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

[blackened23]

Yawn. Now we compare stock vs. overclocked models? I would expect nothing less.

Comparing reference to aftermarket overclocked........That sounds like...

Moving the goal posts eh? Yes, yes it does. Anyway, the R9X will have factory OC'ed models fairly quickly as well, i'd assume. But not at launch. This is aside from the fact that in general, AMD reference PCB cards are generally not voltage locked and have 6 power phases with a dual BIOS. I'm assuming the reference will OC just fine as well and won't deal with the TDP / temperature limits of GPU Boost 2.0 (not that I think GPU Boost is a bad thing).

 

[raghu78]

Lol. Is this serious or what. I'm sure the R9X will get factory OC'ed models in time, comparing reference to aftermarket is just assinine. No credible reviewer would compare anything except reference to reference, and in two months or so the factory OC'ed R9 cards will be released.

the fair comparison would be stock with stock and max OC vs max OC with voltage control. obviously we need to be comparing two reasonably well binned chips which are hitting 1200+ mhz with voltage control. comparing a dud of one card with a golden sample of another is not a relevant comparison. Normally AMD's reference designs are well built for voltage overclocking . so lets see how websites handle these comparisons.

 

[blackened23]

the fair comparison would be stock with stock and max OC vs max OC with voltage control. obviously we need to be comparing two reasonably well binned chips which are hitting 1200+ mhz with voltage control. comparing a dud of one card with a golden sample of another is not a relevant comparison. Normally AMD's reference designs are well built for voltage overclocking . so lets see how websites handle these comparisons.

The bottom line is that we'll be looking at the data of reviewers and they will only be comparing reference to reference. The point stands that from a review perspective, throwing in factory OC'ed models is an asinine practice. I don't think any review sites have done that in recent memory - AT, TH, HardOCP, among others always use reference as the basis. Factory OC'ed models just skew the baseline comparison, and also ignores the fact that the R9 cards will get factory OCed models in a couple of months as well (i'd assume).

Now examining the OC potential of a card is fine. But as a baseline? You compare reference to reference as a fair review test. That's just all there is to it, and this is actually what all review sites will be doing.

 

[sontin]

Is this serious or what. I'm sure the R9X will get factory OC'ed models in time, comparing reference to aftermarket is just assinine. No credible reviewer would compare anything except reference to reference, and in two months or so the factory OC'ed R9 cards will be released.

There are no custom designs for Titan. In your world we need to ignore all the GTX780 custom designs in a comparision between the GTX780 and Titan because it would be unfair?

The market is dictating what you should compare. AMD has every right to come out with custom designs. If they are not doing it - it's their problem.

 

[wand3r3r]

Yeah I think the following would be very thorough.

1. stock vs. stock (90% the most important for most people)

Then in a separate review. Clearly separate and labeled as such.

2. average oc vs. average oc (of course it's hard to get the exact average, but at least ballpark)
3. top oc vs top oc (for those that win that silicon lottery)
4. modded bioses separate because you likely lose the warranty in NV's case.

The only people that want stock vs. oc are ... well you get the picture.

 

[sontin]

The only people who dont want that their product is reviewed under market conditions are people who working for certain companies.

Custom designs are reality. You dont need to like it but you should at least accept it.

And btw: Using your logic reviewer should not compare the GTX770 to other cards because there is no real reference design...

 

[wand3r3r]

There was nothing wrong with Titan. No competition for nearly 8 months.

And custom cards are around 20% faster than the reference GTX780:
http://hexus.net/tech/reviews/graphics/60269-evga-geforce-gtx-780-classified/?page=6



Why should i compare the reference modell when i can get for the same price 20% more performance? You should hope that AMD allows their partners to start with custom designs because if they have only a reference design it's their problem and not mine.

There are no custom designs for Titan. In your world we need to ignore all the GTX780 custom designs in a comparision between the GTX780 and Titan because it would be unfair?

The market is dictating what you should compare. AMD has every right to come out with custom designs. If they are not doing it - it's their problem.

Let me get this straight. Post 1, defending the titan due to no competition the $1k is "ok" in your opinion as there is no competition. ()

Then comes the alleged R9 290 with likely more performance than even the titan.

Wait, did the bar change and now we compare stock R9 290 to overclocked 780's, which were irrelevant a minute ago in your defense of the titan? D:

Oh the logic there. :wub:

Sorry bud, you can't have it both ways. :whiste:

 

[3DVagabond]

I'm sure if it was the other way around, on release they compared custom O/C'd 290X's to reference 780's and when people were running around saying the 290X was 30%-40% faster than the 780, you'd think that was fine?

 

[blackened23]

The only people who dont want that their product is reviewed under market conditions are people who working for certain companies.

Custom designs are reality. You dont need to like it but you should at least accept it.

And btw: Using your logic reviewer should not compare the GTX770 to other cards because there is no real reference design...

What's ironic is that reviews of the Gtx 770 launch were reference designs at launch. Look at the TPU, Anandtech, and TH launch 770 reviews. Reference 770. This is what reviewers will do. Whether you like it or not, the R9X reviews will compare reference to reference. R9X will have aftermarket cards in time, i'm sure, but generally that doesn't happen at launch. Comparing stock to stock is the basis of review integrity, apparently you can't see that stock clocks to overclocked clocks is not an apples to apples comparison.

Keep moving those goalposts, though, bro. You're sitting here trying to tell us that comparing stock clocks to overclocked is a fair comparison. If you say so. While i'm sure AMD could send their golden sample R9X overclocked by 20% (with performance to match), they will send reference cards in order to be compared to stock nvidia reference cards. There is no altered reality where overclocked clocks vs stock clocks is a fair basis of a review - to do so would be undermining journalistic integrity and would create an unfair comparison.

 

[Skurge]

What's ironic is that reviews of the Gtx 770 launch were reference designs at launch. Look at the TPU, Anandtech, and TH launch 770 reviews. Reference 770. This is what reviewers will do. Whether you like it or not, the R9X reviews will compare reference to reference. R9X will have aftermarket cards in time, i'm sure, but generally that doesn't happen at launch. Comparing stock to stock is the basis of review integrity, apparently you can't see that stock clocks to overclocked clocks is not an apples to apples comparison.

Keep moving those goalposts, though, bro. You're sitting here trying to tell us that comparing stock clocks to overclocked is a fair comparison. If you say so. While i'm sure AMD could send their golden sample R9X overclocked by 20% (with performance to match), they will send reference cards in order to be compared to stock nvidia reference cards. There is no altered reality where overclocked clocks vs stock clocks is a fair basis of a review - to do so would be undermining journalistic integrity and would create an unfair comparison.

Save your time. He was telling RS that aftermarket 7970s that ran cooler, used less power and were quieter shouldn't be used in a comparison cause it should be reference cards that should be looked at. Yet, now he wants to compare custom cards.

 

[Fastx]

FWIW 9-22-13

Hawaii GPU Specifications AMD Hawaii GPU

It is more or less confirmed that Hawaii will have 2816 cores. The exact configuration is:

44 GCN Compute Units
2816 Stream Processors
176 Texture Mapping Units
44 Raster Operating Units
So as you can see there are less ROPs than we thought. Sources are also saying that the GPU is around 420 mm2. The leaked review you’ve seen yesterday shows the performance with non-reference clocks. It’s worth noting that R9 290X will have a Turbo Mode. That said we believe 1020 MHz (1GHz) clock was obtained with this mode enabled. The real clock is lower (around 900 MHz). Without Turbo R9 290X will be slower than TITAN.

One other thing. The fastest Radeon R9 290X might cost $650 USD, that’s the latest information on the price.

http://videocardz.com/45837/amd-hawaii-gpu-2816-stream-processors

 

[Jaydip]

I think the competition is going to be intense in the coming months.

1.AMD will bundle 290 with BF4
2.NV Bundles 780 with either AC4 or Watchdogs or COD Ghost.

I think NV is also working with "Evil within(? not sure)" developers and Rockstar.If GTA5 comes to pc before the holidays and is bundled with NV cards it will be tough time for AMD.So I believe in the next three months bundles rather than absolute performance are going to decide many gpu purchases.So BF4 alone won't be enough for AMD they need some other titles too.

 

[OILFIELDTRASH]

He makes a good point though. As a consumer that's only concerned with fps for the money and having a quiet cooler you should definitely consider aftermarket vs stock cards when they are at the same price point.
Of course max performance when both are Overclocked should also be considered.
To me whichever company offers the best value for the money that's what I'm going with. Lately that has been amd so I've went with them.

 

[BoFox]

So I grabbed this image from csbin's thread and used it to approximate the die area again since this image was of better quality:

To make sure the image isn't a composite, I checked the size of the RAM chips in both cards and they more-or-less matched up.

To the best of my judgement, I measured out the die area of Tahiti and compared it to the 290X die and noticed that the width of Tahiti's die was slightly smaller than the 290X's die. I figured this was within the margin of error.

I looked at two cases:
A) The 290X die is the same width as Tahiti's (in RED) => 290X = ~454 mm^2
B) The 290X die is slightly wider than Tahiti's (in CYAN) => 290X = ~476 mm^2

I was unable to produce a case where the 290X die was smaller than ~450 mm^2, but again, this could just be me.

Usually, the officially released die size is usually a bit off from the physical silicon measurement by up to 5% or even more.

Remember how much "off" the GTX 460 official size was? The same goes for Tahiti's official size - there's dispute regarding it.

Anyway, it's probably the usable silicon inside that is up to 5% smaller than the physically measurable silicon that companies sometime like to declare.

 

[el etro]

One thing: If Hawaii needs 2816/2880Sps clocked at 1020Mhz to beat Titan then it was not a good deal. Hope it was 2560Sps clocked at 1Ghz to do this(Architecture efficiency and overclocking potential capabilities matter more than ever this time).

The real 290X enemy is not the terribly overpriced Titan but upcoming Maxwell GTX 880. Will not worth if 290X defeat Titan today(as we seeing it gonna do) but lose to 880 3-4 months after.

 

[wand3r3r]

I think the competition is going to be intense in the coming months.

1.AMD will bundle 290 with BF4
2.NV Bundles 780 with either AC4 or Watchdogs or COD Ghost.

I think NV is also working with "Evil within(? not sure)" developers and Rockstar.If GTA5 comes to pc before the holidays and is bundled with NV cards it will be tough time for AMD.So I believe in the next three months bundles rather than absolute performance are going to decide many gpu purchases.So BF4 alone won't be enough for AMD they need some other titles too.

The alleged price is at least equal or lower.
The alleged performance appears higher.
The game will be one of the top of this year.
If that's not enough I'm not sure what is and people are ignorant.

 

[Skurge]

One thing: If Hawaii needs 2816/2880Sps clocked at 1020Mhz to beat Titan then it was not a good deal. Hope it was 2560Sps clocked at 1Ghz to do this(Architecture efficiency and overclocking potential capabilities matter more than ever this time).

The real 290X enemy is not the terribly overpriced Titan but upcoming Maxwell GTX 880. Will not worth if 290X defeat Titan today(as we seeing it gonna do) but lose to 880 3-4 months after.

Not a good deal, except it uses less power and is about 23% smaller and faster (allegedly).

 

[Jaydip]

The alleged price is at least equal or lower.
The alleged performance appears higher.
The game will be one of the top of this year.
If that's not enough I'm not sure what is and people are ignorant.

1.NV can easily bring down 780 price
2.Yes but if most of the upcoming games perform better on 780 it is a moot point.
3.People don't bother how faster your card is at Metro 2033, AVP as almost no one plays them.They are good as benchmarks but as graphics card recommendations not so much.When people will buy BF4 they will not bother about BF3 benches.

 

[sushiwarrior]

FWIW 9-22-13



http://videocardz.com/45837/amd-hawaii-gpu-2816-stream-processors

Should be called uber mode not turbo mode. That was the intended name at least lol.

AMD will never send cherry picked or non-reference cards. Cards sent out for reviewing are only screened to make sure they aren't a lemon, they are never cherry picked. They just make sure no "bum" cards are sent out. No misleading tactics.

 

[el etro]

BF4 benches will probably be based on BF3 performance... AMD designed Hawaii to beat Kelper in Frosbite titles...

 

[Grooveriding]

So is AMD about to drop a dump on nvidia's gk110 here ? Best part of those board shots of Hawaii is the voltage controller on the PCB. Full control of voltage will make the card a beast if it is already faster at stock than Titan.

 

[RussianSensation]

One thing: If Hawaii needs 2816/2880Sps clocked at 1020Mhz to beat Titan then it was not a good deal. Hope it was 2560Sps clocked at 1Ghz to do this(Architecture efficiency and overclocking potential capabilities matter more than ever this time).

Hold on so now performance per shader matters the most? What difference does it make? Should we have been comparing Fermi vs. Kepler on a per shader basis too? Nope. The two key measurements of stock vs. stock and oc vs. oc is what matters. Specs on paper between NV and AMD alone don't tell us the whole story. We don't directly compare GTX680's 192GB/sec memory bandwidth to HD7970GE's 288GB/sec. Real world performance is what matters. As an example, Tahiti XT has 50% more pixel fill-rate performance vs. HD6970 despite both having 32 ROPs.

If you want to measure efficiency, look at performance/watt.

The real 290X enemy is not the terribly overpriced Titan but upcoming Maxwell GTX 880. Will not worth if 290X defeat Titan today(as we seeing it gonna do) but lose to 880 3-4 months after.

Source?

 

[raghu78]

Hold on so now performance per shader matters the most? What difference does it make? Should we have been comparing Fermi vs. Kepler on a per shader basis too? Nope. The two key measurements of stock vs. stock and oc vs. oc is what matters. Specs on paper between NV and AMD alone don't tell us the whole story. We don't directly compare GTX680's 192GB/sec memory bandwidth to HD7970GE's 288GB/sec. Real world performance is what matters. As an example, Tahiti XT has 50% more pixel fill-rate performance vs. HD6970 despite both having 32 ROPs.

If you want to measure efficiency, look at performance/watt.

Source?

I agree. for perf stock vs stock and max OC vs max OC. perf/clock is a reasonable comparison if the chips are clocking similar ranges for max OC. for efficiency its perf/watt at stock.

 

[AnandThenMan]

Who gives a flying **** what the hardware configuration is, as long as the perf/watt is there. A GPU can have one giant shader for all I care.

 

[ozzy702]

One thing: If Hawaii needs 2816/2880Sps clocked at 1020Mhz to beat Titan then it was not a good deal. Hope it was 2560Sps clocked at 1Ghz to do this(Architecture efficiency and overclocking potential capabilities matter more than ever this time).

The real 290X enemy is not the terribly overpriced Titan but upcoming Maxwell GTX 880. Will not worth if 290X defeat Titan today(as we seeing it gonna do) but lose to 880 3-4 months after.

20nm for gpu isn't close to production. No way wee see Maxwell in 3-4 months from 290x release, more like 6-8 months if we're lucky.

The other thing to consider is yields on 20nm will be low for quite some time so a large 20nm chip to replace Titan is less likely than a smaller gk104 replacement.