Anybody else unimpressed with new midrange Nvidia GPUs, and much higher MSRP?

[airfathaaaaa]

DX12 and the console effect is no longer an advantage for GCN with the changes in Pascal. If devs optimize for frequent compute/copy queues, it will run just fine on Pascal. If devs optimize for wavefront/work groups of 64, it will run great on Pascal.

DX12 Async Compute however, will remain a GCN advantage, so there's potential for 10-20% uplift for titles that support it. But not many will, AMD has to sponsor the development and add it.

This is a NV responding to the chess play AMD set in motion years ago. It works in NV's favor too, because as more games become DX12 or GCN optimized, Kepler is dead and Maxwell takes a dirt nap. Do you know what the result will be? People will upgrade to Pascal. NV is going to bank it big time unless AMD can somehow shift the mindshare back to their side. But to me, it seems like the majority, if they look to upgrade cos their older stuff is not performing, they look to the latest NV GPU.

we only know for sure that they will use fine grained pre emption and we dont know if its going to be on drivers or hardware...
other than that we dont really know anything since if they had anything else to show they would have

 

[maddie]

298mm2(from a die shrink alone) 596mm2

so p10 will be larger than a 390x(438mm2/2) relative speaking

The range of speculation is 2X to 2.2X density of 28nm. Polaris at 232mm^2 would be 464 - 510mm^2 28nm equivalent.

As I posted before, the back end at 20nm gives more leeway to reach these maximums.

IF Polaris can reach the 2 - 2.2X density, then right there you're in Fiji territory. Remember Fiji is not very good at actually using the 4096 shaders relative to previous GCN GPUs.

Add shader and other efficiency improvements and you're in 980Ti territory.

Additional clocks anyone? Where are we then?

 

[book_ed]

AMD isn't going to go cheap on Vega with HBM2. The 4870 and 5870 days a long gone.

Offering 90-100% of the performance of the GTX 280 Ultra for half the price didn't do AMD much good besides marketshare %, they didn't even profit much from that entire "victory" of two generations with the 4800 and 5800 series.

Because the market doesn't reward AMD when they hit success. At best, they are looking at 50% marketshare by selling superior GPUs at bargain break even prices and that was back in the days. NV's mindshare has undoubtedly grown since then, proof with their Founder's tax hike. If AMD strikes a home-run, how much % marketshare do you think they can get? 35%? Should they price it dirt cheap or go high and profit...

Polaris will be priced well because it's small and GDDR5 is cheap. Vega will bring the performance crown and it's gonna be hella-expensive.

The problem with AMD/ATI was that they didn't play the marketing game like nVIDIA did. They demoed some AI working on the GPU around the 4xxx series, some GPU physics and was talk about working with Havok for it, but nothing rather impressive. A lot of games were with "The way it's meant to be played" logo which helps quite a lot.

Having the better product only helps if you know how to market/sell it! It was the same with HD7000/R290 series.

 

[maddie]

The problem with AMD/ATI was that they didn't play the marketing game like nVIDIA did. They demoed some AI working on the GPU around the 4xxx series, some GPU physics and was talk about working with Havok for it, but nothing rather impressive. A lot of games were with "The way it's meant to be played" logo which helps quite a lot.

Having the better product only helps if you know how to market/sell it! It was the same with HD7000/R290 series.

And this is why OEMs, both desktop and notebook are being targeted first by AMD. Nvidia is going first for the person who buys discrete. Business to business might have no glamour, as it appears the world is hooked on, but it sure can be profitable.

 

[Head1985]

It's not that simple. The reviewers will make a huge deal of gaming the system. They will not take it. Heck, I wouldn't like cherry-picked reviews, either.

However, Nvidia, on the other hand, is much better at this. They know how to butter up their reviewers. This past event in Texas was a prime example of clouding their heads. Nvidia threw everything at those "reviewers". They had horseback riding, river wrathing, gambling, DJs, clubing, alcohol, fancy food, etc. Heck, Nvidia even bought out an entire theater so the reviewers can get a special viewing of Captain America: The Civil War. Everything was paid for!! They really butter those reviewers up real good.

It worked. Instead of asking Nvidia the real hard questions. They ate up whatever Nvidia feed them.

Founders Edition? Nice. Pretty cooler.
2X performance of GTX 980SLI? WOW!!!
2x performance, 3x power efficiency of a Titan X? **Head explodes**.

Nevermind the fact that everything Nvidia said were heavily cherry-picked.

Nvidia is the master of marketing. It just makes it a lot easier for reviewers to side with Nvidia. I totally understand why "reviewers" are more forgiving of Nvidia. It makes total sense.

In other words Nvidia BRIBE them.

 

[Sweepr]

VR World rumor says that Polaris 10 will fit in both 480 and 490 series

VR World's Pascal rumours were utterly wrong. Bench-life says Polaris 10 will is R9 480/480X, and AMD itself calls it desktop mainstream.

with Tweaktown saying non-x 480 packs already 390x performance.

Based on the GFXBench comparison I posted in this forum (and mentioned it's not the fastest SKU), and no, that particular SKU is not close to R9 390X at all.

Benchlife says it is a ~115W ACP card, Wccf says it performs like a TitanX on FStrike at far less than the 170W card TDP. And the rumor mill says it packs perf/mm2 and perf/w lead.

WCCFTech got that info from a Chiphell forum thread, and the same user/leaker said Ellesmere XT is expected to almost match R9 390X.

So is not too much optimism to expect Polaris 10 beatimg GTX 1070.

We'll see if it delivers on the expectations of equivalent/faster than Fury X and 90% GTX 1080 performance at half the price.

 

[xpea]

From the leaked 3dmark benchmark, a stock GTX1080 was boosting upward of 1866mhz. 2000mhz isn't that much more than a stock GTX1080. The demo showed it overclocked to 2114mhz. That's only 13% from the stock boost speed.

Honestly, for it to be a worth upgrade to the GTX980TI, it would need to be able to clock around 2.3-2.4ghz. Nvidia told a lot of reviewers after the event that the 1080 is roughly 25% faster than a GTX980TI. That means a 1080 is as fast as fully overclocked GTX980TI (1500mhz). A 1080 @ 2.3-2.4ghz would make it roughly 20-25% faster than a 1500mhz GTX980TI. is it enough to warrant an upgrade? Maybe....

I can't believe that people is building argument based on this 5 years old theoretical benchmark that is useless to measure performance under new 2016 games
under DX12, at same TFLOPS, Pascal will show around 10% better performance due to the uarch improvement and even more with driver optimization that Maxwell won't benefit. When taking into account the >2GHz Pascal clock, in games like Quantum break and Ashes, Pascal 1080 will obliterate GTX980Ti (at least 30% difference). In VR, the lead will be even higher...

 

[guskline]

Honestly, the price of the GTX 1080 accelerated the loss of market value of my GTX 980 TI SC in rig 1 below IF I was going to sell it now to replace it with a 1080. So I guess I would be "depressed" if I was planning to do that.

My plan all along is to replace my 2 R9 290s in CF with a single gpu that significantly outperforms them, preferably an AMD chip ( I like to have each rig with competitor gpu).

Nvidia is doing what they do best. Marketing and maximizing profits. Can't blame them.

I wnt to see the comparison of 2 R9 290s in CF in DX12 games to a single GTX 1080. The 1080 will sip power compared to them but I custom water cool so sheer performance of a single card to 2 in CF is my standard.

I'll probably have to wait for Vega.

 

[airfathaaaaa]

I can't believe that people is building argument based on this 5 years old theoretical benchmark that is useless to measure performance under new 2016 games
under DX12, at same TFLOPS, Pascal will show around 10% better performance due to the uarch improvement and even more with driver optimization that Maxwell won't benefit. When taking into account the >2GHz Pascal clock, in games like Quantum break and Ashes, Pascal 1080 will obliterate GTX980Ti (at least 30% difference). In VR, the lead will be even higher...

then why we didnt saw ANY dx12 games there? only doom with unlocked fps while the open beta is locked on 60 fps?
surely they could have used aots or actually any dx12 that favors them
but no nothing of this...infact we didnt saw anything from dx12 only from vulkan and surely coded on nvidia path

 

[Silverforce11]

then why we didnt saw ANY dx12 games there? only doom with unlocked fps while the open beta is locked on 60 fps?
surely they could have used aots or actually any dx12 that favors them
but no nothing of this...infact we didnt saw anything from dx12 only from vulkan and surely coded on nvidia path

They did, their biggest gains in a game was Rise of Tomb Raider DX12. Note they did not show DX11.

They will have bigger gains in Quantum Break over the 980/Ti.

 

[iiiankiii]

I can't believe that people is building argument based on this 5 years old theoretical benchmark that is useless to measure performance under new 2016 games
under DX12, at same TFLOPS, Pascal will show around 10% better performance due to the uarch improvement and even more with driver optimization that Maxwell won't benefit. When taking into account the >2GHz Pascal clock, in games like Quantum break and Ashes, Pascal 1080 will obliterate GTX980Ti (at least 30% difference). In VR, the lead will be even higher...

1st off, I never said I was using benchmarks from 3dmark. I said the CLOCK speed was 1866mhz from the benchmark.

2nd, NVIDIA was the one that told the press afterwards that the 1080 will be around 25% faster than a REFERENCE GTX980TI. Nvidia said that, not me.

As for DX12, Pascal is expected to do well compared to maxwell because of instant graphic <-> compute switching. I agree with you there. How much of an improvement? Who knows. No hints were given.

edit: nvm. Nvidia showed Tomb Raider. It should 75% increase over a GTX980 in dx12. But, didn't DX12 perform worse than DX11 in Tomb Raider?

 

[Timmah!]

Do you think that possible non-reference 1080 with 2 8-pins instead of just one (or 8+6) could be OCeable well beyond 2GHz, if its apparently possible to run it at 2100+ with just single 8-pin board?

Cant wait for the reviews now. If with the OC it can go up to 50 percent above stock 980Ti in my app of choice, i will probably go for it, even if some things are bit underwhelming (especially when we know P100 properties). Last year i was thinking about getting 980Ti Hybrid from Evga, similar solution for 1080 would be very nice now - not for higher price, though.

 

[airfathaaaaa]

They did, their biggest gains in a game was Rise of Tomb Raider DX12. Note they did not show DX11.

They will have bigger gains in Quantum Break over the 980/Ti.

you saw a live demo of rotr? also if what ppl on beyond3d is saying is true then it surely not even close to x2 even if the rotr was on 80% they say 1.6 for rotr and 1.54 for tw3...

 

[imaheadcase]

The only concerns I have are..is the 1080 going to last 6+ months before a new card is %10+ better?

Historically, what % increase can drivers give new card architecture in performance?

Basically is my 780 TI worth upgrading right now to this 1080(assuming i can get on day 1), or is the 1080TI coming in less than 6 months?

 

[Mopetar]

The only concerns I have are..is the 1080 going to last 6+ months before a new card is %10+ better?

Historically, what % increase can drivers give new card architecture in performance?

Basically is my 780 TI worth upgrading right now to this 1080(assuming i can get on day 1), or is the 1080TI coming in less than 6 months?

Doubtful. Any GP100 chips that manage to come out whole are being sold as $12,000 Teslas for HPC. A chip that big on a node this immature is going to have terrible yields, but when you can sell a slightly cut chip for that much it doesn't matter. Unless they have another bin for chips that can't make the P100 cut, but are still mostly functional, I wouldn't count on it.

Given the performance increase potential over the 1080, NV would piss off a lot of people who paid $700 unless they sold the GP100 chip for $1,500 or something obscene like that.

 

[tential]

Doubtful. Any GP100 chips that manage to come out whole are being sold as $12,000 Teslas for HPC. A chip that big on a node this immature is going to have terrible yields, but when you can sell a slightly cut chip for that much it doesn't matter. Unless they have another bin for chips that can't make the P100 cut, but are still mostly functional, I wouldn't count on it.

Given the performance increase potential over the 1080, NV would piss off a lot of people who paid $700 unless they sold the GP100 chip for $1,500 or something obscene like that.

Honestly, the posts in here make it seem like you guys have 0 clue or idea about ANYTHING Nvidia does.
The whole "This isn't price efficient enough!!!" or "Nvidia will piss off their customers!!!!"
GET OUT
Seriously, if those are your posts, just GET OUT of the discussion, because you clearly have no idea what's going on if you think Nvidia will lose market share over that.

Nvidia has done this for 3 generations STRAIGHT and you think NOW it won't work?

People will HAPPILY buy the 1080, then upgrade to the 1080ti, just like the 980Ti. And that was even worse as the Titan X was severely undercut by the 980Ti.

I get some of you are severely price/performance limited in your purchases.

But in the world of most consumers, they aren't. They purchase based on how a product makes them feel, and a 1070 makes them feel like they got a Titan X for $380-450.

P10 makes them feel like they got a 390x at a good price/perf/watt. The fact that it's already being compared to the 390x is a loss.

But hey, you're welcome to be hopeful and many of the AMD hopeful are now predict even higher performance and even lower prices.

But I'm not naive. The 1070 will dominate, just like the 970.


Next time ask the Mods to take care of it. You're not a mod.

-Rvenger

 

[tential]

The only concerns I have are..is the 1080 going to last 6+ months before a new card is %10+ better?

Historically, what % increase can drivers give new card architecture in performance?

Basically is my 780 TI worth upgrading right now to this 1080(assuming i can get on day 1), or is the 1080TI coming in less than 6 months?

If you hold onto your card for that long, then you really have no right buying Nvidia products. You just are pledging yourself to horrendous performance for long periods of time. Get the 1080Ti if you want, but if you're going to wait 2 generations inbetween high end Nvidia purchases, you're getting outperformed by cards half the price, regularly, and that's highly disappointing.

 

[AtenRa]

But I'm not naive. The 1070 will dominate, just like the 970.

970 dominated because there was not an alternative from AMD at the time. This time we will have Polaris 10, so its not the same as before. Im not saying 1070 will not sell well, but i really have my doubts it will dominate like 970.

 

[airfathaaaaa]

970 dominated because there was not an alternative from AMD at the time. This time we will have Polaris 10, so its not the same as before. Im not saying 1070 will not sell well, but i really have my doubts it will dominate like 970.

we just have to wait and see the q1 results
or wait for the steam survery :sneaky:

 

[exar333]

970 dominated because there was not an alternative from AMD at the time. This time we will have Polaris 10, so its not the same as before. Im not saying 1070 will not sell well, but i really have my doubts it will dominate like 970.

Polaris is releasing in June?

 

[AtenRa]

Polaris is releasing in June?

I believe early July.

 

[sxr7171]

Funny how he showed the 1080 GTX overclocked beyond the boost clock yet all comparisons are vs a reference 980 or 980ti.

I need to know how much a fully o/c 1080 GTX performs vs a Titan X at 1453MHz.

These comparisons which are not like for like are not helpful.

 

[airfathaaaaa]

http://www.ebay.com/itm/EVGA-NVIDIA...=item3f628e66c5:g:2NoAAOSwubRXLpk2&rmvSB=true

prices are droping like a stone..

 

[DooKey]

Funny how he showed the 1080 GTX overclocked beyond the boost clock yet all comparisons are vs a reference 980 or 980ti.

I need to know how much a fully o/c 1080 GTX performs vs a Titan X at 1453MHz.

These comparisons which are not like for like are not helpful.

I suspect you'll see reviews on the 27th that will OC the heck out of 1080 and compare them to last gen cards that are OC'd.

 

[airfathaaaaa]

I suspect you'll see reviews on the 27th that will OC the heck out of 1080 and compare them to last gen cards that are OC'd.

Pretty sure they will instruct the reviewers to compare then with a refrence card.....