Prepare your wallets - GTX 1080 incoming

[Nozzer66]

Nothing wrong with getting info from LinusTechTips. He knows his stuff.

[Spock]

I'll wait to RMA my 780 (it's sort of stable now, underclocked) till they come out then and hope they don't have any left in stock to replace it with and upgrade to 980 after I send a nastygram. Fourth RMA since I got a 780 which is the most I've ever had to rma a graphics card lol. Course I was supremely peeved they released the 900 series with like no warning a couple weeks after I bought my 980 #sad

 

If you can get most of your money back, RMAing that card is a really good idea. Kepler has major weaknesses in it's shader architecture. Once it flew out of the current driver optimizations for NVidia and shaders in games advanced to the next level, it's performance broke down more then one price category (and NVidia cards are pricey to begin with).

 

The 1080 is very interesting because it might get way better performance for stuff like NNEDI3. It also has to be seen if it fixes the async compute problems Maxwell has. Maybe we can finally get an NVidia card with decent openCL compute performance.

[mikegray]

Nothing wrong with getting info from LinusTechTips. He knows his stuff.

I agree completely. It's just that ...well ... I'm not Linus!

[hellanios]

It's too early to pick. We haven't even seen AMD's cards yet. To me it's all about perf/price. BTW, when benchmarks start rolling out for the new gen, make sure to check out this guy's channel.

Edited May 9, 2016 by hellanios

[EssArrBee]

I was just reading the article Anandtech posted and it seems the 1070 is getting regular gddr5. Also, the new cooler they designed is only on the founder's edition. Wouldn't be nVidia if they didn't add some extra fee somewhere.

 

One other thing the presentation didn't mention was a new generation of SLI that will require a new bridge. Apparently they needed something for the higher bandwidth. That should lead to a nice boost for SLI compute efficiency.

[hishutup]

I was just reading the article Anandtech posted and it seems the 1070 is getting regular gddr5. Also, the new cooler they designed is only on the founder's edition. Wouldn't be nVidia if they didn't add some extra fee somewhere.

But who REALLY wants the reference cooler? The only scenario I can see is in one of those ultra slim cases.

One other thing the presentation didn't mention was a new generation of SLI that will require a new bridge. Apparently they needed something for the higher bandwidth. That should lead to a nice boost for SLI compute efficiency.

If they needed more bandwidth then why didnt they use the other pcie lanes that are never used. I wouldnt think the GPU needs anymore than 4x 3.0 lanes.

This is one of those things that AMD got right.

[EssArrBee]

But who REALLY wants the reference cooler? The only scenario I can see is in one of those ultra slim cases.

 The reference cooler is on the regular model. There is a new cooler they had on the founder's edition that kept the GPU at 67C with the core clock over 2K during the presentation. That's pretty insane. 

If they needed more bandwidth then why didnt they use the other pcie lanes that are never used. I wouldnt think the GPU needs anymore than 4x 3.0 lanes.

This is one of those things that AMD got right.

What? The bridges connect to the outside of the cards, not on the board for both nVidia and AMD. They usually go along the outside so they don't mess with anything else.

[hishutup]

 The reference cooler is on the regular model. There is a new cooler they had on the founder's edition that kept the GPU at 67C with the core clock over 2K during the presentation. That's pretty insane. 

Its still a blower which means it needs to spin all the time and at a pretty high rpm

The current card I have is silent even when playing FNV.

 

 

 

 

What? The bridges connect to the outside of the cards, not on the board for both nVidia and AMD. They usually go along the outside so they don't mess with anything else.

AMD doesnt use bridges anymore. why does nVidia still use them when the pcie bus is faster?

[DoubleYou]

Founders Edition = Reference Card

 

They just gave it a new name to get everybody confused.

[EssArrBee]

Guess I haven't paid attention to AMD in so long I didn't even notice they didn't use a bridge anymore. The bridge is used for rendered pixel data to sync timing between both cards and doesn't need a ton of bandwidth. At least it didn't in the past, but now that everyone is moving to higher resolutions, so the move to a 2GB/s bridge is probably necessary. Most the other data is still sent through the PCIe lanes.

 

My guess is that AMD ditched the bridge since PCIe latency has really dropped from 2.0 to 3.0, so it doesn't matter if you use a bridge or not. I'd bet they have some built in tech to handle GPU concurrency in CF setups if they ditched the bridge and nVidia doesn't have their tech developed yet. 

 

The only advantage then of the bridge is if you need to shave time off the data transfer, like fractions of a nanosecond over the PCIe bus. It could just go from card to card instead of from one GPU to the CPU to the other GPU. Those hardware guys can get pretty crazy about that stuff, so I wouldn't put it pass them. 

[hellanios]

[darkside]

 

Very nice analysis

[mikegray]

[....]

Nice info. But FAR more important:

 

Is there any chance I could get that guy to record my own custom-voiced Dragonborn? Add in a 4k Immersive Kilts and Cabers Overhaul, and I'll have the playthrough of the century ...

[hellanios]

[TechAngel85]

Skip to around 7:50 to pass all the stuff we've already heard 100 times by now: