Nvidia Pascal Announcement

[lanek]

Lucky are the guys who have bought a TitanX .....

1080TI vs TitanX .....

- 3854 SP vs 3854SP
- 11.5 vs 11 Tflops ..
- 11GB memory vs 12GB GDDR5x
- 484 GB/s bandwith vs 480 GB/s
- 699$ vs 1200$

If i was own a Titan XP, i will ask for a refund.. I dont even know why Nividia have change the spec on Vram, they could have just do a price cut. ( why do complicate when you can do simple.. )

TitanXP will be unsellable on Amazon or Ebay now ..

 

[Voxilla]

Lucky are the guys who have bought a TitanX .....

1080TI vs TitanX .....

- 3854 SP vs 38584SP
- 11.5 vs 11 Tflops ..
- 11GB memory vs 12GB GDDR5x
- 484 GB/s bandwith vs 480 GB/s
-699$ vs 1200$

If i was own a Titan XP, i will ask for a refund..

Of course nothing stops them overclocking to even better specs

 

[Alexko]

Well, it's pretty simple: if you buy a graphics card that costs as much as minimum wage workers make in a month, expect something better for much less money in short order.

 

[lanek]

Well, it's pretty simple: if you buy a graphics card that costs as much as minimum wage workers make in a month, expect something better for much less money in short order.

Expect the same .. hence why i say, they could have just price cut it.. why use a different memory bus ? lol . I know engineer need to come with something for justify their wage, but well.. useless..

 

[Jupiter]

I know several people with a Pascal Titan X and they are not annoyed at all. At least not the ones who bought it immediately. Who bought it a month ago will surely be annoyed.

Btw. the Titan costs about 1400 USD here with taxes and weak Euro. ^^

 

[pharma]

Like similar Pascal products I expect the 1080 Ti pricing between $625 - $699, with the Founder's price setting the ceiling. Titan XP production can stop at any point since it's only available on Nvidia shop.

 

[Jupiter]

The founders edition should not be more expensive this time.

 

[Kaotik]

Killed AMD Radeon Vega with that price move.

So you actually have Vega card and pricing information on hand? Care to share?

 

[Jupiter]

There are some big video games coming out in March. The GPUs are needed before and not after.

In addition, the availability at AMD GPUs was often a problem.

 

[ImSpartacus]

Expect the same .. hence why i say, they could have just price cut it.. why use a different memory bus ? lol . I know engineer need to come with something for justify their wage, but well.. useless..

They have tons of reasons to cut the memory bus.

• Micron's 11 & 12 Gbps GDDR5X just went into mass production a short while ago, allowing a cut-down memory subsystem to incur no loss of bandwidth compared to Titan X.
• GDDR5X probably isn't cheap and using ~8% less of it helps them cut costs.
• They surely do have some GP102 with defective memory controllers/ROPs, so this helps them salvage more usable chips - another cost control measure.
• A <12GB 1080 Ti allows for a refreshed 2017 Titan to sport a numerically superior amount of vram (like every Titan in history) without forcing them to double up and go to 24GB of expensive GDDR5X.

So doing this put them in a position to maintain the performance of Titan X at a lower cost and put them in an excellent position to release a refreshed 2017 Titan.

So you actually have Vega card and pricing information on hand? Care to share?

I think he's just pointing out that Nvidia's "horse" in the race is a cut down 471 mm2 chip that's been in production for nearly a year while AMD's "horse" is a ~500+ mm2 chip that requires an interposer and a pricier variety of vram.

That's not a pretty sight for AMD.

And I like to think that if Nvidia felt that it needed to configure the 1080 Ti with more compute or memory resources, then they definitely would've done so. The fact that they didn't says something about Vega (or at least about Nvidia's expectations for Vega). They did it before with the 980 Ti, so I have a certain amount of trust that Nvidia could do it again.

 

[Malo]

Well here's the faster memory for the 1080/1060 explained, options for card makers to create faster versions. Makes sense.

http://www.anandtech.com/show/11173...eforce-gtx-1080s-gtx-1060s-with-faster-memory

To be clear, NVIDIA isn’t releasing a new formal SKU for either card. Nor are the cards' official specifications changing. However, if partners would like, they can now buy higher speed memory from NVIDIA for use in their cards. The resulting products will, in turn, be sold as factory overclocked cards, giving partners more configuration options for their factory overclocked SKUs.

As factory overclocking has always been done at the partner level, this doesn’t change the nature of the practice. Partners have, can, and will sell cards with factory overclocked GPUs and memory, with or without NVIDIA's help. However with NVIDIA’s official specs already driving the memory clocks so hard, there hasn’t been much headroom left for partners to play with; factory overclocked GTX 1080 cards don’t ship much above 10.2Gbps. So the introduction of faster memory finally opens up greater memory overclocking to the partners.

 

[Voxilla]

Well, it's pretty simple: if you buy a graphics card that costs as much as minimum wage workers make in a month, expect something better for much less money in short order.

What then you think of a certain brand of smartphone makers, that thinks charging 1000$ is a great idea.

 

[iMacmatician]

But the thing that surprised me was the 9Gbps GDDR5. What's going on there? Who's making that?
[…]
I thought GDDR5 was only supposed to get up to 8Gbps. Is this really just an overclock via aggressive binning?

Samsung mentioned 9 Gbps GDDR5 in a HBM2 news release. The GDDR5 catalog is here but I don't know if the 9 Gbps GDDR5 is among them because I'm not sure how to interpret the speed numbers.

What then you think of a certain brand of smartphone makers, that thinks charging 1000$ is a great idea.

Is that a reference to the rumor that the "iPhone 8" may cost over $1000?

 

[lanek]

They have tons of reasons to cut the memory bus.

• Micron's 11 & 12 Gbps GDDR5X just went into mass production a short while ago, allowing a cut-down memory subsystem to incur no loss of bandwidth compared to Titan X.
• GDDR5X probably isn't cheap and using ~8% less of it helps them cut costs.
• They surely do have some GP102 with defective memory controllers/ROPs, so this helps them salvage more usable chips - another cost control measure.
• A <12GB 1080 Ti allows for a refreshed 2017 Titan to sport a numerically superior amount of vram (like every Titan in history) without forcing them to double up and go to 24GB of expensive GDDR5X.

So doing this put them in a position to maintain the performance of Titan X at a lower cost and put them in an excellent position to release a refreshed 2017 Titan.



I think he's just pointing out that Nvidia's "horse" in the race is a cut down 471 mm2 chip that's been in production for nearly a year while AMD's "horse" is a ~500+ mm2 chip that requires an interposer and a pricier variety of vram.

That's not a pretty sight for AMD.

And I like to think that if Nvidia felt that it needed to configure the 1080 Ti with more compute or memory resources, then they definitely would've done so. The fact that they didn't says something about Vega (or at least about Nvidia's expectations for Vega). They did it before with the 980 Ti, so I have a certain amount of trust that Nvidia could do it again.

Im not quite sure this worth 400$ less .. Titan have 6 memory chips ... you want to count the ratio of what it cost with 5 ?

When i look at the 1080TI.. i know allread Vega y is performing well better than the 1080... 11GB vs 8b HBM2... same amount of SP that Titan ..... An incredible price cut on Titan in the form of the 1080TTI.. no worry to have... Or they will have release a 1080TI with 2 SM less and 8GB of ram. The question is .. Vega is performing better of it or not ? .. the 1080TI will be faster than Titan in every situation.. ( for the standard one, wait the non funders arrive with 1800mhz and more on the table .. )

 

[Voxilla]

Is that a reference to the rumor that the "iPhone 8" may cost over $1000?

There is also a rumor Samsung will be able to match that

 

[mczak]

Samsung mentioned 9 Gbps GDDR5 in a HBM2 news release. The GDDR5 catalog is here but I don't know if the 9 Gbps GDDR5 is among them because I'm not sure how to interpret the speed numbers.

The speed numbers are access time, though it's non-obvious which digits mean what. In any case 25 = 8000 Mhz (really 4000 but well, ddr), 28 = 7000 Mhz, 03 = 6000 Mhz, 04 = 5000 Mhz.
So the 22 part should be 9000 Mhz, albeit it's only listed as a 4gb part (hence, pretty much useless for the 1060 - surely you want the faster memory on the 6GB versions...). Also, if you click on the part (which btw is still saying customer sample) it says vdd/vddq
1.309 to 1.597 V, unfortunately no datasheet where you can see which exact part has which voltage. My educated guess however is that the 9gbit/s part requires 1.6V - so good old factory overclocked/overvolted parts (which has traditionally been the case for new gddr5 speed grades).

edit: the website seems to be outdated. If you take their "product selection guide" pdf from 2017, they list -22 parts both for 4gb and 8gb chips, with 1.5V (albeit don't ask me what the asterisk there means - it could mean it requires more voltage...): http://www.samsung.com/us/samsungsemiconductor/pdfs/PSG2017_1H_HR_singles.pdf
You can also easily see there are no actual separate parts for low (1.35V) vs standard (1.5V) voltage.

 

[LiXiangyang]

The only benefit of Titan XP is that it can be treated as a tesla card without ecc, through enabling tcc mode,. whilst geforce product dont enjoy this feature.

However the problem is at the moment the tcc mode has bugs on Titan X Pascal, such that it cannot coexist with the WDDM, I know that because I reported this bug to nvidia. Anyway it is anyone's guess whether this tcc feature along justitfy the extra $500's cost, but there are people paid thousands of dollars more on cards with this feature+ECC etc.

 

[pharma]

Geforce GTX 1080 Ti - Time Spy Scores + OC - Spotted In Futuremark ORB

In the Futuremark results database several entries of the Geforce GTX 1080 Ti have been spotted. It is not exactly a hard product series to seek with it's unusual 11 GB graphics memory. The scores are based on Time Spy results.

The card is compared with Titan X (OCed at 1772 MHz) with memory at 12251 MHz. In that configuration the 1080 Ti is able to beat that card and config with a graphics score of 9049 points (I am referring towards the GPU score here btw).

The 1080 Ti shows a 1481 Mhz clock frequency and 1376 Mhz on the memory, this card scores 9239 points. There also is an entry to be found with the card tweaked to 1860 MHz, here the cards cores 9303 points. That tweak does mean it would boost over 2000 MHz. Check for yourself here.

Update: there is also a score spotted with the card overclocked seriously, scoring 10952 in Time Spy .. not bad! That card is the GTX 1080 Ti at 2062 MHz (Boost) with memory at 5702 MHz.

http://www.guru3d.com/news-story/geforce-gtx-1080-ti-spotted-in-futuremark-orb.html

 

[pharma]

Nvidia GPU powered DeepStack AI beats poker professionals

Nvidia boasts of passing a new milestone in gaming AI. Scientists from the University of Alberta's Computer Poker Research Group, with help from researchers at Charles University in Prague and the Czech Technical University, managed to create a GPU-trained AI that can defeat poker professionals in heads-up No Limit Texas Hold'em.

Poker is an interesting game for an AI to succeed in. Unlike previous AI gaming milestones where players, including the computer, have 'perfect information' - poker players are challenged by many unknowns. In Chess, Draughts, or Go for example, all the pieces can be seen on the board, the 'imperfect information' aspect of poker comes from the many unknown and unrevealed cards in a game. Add to this uncertainty the intuition and strategy honed by poker masters over years of play, and its difficult to train an AI to win.

The success of the GPU-trained DeepStack AI has repercussions outside of gaming. In the real-world there are very many problems to solve where imperfect information is a feature. DeepStack's training and success in using its training to win shows that similar GPU AI training could be used for complex problems in business, industry, science and so on.

Interestingly, now that it has been trained, DeepStack can "run on a gaming laptop" (with a GTX 1080 GPU). Meanwhile a rival No Limit Texas Hold'em playing supercomputer called Libratus (which uses 'a thousand CPUs', cost nearly $10 million, and packs 274 terabytes of RAM), is about half as fast as DeepStack in gaming decisions, say the CPRG researchers.

http://hexus.net/tech/news/software...wered-deepstack-ai-beats-poker-professionals/

 

[Jupiter]

NVIDIA TITAN Xp: http://www.nvidia.de/graphics-cards/geforce/pascal/titan-xp/#redirected

1349€

Game Ready For The NVIDIA TITAN Xp
"The new NVIDIA TITAN Xp is the world's most powerful graphics card, giving you the power to accomplish things you never thought possible. It features 12GB of next-gen GDDR5X Video Memory (G5X VRAM) running at 11.4 Gbps, has 3,840 CUDA Cores, and runs at 1.6 GHz, for a total of 12 TFLOPS of performance."