wrong but currently I doubt it.
Granted back with the 7970 launched i called it on TS3 with a few other reviews that 7970 was a misbalanced GPU and didnt make sense
7770 = 640 shaders 128bit bus 16 rops 40 TMUs
7870 = 1280 shaders 256 bit bus 32 rops 80 TMUs
7970 = 1920 shaders 384 bit bus 48 rops 120 TMUs
that would have been the usual way amd steps up and steps down a mid range GPU, where as the 7970 seemed to be a one off to regain the crown instead when it launched. Essentially with AMD having fired key personnel they seemed to have lost that ability to scale up and scale down more effectively and efficiently.
another possible option was 2560 384 bit 48 rops but would have been far to power hungry at the time. This revision may change this but w.e it is seems to be more like GK110 was aka heres the GPU we wanted to release but couldn't.
at about 1800 shaders 32 rops seems to be okay but with the 7970 it can be noticed that while not entirely starved it could indeed benefit.
TSMC are on record as saying volume production of 20nm is due (presumably) late in Q1 2014, which is also the timeline for GDDR6/DDR4. I'd find it hard to believe that AMD (and likely Nvidia) wouldn't like to jump on that particular train at the earliest opportunity.Seeing the way Nvidia model their strategy, its not beyond the realms of possibility that they could release a full 2880 core GK110 (in whatever numbers/price), rename the Titan the GTX 880, and the GTX 780 the GTX 870 and still have AMD's top single GPU SKU paired against Nvidia's third best performance card.TSMC's contract with Apple for A8 production likely fits the bill for initial production for them. Small, relatively simple chips (low transistor count), that yield high quantities from limited wafer production.
Only for $999 and includes 9 games.
That being said I have no doubt the new cards will be competitive.