Topic: What's the latest on 7xxx series from radeon? - page 2. (Read 10556 times)

how the hell you got that I don't want to bet I don't know. read... for your purpose its on, at stock, no oc as someone else who can't read above commented. what you said stock is, its got to double that for 2 btc, and if.it lands at less than £240 you've got.yourself 4 btc. anyone got a release date?

looks like a nice upgrade based on the unconfirmed leaked info, although pricing will determine if it's a worthy upgrade to current mining cards. here's a summary that's relevant to mining:

7990 - 4000?SP, @ ???mhz,  300w TDP
7970 - 2048 SP, @ 1000mhz, 190w TDP
7950 - 1920 SP, @ 900mhz,  150w TDP
7870 - 1536 SP, @ 950mhz,  120w TDP
7850 - 1408 SP, @ 850mhz,  90w TDP
7670 - 768 SP,  @ 900mhz,  60w TDP

Not bad!!  Any info on pricing and/or release schedule?

Also, for the curious, assuming HD 7990 is a total of 4096 ALUs at 1GHz (two Tahiti XT), this single card will be capable of ~1.1Ghash/s.

Nice find. This shows the 7850 will be identical to the 6950 (VLIW4, same # of ALUs) but clocked at 850 MHz instead of 800MHz, and of course 28nm instead of 40nm.

Estimated HD 7850 hashing performance:
    1408 ALUs * 850 MHz / 3670 instructions per hash = 326 Mhash/s  (you can use this formula to estimate the performance of any VLIW4-based 69xx card)

Performance per Joule:
    326 Mhash/s / 90 W = 3.62 Mhash/Joule

Looks like my prediction will be shown as correct (>3.61 Mhash/Joule).

Yeah, I guess he doesn't want to bet as he basically ignored my entire last post...

http://bbs.expreview.com/thread-46257-1-1.html

Also more info.
http://www.overclock.net/rumors-unconfirmed-articles/1109848-expf-amd-gcn-more-rumours-xdr2.html

HD 7000 Southern Islands 28nm
HD 7990 New Zealand GCN
HD 7970 Tahiti XT GCN 1000MHz 32CUs 2048ALUs 128TMUs 64ROPs 256bit XDR2 8.0Gbps 256GB/s 2GB 190W HP
HD 7950 Tahiti Pro GCN 900MHz 30CUs 1920ALUs 120TMUs 64ROPs 256bit XDR2 7.2Gbps 230GB/s 2GB 150W HP
HD 7870 Thames XT VLIW4 950MHz 24SIMDs 1536ALUs 96TMUs 32ROPs 256bit GDDR5 5.8Gbps 186GB / s 2GB 120W HPL
HD 7850 Thames Pro VLIW4 850MHz 22SIMDs 1408ALUs 88TMUs 32ROPs 256bit GDDR5 5.2Gbps 166GB/s 2GB 90W HPL
HD 7670 Lombok XT VLIW4 900MHz 12SIMDs 768ALUs 48TMUs 16ROPs 128bit GDDR5 5.0Gbps 80GB/s 1GB 60W HPL
HD 7570 Lombok Pro VLIW4 750MHz 12SIMDs 768ALUs 48TMUs 16ROPs 128bit GDDR5 4.0Gbps 64GB/s 1GB 50W HPL

Great. One is talking about stock performance and the other is talking about overclocked performance.   

Haha, Im too hardware enthusiest. If you cant get 400 m/hash out of that card your rubbish. But, for the purposes of the bet, Its on. People here can see. They are your witness. You see, Its win win for me anyways if your right or wrong. You really think it will be less than £240 as well, thats a brave man. Remember, the 5850 is the 6950(7950), not 6850(7850), but you know that with your knowledge Ill be around on here from time to time but i dont believe its a long wait. Im a betting man The bet is 2 BTC, 4 if your right AND its less than £240, not OR

And i do claim that yes, and ill show you...

http://imageshack.us/photo/my-images/197/imag0031cd.jpg

http://imageshack.us/photo/my-images/708/imag0030f.jpg/

1.6 each, signiture should be there or there abouts... Theres another comp with 3 5870's but dont overvolt. They do 410 at 900/300. 5850's are best as 5870 THAT CAN BE OVER VOLTED are RARE! and expensive usually.

You want to bet against someone who has been developing ATI/AMD GPU apps for 3 years, who has been tracking and estimating the perf of their past and future GPUs for as much time, and who wrote the first CAL IL Bitcoin miner for Radeon cards? Sweet

Let's craft the terms of the bet more precisely. Especially let's take overclocking out of the equation as it is very variable from card to card (what if I claim I own a 7xxx card doing 800 Mhash/s? And you accuse me of photoshopping? We want results anybody can replicate.) As of today, the 5850, with a 151 Watt TDP, does 295 Mhash/s with the latest cgminer, at stock GPU clock, mem clock, and voltage. This is 1.95 Mhash/Joule. I predict that a 28nm 7xxx GPU will achieve at least 1.85 times this performance (3.61 Mhash/Joule) at stock GPU clock, mem clock, voltage, and all other settings, as benchmarked with cgminer 1.6.2 with the phatk kernel. If AMD fails to release a GPU matching my description by Sep 4, 2012 (a year from now), you win.

For example, I win if:
- a 100 Watt 7xxx card is released and achieves at least 100*3.61 = 361 Mhash/s
- a 150 Watt 7xxx card is released and achieves at least 150*3.61 = 542 Mhash/s
- a 200 Watt 7xxx card is released and achieves at least 200*3.61 = 722 Mhash/s

Let's bet 4 BTC. Deal? My email address is m.bevand at gmail.com, send me yours.

The top end Mobile part is going to be 80W with performance on par with a hd 6850

Haha, if i see the equivilant of a 5850 able do 800 m/hash i will give you 2 BTC, and you can hold me to that haha. I have loads of 5850's and all do over 400 without breaking a sweat. And im sure they wont be £240 if they do, equivilent to double the price of a 5850(6950) Ill give you 4 BTC if you hit that one as well

Except that for embarrassingly parallel applications, like Bitcoin, we actually saw a doubling of the performance (in fact more because the switch from the 55nm 4000 series to 40nm 5000 series also brought the BIT_ALIGN_INT integer rotate instruction.)

Bitcoin is not like "normal real-world applications"  

I wouldnt quote twice the performance, IT NEVER IS IN THE REAL WORLD and thats where i live, unfortunatly.

This figure is realistic. AMD has done it the past. With high-powered cards. They will do it again with the 28nm 7000 series.

- HD 4870 X2: 1200 billion instruction/sec / 289 Watt = 4.15 billion instruction/Joule
- HD 5970: 2320 billion instruction/sec / 294 Watt = 7.89 billion instruction/Joule -> 90% perf/Watt improvement

When performance/Watt is doubled, across the board, it doesn't mean we will see an end to ~300W cards like the 5970 or 6990. It means the same performance level will be achievable at half the power consumption, 150W, and AMD is going to continue to design cards for the 300W power envelope, except they will pack twice as much performance.

Yes, but a few years ago anything below 20 was said to be impossible, or at least be too expensive, as can be seen with intels ivy bridge processors at 22nm. It looks like there having a few problems, and broadwell when shrank to 14nm will yet lead to more problems due to 'quantum mechanics' non the less. Apparently. And then 10nm Skylake will more and likly lend less power advantages. The problem is they need to use something else than silicon. IBM have grapene chips but there not ready. Intel cant hold onto there advantage in nm scale which as held them up so long (everyone else is 28nm+, they have working 14nm chips) and with arm and that all coming into play by about 2015, intel are looking elsewhere. There lawers mostly But theve also invested ALOT of money into silicon and have to make there money back.... I've heard theres already better materials than silicon out there but are not allowed to flurish with the powers that be ^^

ANyways, 75% seems abit high to me, maybe with the low power parts but not with the high power stuff. The processes used now are ever more harder and intels Tri-gate doesnt look to be as good as they are boasting. Infact may be the problem in ivy bridge. Cudos though for even atemping such silly crazy sizes. Tech been boring a while, Its time for change, Hopfully without losing control.... i.e, apple lockdown

28nm process is supposed to reduce power consumption  75%. That would be nice.

Huh? AMD did precisely that when going from 55nm (HD 4870 = 800 ALUs) to 40nm (HD 5870 = 1600 ALUs).

They will do the same from 40nm to 28nm.

In fact it will be a bit easier this time to double performance because 55**2/40**2 = 1.89 but 40**2/28**2 = 2.04.

It is not. Do not forget that there are two items at the same time:

* More transisteos per surface
* Higher clock speeds possible.

Find that bad argumentation? Look at what happens to the CPU front - same thing.

Even with only 50% more transistors they can more than double performance by ramping up clock speed.