Possible AMD counter to Titan...

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
Or New Zealand. :D I think the only consumer 'New Zealand' card we've seen is the one Mark Papermaster showed at AFDS. Those cards may have been snatched up by users of ElcomSoft software. The dual chip cards by AMD's partners may not be New Zealand, but dual 7970's.

ElcomSoft: Nvidia Tesla K20 Beats a Dual GPU Configuration in Password Cracking

If we look at Wireless Security Auditor, ElcomSoft's most popular tool the situation changes slightly, as a single K20 delivers 85,000 passwords per second, compared to the 65,000 on the GTX 690. Then again, Nvidia still lags behind AMD, as the three year old Radeon HD 5970 handles 103,000 passwords per second, and HD 6990 increased that to 129,000. In that aspect, not even the K20 can reach performance achieved by a single consumer AMD card. This is also the reason why a sea of secy agencies went forward and acquired AMD Radeon HD 5990 and HD 6990 cards, instead of going professional with the Tesla and FirePro cards. AMD now has a dual-GPU card for professionals and we were told the results blew the competition out of the water and are one of major reasons why AMD never released the consumer version of the card.
 
Last edited:

RussianSensation

Elite Member
Sep 5, 2003
19,458
765
126
I think that 48Rops + 2560 shaders @ 1000 mhz is enough to match titan, if it is really 50% faster than 680.

Even the specs you listed don't sound probable on 28nm node. HD7970 GE reference review sample card's cooler was pretty loud at just 1050mhz. After-market 7970s are more efficient, but they have superior components and much more expensive heatsinks. The HIS 1180mhz draws 239W for roughly 8% more performance over the 1050mhz and that's with only 2048 SPs and 32 ROPs. Don't forget that with 48 ROPs, you'd end up with 192 TMUs. The die size would need to increase substantially from 365mm2 and with it power consumption.

My fear is that if HD8000 series is delayed to Q4 2013/Q1 2014, it might mean 20nm Maxwell/Volcanic Islands could be pushed back. I was hoping they'd launch 1H of 2014.

AMD-Radeon-January-2013-Slide-2.jpg


At the same time until I see next generation PC games, the only thing that keeps upgrades exciting for me is bitcoin mining. Unless AMD releases a 3000 shader card, the bump in bitcoin mining from a 2560SP part will be just 25%, which is nothing special.
 

parvadomus

Senior member
Dec 11, 2012
685
14
81
Even the specs you listed don't sound probable on 28nm node. HD7970 GE reference review sample card's cooler was pretty loud at just 1050mhz. After-market 7970s are more efficient, but they have superior components and much more expensive heatsinks. The HIS 1180mhz draws 239W for roughly 8% more performance over the 1050mhz and that's with only 2048 SPs and 32 ROPs. Don't forget that with 48 ROPs, you'd end up with 192 TMUs. The die size would need to increase substantially from 365mm2 and with it power consumption.

My fear is that if HD8000 series is delayed to Q4 2013/Q1 2014, it might mean 20nm Maxwell/Volcanic Islands could be pushed back. I was hoping they'd launch 1H of 2014.

AMD-Radeon-January-2013-Slide-2.jpg


At the same time until I see next generation PC games, the only thing that keeps upgrades exciting for me is bitcoin mining. Unless AMD releases a 3000 shader card, the bump in bitcoin mining from a 2560SP part will be just 25%, which is nothing special.

That is true, however I can see a 15% speed increase by just adding 16Rops. There is something strange in 79xx series, where a single memory channel is not tied to an X quantity of rops. 32/6 (6 64-bit channels) = 5.333. There must be some type of crossbar between rops and memory channel to match them in tahiti, non existent on 78xx and 77xx series.
That is free performance for some die space, add 8% more for going from 925 to 1Ghz, that`s 23%, then add 25% more shaders and what do you get? I think that should be at least 40% faster than 7970 stock, and maybe at same power envelope of 7970ghz (this has too much stock vcore over 7970, that wont be needed for the hypothetical model without turbo and not surpassing 1 ghz).
This is not taking into account any architecture improvements for HD8000s.
 

BoFox

Senior member
May 10, 2008
689
0
0
True.. Dave Baumann (AMD Product Manager) in fact said:

"with a 8 RB to 12 Channel mapping there has to be a crossbar somewhere - whether it is from the engine to the RB's or the RB's to the memory channels. You also have to consider how the engine upstream is working and how you tie in RB redundancy as well.

Note, with Tahiti there is not a full RB to memory channel Crossbar - each RB can access just 3 memory channels in order to keep the crossbar complexibility and size down, but still providing some flexibility."

I think AMD can easily do it.. perhaps it's a strategy game move to try to make Nvidia think AMD is not yet "counter-attacking" (so that Nvidia just prices their products ridiculously high like they did with GTX 280) then launch a surprise move that causes NV to issue $100 cash rebates all over again?