AMD Radeon HD 9970 Specifications Leaked – Twice as fast as GTX 780 (ChipLoco rumor)

[boxleitnerb]

I said "680". The 770 has significantly better memory speeds. Especially at these extremely high overclocks (1.3 GHz for the 780 is extreme), memory bandwidth will definitely bottleneck the faster card. In that review the 780 OC has 6.3 GHz memory, the 770 OC has 8.0 GHz memory. Besides: Does HardOCP monitor GPU clocks throughout their benchmarks? Or do they only look for a couple of seconds in the beginning and later the frequency might drop below 1.3 GHz under heavy and sustained benchmark load? There is simply no guarantee what the clocks actually are for the whole duration. You would have to disable boost completely with a mod bios or with K-boost to be absolutely sure.

Without such massive overclocking, the tendencies are clear, however:
http://www.pcgameshardware.de/Gefor...AMP-im-Test-preiswerter-Titan-Killer-1082941/
About 50% at similar clocks (1070 MHz).

https://www.computerbase.de/artikel/grafikkarten/2013/palit-geforce-gtx-780-super-jetstream-im-test/
53% at similar clocks (1080 MHz).

With the HT4U review I linked I assumed perfect scaling with overclocking, thus keeping the perfect scaling ratio between GK104 and GK110 at the same clocks. This is a reasonable assumption since every GPU scales perfectly with overclocking as long as memory bandwidth is increased accordingly. This is because the frontend is overclocked as well since it runs with core speed.

 

[3DVagabond]

How about a bonaire x2 x2? A super tweaked bonaire chip doubled all around. Put two of these on a PCB with 8GB GDDR5.
That should put a hurtin on any 780 or Titan. Objections?
No way is AMD coming out with a 480mm2 die. I'll :::expletive::: myself if that happens. hehe.

Pics or it didn't happen

 

[raghu78]

I said "680". The 770 has significantly better memory speeds. Especially at these extremely high overclocks (1.3 GHz for the 780 is extreme), memory bandwidth will definitely bottleneck the faster card. In that review the 780 OC has 6.3 GHz memory, the 770 OC has 8.0 GHz memory. Besides: Does HardOCP monitor GPU clocks throughout their benchmarks? Or do they only look for a couple of seconds in the beginning and later the frequency might drop below 1.3 GHz under heavy and sustained benchmark load? There is simply no guarantee what the clocks actually are for the whole duration. You would have to disable boost completely with a mod bios or with K-boost to be absolutely sure.

Without such massive overclocking, the tendencies are clear, however:
http://www.pcgameshardware.de/Gefor...AMP-im-Test-preiswerter-Titan-Killer-1082941/
About 50% at similar clocks (1070 MHz).

https://www.computerbase.de/artikel/grafikkarten/2013/palit-geforce-gtx-780-super-jetstream-im-test/
53% at similar clocks (1080 MHz).

With the HT4U review I linked I assumed perfect scaling with overclocking, thus keeping the perfect scaling ratio between GK104 and GK110 at the same clocks. This is a reasonable assumption since every GPU scales perfectly with overclocking as long as memory bandwidth is increased accordingly. This is because the frontend is overclocked as well since it runs with core speed.

big mistake. I posted HWC and hardocp links of actual game benchmarks showing GTX 780 running 25 - 30% faster than GTX 770 at similar clocks . We are talking here about 30% performance for 50% more shaders.

the chips are different. GK104 is gaming focussed with better gaming perf/sp and GK110 is compute focussed with lower gaming perf/sp. reasons too are very clear. GK104 has an excellent balance of front end resources to cuda cores.

770 - 1536 cc, 8 SMX, 4 GPC, 4 raster engines.
780 - 2304 cc, 12 SMX, 4 GPC, 4 raster engines.

 

[boxleitnerb]

And you still ignore the question of bandwidth and the continuous surveillance of clocks during the benchmarks. And you don't respond to my counterexamples. Do you really think a 780 which is overclocked by 44% (!!!) vs the average boost clock can scale well if you increase memory clock by only 5% at the same time? THAT assumption is a big mistake:

http://www.pcgameshardware.de/Grafi...rce-GTX-780-Taktskalierung-im-Test-1082208/6/

This review uses fixed clocks via NV inspector, no boost. I know the reviewer personally and he is very meticulous with his methodology. 9 games were tested.
Already at 1046 MHz vs 902 MHz the gain from overclocking alone is not the expected 15.5% but only a good 11%. Only when overclocking the memory by 16% (7 GHz) is the clock scaling perfect. You can clearly see that another increase to 1.3 GHz for the core would require an adequate increase in memory bandwidth, too, for perfect scaling. Therefore my assumptions are correct.

Just because GK110 and 104 have the same frontend resources doesn't automatically mean it's not enough for GK110.

 

[raghu78]

And you still ignore the question of bandwidth and the continuous surveillance of clocks during the benchmarks. And you don't respond to my counterexamples. Do you really think a 780 which is overclocked by 44% (!!!) vs the average boost clock can scale well if you increase memory clock by only 5% at the same time? THAT assumption is a big mistake:

http://www.pcgameshardware.de/Grafi...rce-GTX-780-Taktskalierung-im-Test-1082208/6/

This review uses fixed clocks via NV inspector, no boost. I know the reviewer personally and he is very meticulous with his methodology. 9 games were tested.

Already at 1046 MHz vs 902 MHz the gain from overclocking alone is not the expected 15.5% but only a good 11%. Only when overclocking the memory by 16% (7 GHz) is the clock scaling perfect. You can clearly see that another increase to 1.3 GHz for the core would require an adequate increase in memory bandwidth, too, for perfect scaling. Therefore my assumptions are correct.

Just because GK110 and 104 have the same frontend resources doesn't automatically mean it's not enough for GK110.

the HWC links I posted showed a EVGA GTX 780 classified at 1.32 ghz core and 7.4 ghz memory and the galaxy GTX 770 was 1.31 ghz core and 7.4 ghz memory.

http://www.hardwarecanucks.com/foru...s/63051-evga-gtx-780-classified-review-8.html

http://www.hardwarecanucks.com/foru...ews/62594-galaxy-gtx-770-gc-4gb-review-8.html

Crysis 3 2560 x 1600 Very High FXAA

EVGA GTX 780(1.3 Ghz) - 47.64 (25% faster)
Galaxy GTX 770(1.3 Ghz) - 37.99

if you are saying GTX 780 is 50% faster than GTX 770 at same clocks atleast post some actual comparisons. All you are doing is making statements without any data or proof to back it.

 

[flopper]

the HWC links I posted showed a EVGA GTX 780 classified at 1.32 ghz core and 7.4 ghz memory and the galaxy GTX 770 was 1.31 ghz core and 7.4 ghz memory.

http://www.hardwarecanucks.com/foru...s/63051-evga-gtx-780-classified-review-8.html

http://www.hardwarecanucks.com/foru...ews/62594-galaxy-gtx-770-gc-4gb-review-8.html

Crysis 3 2560 x 1600 Very High FXAA

EVGA GTX 780(1.3 Ghz) - 47.64 (25% faster)
Galaxy GTX 770(1.3 Ghz) - 37.99

if you are saying GTX 780 is 50% faster than GTX 770 at same clocks atleast post some actual comparisons. All you are doing is making statements without any data or proof to back it.

so 9970 thread is about assholes debating old nvidia cards?

anyhow,
http://www.techpowerup.com/gpuz/4g95w/

 

[boxleitnerb]

raghu, just leave it be. You clearly don't understand the bandwidth problem and are learning resistant. Your data is flawed in that regard and I've told you why and have given proof that bandwidth is very important especially with high OC. The comparison at the same clocks only makes sense if any other limiting factors can be excluded. The two reviews (PCGH and computerbase) I've linked to corroborate what I'm saying - yet you ignored them and claim there is no data. I will not discuss with you further if you exhibit such a behavior.

 

[raghu78]

raghu, just leave it be. You clearly don't understand the bandwidth problem and are learning resistant. Your data is flawed in that regard and I've told you why and have given proof that bandwidth is very important especially with OC.

you have not provided a shred of evidence to back up the 50% perf increase claim at same clocks between GTX 780 and GTX 770. Also you want to handicap GK104 when Nvidia specifically has released a higher bandwidth GK104 SKU aka GTX 770. anyway if you have any data to share PM me and lets not derail this thread more than we already have.

 

[Keysplayr]

leave it. a few weeks and then all this will be settled. :biggrin:

You act like you know something. But you don't know anything. Why?

 

[Keysplayr]

so 9970 thread is about assholes debating old nvidia cards?

anyhow,
http://www.techpowerup.com/gpuz/4g95w/

What is this supposed to be?

 

[BallaTheFeared]

What is this supposed to be?

Clocks make me think it's a 7970.

 

[Keysplayr]

So you truly believe that AMD can't and/or won't release a single-GPU card faster than a 7970Ghz?

Where did you draw that conclusion from? Unless this is a steering attempt?

 

[raghu78]

You act like you know something. But you don't know anything. Why?

the facts will be out in 40 days. btw I hope you will load pics of what you said you will do if AMD went big die (450+ sq mm). :biggrin:

 

[Keysplayr]

the facts will be out in 40 days. btw I hope you will load pics of what you said you will do if AMD went big die (450+ sq mm). :biggrin:

Seems like trolling, but here is yiur crack at redemption. If you have inside info, say so. If you are just guessing, say so.
Otherwise, stop the troll games.

 

[raghu78]

Seems like trolling, but here is yiur crack at redemption. If you have inside info, say so. If you are just guessing, say so.
Otherwise, stop the troll games.

what an irony. sushiwarrior has inside info and he has clearly hinted that Hawaii is a massive single die GPU and GK110 competitor. obviously he won't reveal his sources and definitely you don't need to believe him.

http://forums.anandtech.com/showpost.php?p=35210286&postcount=47

 

[boxleitnerb]

you have not provided a shred of evidence to back up the 50% perf increase claim at same clocks between GTX 780 and GTX 770. Also you want to handicap GK104 when Nvidia specifically has released a higher bandwidth GK104 SKU aka GTX 770. anyway if you have any data to share PM me and lets not derail this thread more than we already have.

So this is what, nothing? Stop claiming I didn't show anything when it is not true! You're only damaging your reputation by doing that.
http://www.pcgameshardware.de/Gefor...AMP-im-Test-preiswerter-Titan-Killer-1082941/
https://www.computerbase.de/artikel/grafikkarten/2013/palit-geforce-gtx-780-super-jetstream-im-test/

The necessity of sufficient bandwidth I showed is...nothing? So it is you who has not shown anything of relevance on the issue of frontend GK104 vs GK110 and scaling. Your examples show a clear bandwidth limitation that holds the oc'ed 780 back. Thus they are of no value for this question.

 

[raghu78]

So this is what, nothing? Stop claiming I didn't show anything when it is not true! You're only damaging your reputation by doing that.
http://www.pcgameshardware.de/Gefor...AMP-im-Test-preiswerter-Titan-Killer-1082941/
https://www.computerbase.de/artikel/grafikkarten/2013/palit-geforce-gtx-780-super-jetstream-im-test/

The necessity of sufficient bandwidth I showed is...nothing? So it is you who has not shown anything of relevance on the issue of frontend GK104 vs GK110 and scaling. Your examples show a clear bandwidth limitation that holds the oc'ed 780 back. Thus they are of no value for this question.

I repeat you have not shown a single comparison of GTX 780 and GTX 770 at 1.3 Ghz.

anyway here is HWC comparison of GTX 780 and GTX 770 both running at 1.3 Ghz core and 7.4 Ghz memory. so with 50% more bandwidth and 50% more cuda cores the increase is still 25%.

http://www.hardwarecanucks.com/foru...s/63051-evga-gtx-780-classified-review-8.html

http://www.hardwarecanucks.com/foru...ews/62594-galaxy-gtx-770-gc-4gb-review-8.html

Crysis 3 2560 x 1600 Very High FXAA

EVGA GTX 780(1.3 Ghz) - 47.64 (25% faster)
Galaxy GTX 770(1.3 Ghz) - 37.99

I have shown enough benchmarking data from hardocp and HWC to prove that GTX 780 is only 30% faster than GTX 770 at same clocks on average while you have not shown any data to back the 50% assertion.

 

[Keysplayr]

what an irony. sushiwarrior has inside info and he has clearly hinted that Hawaii is a massive single die GPU and GK110 competitor. obviously he won't reveal his sources and definitely you don't need to believe him.

http://forums.anandtech.com/showpost.php?p=35210286&postcount=47

I see no irony.
So you would blindly take sushiwarriors word?
That's up to you. But dont go preaching it to everyone else as if its fact. I don't know who sushiwarrior is.

 

[f1sherman]

I see no irony.
So you would blindly take sushiwarriors word?
That's up to you. But dont go preaching it to everyone else as if its fact. I don't know who sushiwarrior is.

NVIDIA rep?

Hawaii is targeted to beat a hypothetical Titan Ultra already

 

[sushiwarrior]

so 9970 thread is about assholes debating old nvidia cards?

anyhow,
http://www.techpowerup.com/gpuz/4g95w/

Looks like a Bonaire Pro if I had to take a guess, dno why they're on 13.20 driver thread though :sneaky: that's out of date by now.

 

[Erenhardt]

Looks like a Bonaire Pro if I had to take a guess, dno why they're on 13.20 driver thread though :sneaky: that's out of date by now.

You mean downlocked/trimmed down 7790?
7790 is known to be memory bandwidth starved, Bonaire pro may be more balanced sku.
Any changes to existing 7790, or it will be rebranded as is?

 

[Jaydip]

So far it looks like the same hype before 69xx series came out.It was touted as performance champ but got beaten handily by 580.I also love at the notion that somehow this will take the gpu crown and NV will be left high and dry.NV hasn't lost a single gpu crown battle in a long time and I don't see it happening any time soon.

 

[SiliconWars]

I also love at the notion that somehow this will take the gpu crown and NV will be left high and dry.NV hasn't lost a single gpu crown battle in a long time and I don't see it happening any time soon.

It's curious that you remember the 6970 vs 580 yet have already forgotten the 7970 GHz edition vs the 680.

 

[Jaydip]

It's curious that you remember the 6970 vs 580 yet have already forgotten the 7970 GHz edition vs the 680.

No I do, but 680 was and still faster than 7970 it's main competitor.7970 GHz was a new sku which was beaten by 680 occasionally and then completely dethroned by Titan/780.Even 770 is faster than GHz at this moment.

 

[AnandThenMan]

Well if you don't factor in chronology, sure Nvidia never loses the performance crown. Think about your argument for a second....