NVIDIA Fermi: Architecture discussion

NVIDIA GF100 Fermi Video Card Triple SLI Rocket Sled Demo


http://www.legitreviews.com/news/7137/
Legit Reviews got a chance to look at test system that had not one, but three GF100 Fermi graphics cards running in triple SLI demo. The demo is being narrated by NVIDIA and all looks great until the very end. GF100 is due out in March 2010, so the driver team is still tweaking the software on these graphics cards. NVIDIA also announced that GF100 was in volume production today, so it looks like Fermi will still be out in Q1 2010.

http://www.youtube.com/watch?v=3-1GMbqzGX0
 
So, you admit that it does consume more than a GTX 280, the hotest/highest power consumption single chip GPU ever? I know I had a BFG GTX 280 about a month after launch and it would overheat if I didn't manually adjust the fan to max, and they said nothing was wrong with it, it just runs that hot.


Hmm its not the hottest and highest power consumption either. I don't know about you but my gtx280 is pretty much on 24/7 (don't turn my computer off) and had no issues with heat.
 
http://www.anandtech.com/tradeshows/showdoc.aspx?i=3719&p=3

anand is speculating 20% faster than the 5870. Should be another good generation for all us costumers as a price war is bound to happen.

Actually I would predict the opposite, relatively stable pricing. Nvidia don't have any incentive to engage in price competition with AMD with a more expensive part. They will use their natural brand name advantage to extract a price/performance margin whilst AMD will want to keep their margins and not squander them. They made very little money with the RV770 considering their advantages, they need profit more than anything. I doubt in that situation they would do little more than price to balance margins/unit sales and the price cuts will be more inline with how the cost of production scales over time.

Edit: He said at least 20% faster not 20% and no more.
 
Actually I would predict the opposite, relatively stable pricing. Nvidia don't have any incentive to engage in price competition with AMD with a more expensive part. They will use their natural brand name advantage to extract a price/performance margin whilst AMD will want to keep their margins and not squander them. They made very little money with the RV770 considering their advantages, they need profit more than anything. I doubt in that situation they would do little more than price to balance margins/unit sales and the price cuts will be more inline with how the cost of production scales over time.

Edit: He said at least 20% faster not 20% and no more.

yes at least 20% faster. However a refresh of the 5870 and perhaps the 1ghz verisons we will see will end up being faster in some situations than the fermi high end also.

Which will cause a price war as i'm sure that amd with the smaller chip will want to press nvidia with the larger chips
 
It seems that Anand wasn't told anything other than that Fermi will supposedly be "very competitive" even with release date several months after HD 5870. He is guessing that this means at least 20% faster, but that appears to be just a guess on his part (after all, who really knows what the true differences will be).

Due to supposed launch date two months from now, looks like it's going to be a relatively quiet period of time for the next few weeks before the storm hits. Can you believe it's been more than 3 years since G80 launched?
 
Last edited by a moderator:
GF104 specs?

Take this with a grain of salt, but did anyone notice the Anonymous post on BSON with GF104 specs:

NVIDIA GeForce GTX 395 Specs :

- Codename "GF104".
- Dual Core GPU Design (Two GF100 "Fermi" Cores).
- 6.4 Billion Transistors In Total (TSMC 40nm Process).
- 32 Streaming Multiprocessors (SM).
- Each SM has 2x16-wide groups of Scalar ALUs (IEEE754-2008; FP32 and FP64 FMA).
- The 32 SMs Have 1536KB Shared L2 Cache.
- 1024 Stream Processors (1-way Scalar ALUs) at 1350MHz.
- 1024 ALUs In Total.
- 1024 FP32 FMA Ops/Clock.
- 512 FP64 FMA Ops/Clock.
- Single Precision (SP; FP32) FMA Rate : 2.76 Tflops.
- Double Precision (DP; FP64) FMA Rate : 1.38 Tflops.
- 256 Texture Address Units (TA).
- 256 Texture Filtering Units (TF).
- INT8 Bilinear Texel Rate : 153.6 Gtexels/s
- FP16 Bilinear Texel Rate : 76.8 Gtexels/s
- 80 Raster Operation Units (ROPs).
- ROP Rate : 48 Gpixels
- 600MHz Core.
- 640bit (2x320bit) Memory Subsystem.
- 4200 MHz Memory Clock.
- 336 GB/s Memory Bandwidth.
- 2560MB (1280MB Effective) GDDR5 Memory.
- New Cooling Design.
- High Power Consumption.

GTX395 is 60% faster than GTX380
GTX395 is 70% faster than HD5970
Release Date : May 2010, Price : 499-549 USD.
 
A full chip but than only 320bit memory interface doesn't make sense. Fake. ^^

But a pretty good guess. 600 Mhz Core for the dual chip when the single chip fermi will come with about 650 Mhz.
 
What is wrong with 320-bit memory bus for each GPU? Wouldn't that be akin to [rumored] GTX 360 SLI setup (albeit with a bit more processing power due to not having ALU's disabled)?
 
Take this with a grain of salt, but did anyone notice the Anonymous post on BSON with GF104 specs:

*snip*

GTX395 is 60% faster than GTX380
GTX395 is 70% faster than HD5970
Release Date : May 2010, Price : 499-549 USD.

Link to this "speculation"?

The price listed is a pretty big reason to doubt this "GF104" off the bat anyways, since there's no way Nvidia would sell a card with those supposed performance numbers at that price. And second, the rest of the specs aren't exactly hard to figure out... they basically took a full 512-core GF100 and gave it the GTX295 treatment..
 
Link to this "speculation"?

The price listed is a pretty big reason to doubt this "GF104" off the bat anyways, since there's no way Nvidia would sell a card with those supposed performance numbers at that price. And second, the rest of the specs aren't exactly hard to figure out... they basically took a full 512-core GF100 and gave it the GTX295 treatment..

http://www.brightsideofnews.com/news/2010/1/7/amd-sold-two-million-directx11-gpus!.aspx

Yes, nothing too unimaginable here. With a purported launch in May (five months from now!), the competitive pricing really cannot be set properly right now.
 
yes at least 20% faster. However a refresh of the 5870 and perhaps the 1ghz verisons we will see will end up being faster in some situations than the fermi high end also.

Which will cause a price war as i'm sure that amd with the smaller chip will want to press nvidia with the larger chips

Thats simply speculation at this point, even going from the recent interview we saw that AMD expects Fermi to be faster on a single chip vs single chip basis. They will of course be pleasantly surprised if the situation is different.

A price war requires at least one party which wishes to engage in one, since there are of course only two parties. In this case Nvidia has no incentive to engage in price based competition and AMD aren't exactly the most profitable company in the semi-conductor industry. If theres to be pain for Nvidia it will come from an uncompetitive Fermi chip in the short term and be purely self inflicted and if theres pain in the longer term it can only really likely come from either R9xx releasing this year OR an early die shrink of the current AMD lineup.

At this point im not willing to call Fermi uncompetitive nor AMD willing to give up much needed profit for market share.
 
The clocks are completely unrealistic. Two full GF100s with shaders at 1350MHz? Certainly not within a 300W power envelope.
 
What is wrong with 320-bit memory bus for each GPU? Wouldn't that be akin to [rumored] GTX 360 SLI setup (albeit with a bit more processing power due to not having ALU's disabled)?

Don't know if it's possible to disable only the ROPs (to which the bus is linked). I think that in order to castrate a chip you should disable entire blocks (ALUs, TMUs, ROPs).
Thus, F A K E. :LOL:
 
Back
Top