GT300 new leaks?

[Talonman]

1. GDDR5 memory

2.use 512-bit memory bus

3. To provide 280 GB / sec bandwidth of about (higher than RV870)

4. GM is now calculated to be 2.2 times higher than the GT200

5.1700MHz frequency shader

Bold guess 6.die size

Regards

The GTX 280 has a 141.7 BG/s memory bandwidth, and the GTX 285 has 159.0 GB/s bandwidth before OC'ing the cards memory...

280 GB / sec bandwidth on the 300 would be a decent jump...

1700MHz frequency shader's would be nice, especially with the large number of shaders that might end up being on the chip.

There are few guys who can run 1700 Shaders on a 285.

http://www.evga.com/forums/tm.asp?m=...65533;�

If 1700 was the 300's stock clock setting.... Ya have to wonder what the 300 might OC to.

I think all the latest Nvidia cards respond nice to a little OC.

[trinibwoy]

Funny how speculation in Chinese forums becomes news after a little google translation magic. At this point I don't think even Nvidia knows what GT300 is!

[Talonman]

True...

Pure guessing...

[zanzabar]

The GTX 280 has a 141.7 BG/s memory bandwidth, and the GTX 285 has 159.0 GB/s bandwidth before OC'ing the cards memory...

280 GB / sec bandwidth on the 300 would be a decent jump...

1700MHz frequency shader's would be nice, especially with the large number of shaders that might end up being on the chip.

There are few guys who can run 1700 Shaders on a 285.

http://www.evga.com/forums/tm.asp?m=...65533;�

If 1700 was the 300's stock clock setting.... Ya have to wonder that the 300 might OC to.

I think all the latest Nvidia cards respond nice to a little OC.

a basic 512bit gddr5 would do 280GB/s

[570091D]

Funny how speculation in Chinese forums becomes news after a little google translation magic. At this point I don't think even Nvidia knows what GT300 is!

oh i'm sure things are finalized, they're probably working through the last revisions before ramping production. obviously ati has already begun production on evergreen....

[Pontos]

Funny how speculation in Chinese forums becomes news after a little google translation magic. At this point I don't think even Nvidia knows what GT300 is!

Well, most hardware is made there, almost every early sample and retail product leak comes from someone in there, and also many things get released officially in their retail channel earlier than anywhere else.

So when you get info from a Chinese source it's more likely to be true than when you get it from the inq :P

[Talonman]

I still think there might be a few guys buying into the entire GPU revolution thing, like me. It actually might sway a few customers Nvidia's way?

More might be looking at CUDA, Video Transcoding, Ray-tracing, and dare I say PhysX, than some might think?

Yes, ATI and Nvidia, and Intel will be into OpenCL, and the app game will change then. But I bet getting OpenCL up near CUDA level, will take longer than we want. Heck, I want it to happen fast too.

Another thing that I have to say I think about, is when people say Nvidia's experience with CUDA, will for sure help them with OpenCL. I do understand that each companies implementation of OpenCL, could be a make or break deal. When an OpenCL benchmark comes out, and it only would run on 2 of the 3 platforms, that would be huge in my mind.

I simply believe that Nvidia has a leg up here, and by their actions, they are telling us that they are all about making the GPU more important in the performance equation.

I hope they succeed.

[Oberon]

2.2x figure would prolly be the theoretical throughput which certainly differs from the practical gain, maybe 70~90% even.

I'm not going to put too much stock in that value, then, personally. RV770 has higher peak throughput than GT200 when considering MADD instructions, but that doesn't translate into better performance in most cases.

[Stukov]

2.2x figure would prolly be the theoretical throughput which certainly differs from the practical gain, maybe 70~90% even.

Anyway any1 else besides me have a strong feeling of history repeating like GT200 vs RV770, except I think nvidia will have slightly faster cards than last time.

September, HD 5870 for $299 (and HD 5850 for $199?) is launched that provides 30~50% gain over HD 4890 and will be the fastest single gpu card, may lead to slight price adjustments of other cards for especially nvidia but maybe also ati. Relatively energy efficient cards with great performance/power/cost factor.

Early/mid December, NVIDIA launches the new GT300 single gpu monster with huge power consumption but will have like 70~90% gain over GTX285 but at a price of $599~650. But very few cards will be in stock but hey Santa needs to get a few cards for xmas so the kids won't cry. Real stockage will grow in January. At same time ATI releases HD 5870 X2 to give the new GT300 single gpu card a run for it's money @ $499. This will eventually lead to some price reduction for nvidia and eventually release the dual gpu solution to more specificly point out the so called "performance crown" as they used to talk about a few years ago (yay). ATI then releases quite shortly after (maybe half year after GT300 launch or so) a refresh series with higher clocks with a bit higher performance and some tweaks due to the prices of HD58xx series have become so low at this point and to remain more competitive again (although HD 58xx series have sold well anyway) as ATI knows this is the pricerange were most money is spent.

Just for the luls, have to remember this statement if it actually would turn true. :p

Why do I think a similiar scenario will happen? Because both parts seems to be happy like this when they don't fight each other, they stick to different segments so both parts will get sales, playing safe. Of course this doesn't mean it would be less competition, it would look silly if ATI suddenly would match NVIDIA in performance at a totally different price segments (NVIDIA has beefier cards meant to be sold at a higher price cuz of the higher cost). It's a lot smaller chance of complete disaster this way than directly competing against each other in the entire price segment.

Sorry, but the X2 version of whatever is going to either launch same day, or few months later. It will most likely already be out before Nvidia has GT300 out.

Also, AMD has been really tight lipped about the Evergreen family, launch is sometime in September and we don't have rock solid rumors yet, so I'm pretty impressed with that.

We must not forget that AMD/ATI has been working on a MCM graphics chip, I'm not saying the RV870 series is going to offer that as R800, but it sure might. Instead of dual GPU cards with separate memory we might see an MCM chip sharing memory and actually not require any driver work as separate chips (as it is dual die single chip). If that is the case we may see an X4 depending on the size of each die.

[OBR]

Spec are true ... new DX11 GFX (GT300 - this is not correct name) will have +/- the same performance like high Oced GTX 295 ... i think 10-20 percent above GTX 280 SLI

[Sh1tyMcGee]

Spec are true ... new DX11 GFX (GT300 - this is not correct name) will have +/- the same performance like high Oced GTX 295 ... i think 10-20 percent above GTX 280 SLI

Wow, that would be impressive! Cant wait for these new cards to come out!

[Sly Fox]

Wow, that would be impressive! Cant wait for these new generation cards to come out!

He's lied about products several times in the past, so I wouldn't bank on it.

In fairness he's been right sometimes as well.

[Barys]

How do you know this when there is no running GT300/G300 yet?

Performance like OCed GTX295:? Hm according to rumours ATIs Cypress is going to be as fast as GTX295 (based on Vantage Performance numbers at least) so if we take that ATIs GPU will be much cheaper and much smaller than GT300/G300 then NVIDIA would get trouble again.

About OBR : You are saying he lied about some products in the past. Well, but i remember when he tested HD4850 and GT200/GT200B before their launch and he said true about their performance but many people didn`t trust him. He said true about new NVIDIA GPU called "GTX275"pushed against Rv790. As i remember he said thatHD4850 are faster than GT200 when he tested them and it was true too so i wouldn`t be so sure that he lies about new NVIDIA GPU at now

[Andrew LB]

Maybe...just maybe... we can finally play Crysis now!!


sorry I couldn't resist, although I'm semi-serious... it's about time that the hardware keeps up with the software.

Seriously though... that joke is getting old.

I play Crysis just fine @ 1920x1200 with 8xAA and 8xAF with everything on "Very High".

.... until i'm in the snow.

Not bad for a GTX-285.

[570091D]

How do you know this when there is no running GT300/G300 yet?

Performance like OCed GTX295:? Hm according to rumours ATIs Cypress is going to be as fast as GTX295 (based on Vantage Performance numbers at least) so if we take that ATIs GPU will be much cheaper and much smaller than GT300/G300 then NVIDIA would get trouble again.

i've heard that nvidia have been making engineering revisions with TSMC, and testing the 40nm process with this new chip.

this will be the first time i've heard of cypress = gtx295, been hearing more consistently that cypress = r700. but if it does reach that performance level, it would give credence to the 1600sp rumor.

both camps have been REALLY tight on info with this generation, release day will be like christmas!

[LordEC911]

i've heard that nvidia have been making engineering revisions with TSMC, and testing the 40nm process with this new chip.

this will be the first time i've heard of cypress = gtx295, been hearing more consistently that cypress = r700. but if it does reach that performance level, it would give credence to the 1600sp rumor.

both camps have been REALLY tight on info with this generation, release day will be like christmas!

Cypress is supposedly performing better than 4870x2, around 4890CF, which is obviously close to a GTX295.

[Barys]

Cypress is supposedly performing better than 4870x2, around 4890CF, which is obviously close to a GTX295.

Cypress scores P16-18k - depends on configuration. GTX295 scores P18-20k. With good OC could get P22-23k easily. So it seems that GT300 could be about 30-40% faster than Rv870. But we should remember that this is only 'Performance" scores and real power of these GPUs should show "Xtreme" benchmarks.

[Smartidiot89]

I wonder how they came about that 2.2x figure? Peak throughput?

My guess would be Gflops, or that the GT300 has 512 SPs compared to GT200 240?

2.2x faster than GTX285 could be true on paper now show it in practice and perfect driversupport

[ajaidev]

On the op's link i failed to find the GT300 but i found RV870 instead


RV870 (Cypress)

SIMD : 20 SIMD: 20
Shader Clock : 850 Shader Clock: 850
Memory Clock : 1200 Memory Clock: 1200
Bandwidth : 153Gbps Bandwidth: 153Gbps
1G GDDR5 1G GDDR5

RV870--Cypress ~P16xxx - P17xxx - P18xxx RV870 - Cypress ~ P16xxx - P17xxx - P18xxx
Rv830 XT--Juniper XT ~P95xx Rv830 XT - Juniper XT ~ P95xx
RV810--Redwood ~P46xx RV810 - Redwood ~ P46xx


什么384bit之类的料进垃圾桶吧. 384bit like, what material it into the trash.

http://74.125.153.132/translate_c?hl...RlGATcDWiLDmFw

[hurleybird]

He's lied about products several times in the past, so I wouldn't bank on it.

In fairness he's been right sometimes as well.

Seeing as the card is supposed to be completely different architecture, and it's unlikely that completed wafers much less cards are out yet, I'd say this one is safe to throw into the BS file.

[jaredpace]

Yah if someone says they have an evergreen card at this time - it's believable. But to say you have GT300 card now.... Kinda bs. But who knows? maybe some people have gt300 protoype cards of A1???

[Chrono Detector]

Hope this is true because I'm getting sick of GDDR3 and a 256 bit memory bus. Bring on GDDR5 and 512 bit (Hope ATI does the same as well)

[zalbard]

I keep wondering if I should let my 295 go, and get ready for a 300...

Way too early to do that IMO.

[zanzabar]

How do you know this when there is no running GT300/G300 yet?

Performance like OCed GTX295:? Hm according to rumours ATIs Cypress is going to be as fast as GTX295 (based on Vantage Performance numbers at least) so if we take that ATIs GPU will be much cheaper and much smaller than GT300/G300 then NVIDIA would get trouble again.

About OBR : You are saying he lied about some products in the past. Well, but i remember when he tested HD4850 and GT200/GT200B before their launch and he said true about their performance but many people didn`t trust him. He said true about new NVIDIA GPU called "GTX275"pushed against Rv790. As i remember he said thatHD4850 are faster than GT200 when he tested them and it was true too so i wouldn`t be so sure that he lies about new NVIDIA GPU at now

obr calls every card that ive seen a call for correct, and the original benches with the 260gtx 192 v 4850 with early drivers they were vary close and the original 260 192 and 9800gtx were close when they came out. there were hhuge changes with the 216 and clock increases + drivers make a huge difference on the gt200

[nn_step]

With the Single fact that it has a 512bit bus, we know it is going to be a huge chip.... All those freakin' traces....