Entire GeForce 400M lineup revealed

So far we've only officially seen the 480M, but now the rest of the cards (from 470M to 415M) have been revealed. Nvidia's lowest cards have approximately doubled in performance since the previous generation. The higher end cards have significantly higher processing power, but memory bandwidth may be an issue. I am curious what the TDP and battery life is for these.

 

I have a question, is the Nvidia 400M generation a response to the ATI 5xxx or the ATI 6xxx generation?

 

heres the official newsroom release from Nvidia NVIDIA Newsroom

 

Memory bandwidth being a problem? It has a 192bit GDDR5 configuration. That's a huge bandwidth specially compared to their last gen's offerings.

And well, as they say in the article, the cuda cores of fermi ar actually weaker, so the jump from 200m to new 400 in the high end area, is not that much.

 

Geez can they confuse customers any more with their GT 445M. Huge difference comparing 128-bit GDDR3 vs. 192-bit GDDR5 cards... The should have just made the higher end one the GTS 450M. At least the entry level cards are now 128-bit...

 

The issue of bandwidth is iffy. I don't know why, but even though Nvidia's GDDR5 bandwidth is huge, it's not that much faster than AMD's if any. The HD6xxx I thought I saw was still 256 bit, but it's bandwidth is way higher, destroying the 480M

I know the 480 will it's 384 bit, it's only 900mhz. And believe it's even slower for 480M.

For example, I know an overclocked HD5770 runs at 1400 for memory stable at 75C under load.

Granted the GDDR5 on HD5870M is only 128 Bit, but it runs at 1100-1200. My guess the head to head competition, 460M, the memory speed likely much slower.

 

Just the charts, for those who can't access Anandtech:

Yeah, the 445M packs way more power than I was expecting. If it's wattage is low, it will absolutely crush the mid-range.

 

No, it is not. The GTX260M actually has a marginally higher bandwidth (60.8 GB/s as opposed to 60.0 GB/s for the 460M and 470M). Bandwidth is not just the bus width and memory type, the clock speed also matters.

This is true, but the processing power did in fact go up (relative to the 260M a little bit for the 460M, a lot for the 470M) whereas the bandwidth essentially stayed constant.

My guess is that they're counting on the fact that nearly all displays currently on the market are inferior to what was available in the last generation. You can't get 1920x1200 anymore, it has to be 1920x1080. Likewise, 1680x1050 becomes 1600x900. And of course most of the laptops sold today come with 1366x768 garbage for which the bandwidth of these GPUs is way, way overkill.

 

You seem to know your stuff. Any idea why the memory bandwidth is only 60 GB/s considering it's 192 bit GDDR5? The HD5870M has 128 Bit @ 64 GB/s. Granted I'm still thinking it may have to do with 5870s faster ram speed.

 

It's due to the memory speed. And I meant it is huge because for laptop GPUs, it is still 60GB/s.

I was also counting on my G51J memory bandwidth which iwas 51GB I think, I forgot it was underclocked so I thought it was better on 460m.

 

Yes, it's because of the faster memory clock. You can calculate the bandwidth by simply multiplying:

192 bits * (1 Byte / 8 bits) * 2 (due to DDR) * 1.25 GHz = 60 GB/s

The only tricky part is that with GDDR5, sometimes the memory clock speed given is divided by 2 (so you have to multiply) and sometimes it is not (here it appears to be the latter because otherwise the 470M would have a higher clock speed than any of the desktop Fermi cards).

 

NVIDIA?s GeForce GTX 480 and GTX 470: 6 Months Late, Was It Worth the Wait? - AnandTech :: Your Source for Hardware Analysis and News