Rumor: Nvidia GeForce 1180 / 2080

[huhn]

the gaming performance deficit comes mostly from the slower infinity fabric that has higher latency than the ring bus from intel.

that's why high speed low latency ram give an AMD more speed boost than an intel CPU because that closes the gap.

the clock speed difference shouldn't be anything more than a worse process and this should be fixed with TSMC 7nm and not a design problem from AMD.

especially for high 120+ FPS gaming intel CPUs are still far better and the reason i still don't have a new CPU...

[Blue_MiSfit]

Mind the toxic behavior, guys. Rule 15 and Rule 4 are both being broken.

[NikosD]

Especially for high 120+ FPS gaming intel CPUs are still far better and the reason i still don't have a new CPU...
So, you are telling us that the reason of not buying Ryzen is because it can play games at 1080p at 140 fps instead of 155 fps of Intel.

Interesting...

[huhn]

yes because your example is the difference between tearing and no tearing using a 144 screen run by fast sync and a ghosting free experience using BFI/back light strobing.

and the difference can be 20 % or more FPS.

[NikosD]

I knew that you would use the 144 Hz excuse, that's why I gave you that example...

So many people nowadays are using ultra expensive displays and ultra expensive GPUs in order to play games at 1080p resolution to reach 144 or more fps.

Right ?

[huhn]

144 hz screen have no special price tag any more and no you don't need expensive GPU it is relative easy to tweak the GPU usage but not the CPU usage at high frame rate.

and yes i personally plan to get a 240 hz screen.

[ShogoXT]

https://www.nvidia.com/content/dam/en-zz/Solutions/design-visualization/technologies/turing-architecture/NVIDIA-Turing-Architecture-Whitepaper.pdf

Bitrate efficiency and quality has gone up! This is a big deal see page 29.

[huhn]

"typical CPU"

there tests shows results where pascal is supposed to be close to x264 using PSNR. but it in term of visual quality it is not even close. x264 CPU utilisation 13 % good job nvidia...

and the 25% bitrate savings on HEVC well maybe they added b frames...

[NikosD]

Ok.

After all the reviews been released to the public, my original assumption still stands alive and kicking.

Turing RTX cards is the worst release ever for nVidia and maybe their biggest mistake.

I don't personally believe that RTX pseudo-raytracing is going to last if AMD doesn't support it.

Regarding RTX cards:

2080 Ti vs 1080 Ti is ~10% at 1080p, 25% at 1440p and 30% at 4K

2080 vs 1080 is ~ 35%

2080 vs 1080 Ti is almost a tie ~ 5%

1080 Ti had an MSRP around 50$ more than 980 Ti but ~ 70% more performance

1080 had an MSRP around 50$ more than 980 but ~ 60% more performance.

The actual difference in price of 2080 Ti vs 1080 Ti is ~ 500$

What a disastrous combination!

Stay away from this fraud.

[videoh]

Turing RTX cards is the worst release ever for nVidia and maybe their biggest mistake. It's so bad that no-one wants these things:

https://www.nowinstock.net/computers/videocards/nvidia/rtx2080ti/

/sarcasm

Take a look at the 5 year stock price history of nVidia to get the true picture.

I don't personally believe... Get over yourself, nobody cares.

NikosD can't afford the world's premiere GPU, so he goes the sour grapes route. What a joke!

[huhn]

so where is the competitions answer to this "fraud"?

it's not like people are happy about this release there is no alternative so they can do what ever they want. and it is not nvidias "propaganda" that created this situation is the fact that vega is a bad card.

[NikosD]

It's so bad that no-one wants these things:

https://www.nowinstock.net/computers/videocards/nvidia/rtx2080ti/

/sarcasm

Get over yourself, nobody cares.

Poor NikosD can't afford the world's premiere GPU, so he goes the sour grapes route. What a joke!Hahaha.

You are so funny guy!

nVidia was preparing for years this release with a leather-jacket-man show, a huge blog regarding RTX, a huge PDF file of 87 pages for...pre-orders.

Because after the release of real reviews, we had a world's first, the stock of the company with a huge marketing campaign to go down.

What a fail!

From your beloved pure capitalistic company Morgan Stanley with love:
We are surprised that the 2080 is only slightly better than the 1080ti, which has been available for over a year and is slightly less expensive.
With higher clock speeds, higher core count, and 40% higher memory bandwidth, we had expected a bigger boost.
https://www.techpowerup.com/247780/nvidia-stock-falls-2-1-after-turing-gpu-reviews-fail-to-impress-morgan-stanley

Poor videoh...Poor Turing...Poor nVidia.

It probably made its biggest mistake in recent history.

RTX for nVidia is like 10nm for Intel or MS Vista for Microsoft.

Big failures.

Let's see how nVidia is going to cope with this my poor friend.

[videoh]

Look at the 5-year plot. Stocks always vary over the short term. Your rhetoric and tone really expose you. "nVidia is a fraud" "leather-jacket-man" blah blah blah. Keep it up. You don't need others to discredit you.

When I get my pair of 2080 Ti's installed I'll think of you. :p

[NikosD]

No, stocks going down after real reviews is not a good sign at all.

But I never said nVidia is s fraud.

I said Turing is a fraud with that price for that performance.

Regarding leather-jacket-man, I don't like a guy presenting such a technical product to have a moto "the more you buy, the more you save"

But it seems he convinced you, at least.

So now you are definitely exposed as a blind nVidia fanatic.

What a pity :(

[huhn]

why don't you provide this blind person an eye opening alternative?

[NikosD]

Good news!

The worst stock price drop in the last 10 years for nVidia :)

https://www.techspot.com/news/77479-nvidia-stock-takes-nosedive-after-disappointing-earnings-report.html

[Atak_Snajpera]

Good news!

The worst stock price drop in the last 10 years for nVidia :)

https://www.techspot.com/news/77479-nvidia-stock-takes-nosedive-after-disappointing-earnings-report.html

Not so good if you are shareholder ;)

[foxyshadis]

We've officially entered another GPU recession; prices and availability will be better, but new designs and technology will be kicked another generation or two down the road, and investment will mostly center around refinements of current tech.

That's not an entirely bad thing, Nvidia's been repositioning itself as a luxury brand for too long now, and it will be nice to have affordable TFlops again.

[huhn]

the real problem here is the process. TSMC 12 nm "is" TSMC 16 nm so to make a faster cards that has place for special cores you have to make even bigger dies which result in higher prices the vram is more expensive too. nothing what people want but what we got the fact that AMD has no present with there vega cards doesn't make it any better.

so the question now is 7nm still to expensive to make smaller but faster GPUs or is nvidia trying to milk the end user here until AMD gets competitive again?

i don't understand how nvidia is a luxury brand when you get more or even performance for the same price as the AMD alternatives (atleast in the past two years).

and why does AMD need 2 different products for high end and main stream. you can downscale vega smaller than polaris much smaller but it doesn't look like it is worth it or where are the mainstream vega cards all these years?
but we got vega 64 which is far more expensive in production than a 1080 the HBM2 alone cost 3x GDDR5 and the die is bigger too...

[NikosD]

More and more I think that Raja Koduri doomed AMD GPUs with his Vega design.

He released a chip at 14nm with maximum CU (Compute Units) of 4096 and he couldn't match even 2nd place card of nVidia - Pascal 1080.

Now, even using 7nm, Vega can't upscale to more than 4096 CUs!

Practically we have to wait for another architecture from AMD to hit high-end cards and that would be probably the large Navi 20 at the end of 2019 (!)

If Raja wasn't paid by nVidia or Intel, he should.

Maybe that's why he is working officially for Intel now at discrete cards section.

He managed to ruin a generation of AMD's GPU cards in order for Intel to catch up and nVidia to try crazy things like pseudo-raytracing without consequences.

Anyway, I'm not interested in >1080p gaming so I'm going to wait for my 7nm RyZen 3000 series along with the small Navi 12 also at 7nm in Q2 of 2019.

[huhn]

well someone has to say it.

the major problem with RTX was that user didn't get more performance for there money. so now that AMD plans a card that does exactly the same thing by matching the price and performance with a 2080 which pretty much justified the RTX cards pricing...

worst possible outcome.

[nevcairiel]

The Vega architecture just wasn't that great. 300W on 7nm, with memory bandwidth to the moon and back, and they can only match a 2080, which does all of that on 12nm at 225W and only half the memory bandwidth. And not even for cheaper (MSRP is the same as 2080 non-FE)
How bad Vega was/is really shows with this release. Imagine what Turning could do with 300W on 7nm and that massive memory bandwidth.

The only people that might truely be interested in Radeon VII are possibly workstation users that can make full use of its 16GB memory and fast FP64 calculations.

Its a shame though that Navi just doesn't seem ready yet, and even if it was, its likely they are not going to launch anything high-end with it anytime soon, because that would directly undermine the Radeon VII.
Its no secret that I prefer NVIDIA cards at this point, but thats just a product of how underwhelming AMDs recent generations have been. And even if one isn't necessarily interested in buying an AMD card, competition drives innovation and keeps prices in check. And there just is none.

At least they managed to step back onto the CPU playing field, even though they are still playing catch-up in IPC performance. Hopefully the next 5-6 months of finalizing Ryzen 3000 can at least close the gap to Intel, and maybe even make a bit of headroom.
H2/Q4 2019 and 2020 will be interesting for CPUs. For GPUs, we'll have to see.

[NikosD]

Radeon 7 is an all around player.

Yes, it can play games around 30% faster than Vega so it matches 1080 Ti/ 2080 cards, but it can also be used as a workstation card.

It's not only the extremely fast FP64 performance, the out of this world memory bandwidth or the massive 16GB VRAM

It is also the drivers for professional apps, which should be similar to Vega Frontier.

So, a game dev for example could design and play his game on the same card, just by using a switch in the drivers.

HBM2 is still expensive obviously.

Now, Ryzen 3000 series shown at CES completely destroyed the competition.

A medium Ryzen 5 with only 8 cores, matched the performance of the top Core i9 9900K (!) using 130W vs 180W for Intel and without top clocks yet, because it was an engineering sample.

You should expect Ryzen 7 with 12 cores and Ryzen 9 with 16 cores.

Regarding Navi, it will come in summer like Ryzen 3000 series, smashing the middle category performance in terms of price/performance ratio.

Definitely this year belongs to AMD.

[huhn]

a german news site reported that the FP64 well be crippled.
and the 128 rop are wrong is the same ROP as mi50 64.

https://www.computerbase.de/2019-01/vega-20-amd-radeon-vii-fp64/

this is yet again a professional card salvaged for the consumer market.

[nevcairiel]

AMD wants to sell the MI50 after all, the only real reason to get it is the FP64 performance then, it's otherwise pretty close to a Radeon 7.

[Atak_Snajpera]

Personally I'm waiting for 75W Navi with RX580 performance.

[NikosD]

The last few days I'm under the impression that 1160 Ti or 1660 Ti or whatever name nVidia eventually decides to give to the Turing card with Tensor and RT cores disabled, will be a good card.

It seems to be a lot faster than 1060 6GB, RX 580 & RX 590, with a performance closer to 1070 - a little slower actually.

The initial estimation of its price is ~250$ and if that is true, I think I'm going to buy it.

I'll wait for Navi in summer along with Ryzen 3000 series for my new PC rig.

[huhn]

i highly doubt that a card with disabled feature will cost this much less the production cost is still the same.

if nvidia is able to produce new chip with no tensor or RT cores i can see a "huge" price difference.

but still if you don't need a new GPU right now i would wait for 7nm card which should give better performance for the money.

not sure if tensor cores should be disabled at all they will or should have a use case in "all" games at one point.

[clsid]

The chips are designed in such a way that broken parts can be disabled, while keeping a functional chip. So imperfect chips get sold as a lower model.

[huhn]

yes that'S known but the tensor core are a very specific part.

not like one of these core cuda packs where the defect can be at any of these.

[NikosD]

It seems that we only have to wait till February 15 and spend around 280$ in order to get a 1660 Ti.

Let's wait for reviews to find out if the performance is closer to 1070 or 1060.

https://www.tomshardware.com/news/nvidia-geforce-gtx-1160-1660-ti-specs-palit,38505.html

[wanezhiling]

I am interested in Intel's dGPU(Gen 12) in 2020

[NikosD]

Tough times for nVidia...But why ?

RTX is such a successful implementation with such a good price [emoji1787]

https://www.anandtech.com/show/13965/nvidia-earnings-report-q4-2019-crypto-pain-but-full-year-gain

[videoh]

Not so tough:

"For the full fiscal year though, earnings were still very solid, with revenue up 21% to $11.7 billion, and an overall gross margin of 61.2%, up 1.3% from 2018. Operating income was $3.8 billion, up 19%, and net income was $4.1 billion, up 36%. Earnings-per-share for all of 2019 came in at $6.63, 38% higher than 2018."

And don't forget nVidia also has about a 2 to 1 market share advantage over AMD, whose Radeon 7 has been a disappointment.

I'm loving my RTX 2080 Ti! The Turing SDK is out now and life is swell.

[huhn]

doesn't change that the company lost more than 30 % of it's worth in the last couple of month.

reviews and customer are not thrilled by these RTX cards.

I'm loving my RTX 2080 Ti! The Turing SDK is out now and life is swell.
this wasn't available on release date?

[nevcairiel]

Losses or not, without competition its not going to really matter, and AMD has only had very weak showings.

[huhn]

it clearly doesn't matter for the consumer and it shouldn't matter in the long run.

but if you are at the stock market than this is important. if is justified or not doesn't matter for the stock market.

[videoh]

Yeah, every time the nVidia stock price changes, the performance of my RTX 2080 Ti follows right along.

nVidia haters are getting desperate!

[huhn]

did you even read what i said?

i take it now as an compliant to be known as an nvida and AMD hater.

[videoh]

Haters gonna hate.

Work on your English and grammar; you're embarrassing yourself.

[huhn]

thank you for your contribution to this topic.

[NikosD]

One of the most terrifying moments that I recently had, was without doubt the reading of the following article:

https://www.tomshardware.com/news/softbank-sells-nvidia-shares,38564.html

SoftBank, a large investment fund, sold all of its nVidia shares - around 3.6 billion $ ! - due to low sales and no bright future ahead.

I have to repeat myself unfortunately.

Tough times for nVidia...

[Blue_MiSfit]

Let's keep it on topic, folks :)

[NikosD]

1660 Ti is eventually out as a new TU116 which is a Turing architecture die but without RT and Tensor cores.

The performance is close to 1070, so it's like 40% faster than 1060 6GB in games, but with the same TDP of around 120W.

With an MSRP price of 280$ it's definitely the most interesting Turing card to buy nowadays.

[huhn]

the 1660 ti give me a lot of question marks.

first is DLSS which pretty much lowers picture quality to increases performance so something i see as useful for a not high end GPU but it's missing i totally see why RT core are missing they are pretty useless on such a card but tensor cores?

next is the GPU size the relative TU116 size compare to a TU106 show it's only a little bit smaller per cuda core. the next thing are the design diagrams: https://www.pcgamesn.com/wp-content/uploads/2019/02/Nvidia-RTX-and-GTX-Turing-SM.jpg
so TU116 has dedicated FP16 cores where we can't see them on TU106 so the tensor core are doing the FP16 math?

[nevcairiel]

It's quite likely that the tensor cores are being recycled to do the FP16 math, it's far more likely then actually making new hardware.

The reason full tensor cores are disabled is the same old story, feature segmentation. They reckon gaming doesn't really need it, and they don't want some workstation loads going for cheap cards.

DLSS is specifically designed to offset RT cost. It's not designed to allow 4K gaming on low-end hardware. These are 1080p GPUs, the quality loss would probably be too extreme on that resolution (since it's more apparent in lower DPI)

[huhn]

i highly doubt DLSS is designed to off set RT it clearly very important for RT with it's very medicore performance but there are games that support DLSS but no RT like final fantasy 15 and DLSS can be used an alternative form of AA.
so even on a 1080p screen it has it's uses it just depends how it is used.

if i'm not mistaken metro exodus "struggles" to say it friendly with DLSS at "low" resolution but it wasn't used for super sampling in that game AFAIK.
The reason full tensor cores are disabled is the same old story, feature segmentation. They reckon gaming doesn't really need it, and they don't want some workstation loads going for cheap cards.
i totally understand your argument.
but i have to argue giving more user access to this specialised feature should create more program that use it and so more demand for these cards.

It's quite likely that the tensor cores are being recycled to do the FP16 math, it's far more likely then actually making new hardware.

i didn't knew they could do that i through they where extremely specialized in matrix operations.
i mean they can do some operation/outputs in FP32 so clearly not optimised for pure FP16

and the next thing is FP16 processing affecting tensor core performance unlike int32/FP32? if they are the reason turing can do great FP16 it has to be like this.
as a comparison polaris can do FP16 at the same speed as FP32 while pascal can do 1/64 of the FP32 so this is huge opportunity to write/modify programs that benefit from this.

(i take on paper numbers not real world test here)

[NikosD]

According to anandtech, based on nVidia sources the design of TU116 is unique regarding FP16 support, introducing for the first time dedicated FP16 cores.
Of course, as we just discussed, the Turing Minor does away with the tensor cores in order to allow for a learner GPU. So what happens to FP16 operations?
As it turns out, NVIDIA has introduced dedicated FP16 cores!

These FP16 cores are brand new to Turing Minor, and have not appeared in any past NVIDIA GPU architecture.
Their purpose is functionally the same as running FP16 operations through the tensor cores on Turing Major: to allow NVIDIA to dual-issue FP16 operations alongside FP32 or INT32 operations within each SM partition.
And because they are just FP16 cores, they are quite small. NVIDIA isn’t giving specifics, but going by throughput alone they should be a fraction of the size of the tensor cores they replace.

[huhn]

there is the next thing that puzzles me.
turing is supposed to be able to do integer and float at the same time but can it do FP16, FP32 and int32 at the same time? they are different cores aren't they?

[nevcairiel]

The normal Turing SM core can do int and FP at the same time now. On big-turning, FP16 can be done by the tensor cores, so it can also run in parallel - and on little-turing, the dedicated FP16 cores, if they are recycled tensor cores or not.
In Volta, the SM cores did FP16 still. This is also a Turing change.

I don't think they can tripple-issue commands to the same SM unit though, so it'll probably not actually do all three at the same time, but it may be flexible to do any two at the same time.