[Rumor, Tweaktown] AMD to launch next-gen Navi graphics cards at E3

[JDG1980]

Also, I noticed no mention of video decoding - is Navi still missing full hardware decoding for VP9?

 

[JasonLD]

The cards look okay for what they are. The problem is that AMD has stranded the sub-$300 market completely. Now it's 1660Ti or bust. AMD won't save you, midrange gamer.

Isn't there still a lot of inventory leftover from mining bust? I think they probably want to clear those first for sub $200-300 markets.

 

[Glo.]

Since when? That isn't "price market leadership", whatever that really is.

5700 vanilla is already $50 more than its direct competitor, and NV hasn't even cut prices yet. I'll believe "90% of 2070" when it shows up in the benchmarks. AMD didn't even make that comparison.

Radeon 5700 is not RTX 2060 competitor...

Its much faster GPU...

 

[JDG1980]

Radeon 5700 is not RTX 2060 competitor...

Its much faster GPU...

I'll bet money it will be within 5-10% of 2060 in unbiased third-party rankings (e.g. TPU).

 

[Paul98]

Also, I noticed no mention of video decoding - is Navi still missing full hardware decoding for VP9?

They mentioned something about 4k encoding and decoding

 

[Stuka87]

I'll bet money it will be within 5-10% of 2060 in unbiased third-party rankings (e.g. TPU).

TPU is known for using games that skew results.

I would be surprised if the 5700 is somehow way slower than what they showed.

 

[crisium]

I think they need a more cut down card to try to get the masses to try Navi. That with some more separation would still get big margins on the full-fat chip, but get the masses to adopt more.

$450 2560 Shaders
$360 2304 Shaders
$280 2048 Shaders, 192-bit

Go after that 1660 Ti and kill it. I think the market would respond better to this product pricing and offering. I'm shocked they are ignoring the $300 and under market with a 250mm2 chip, even just a heavily cut one.

 

[DrMrLordX]

Isn't there still a lot of inventory leftover from mining bust? I think they probably want to clear those first for sub $200-300 markets.

Not really. Case in point:

https://pcpartpicker.com/products/v...92,391,404,405&sort=price&page=1&P=8589934592

Here's every 8GB AMD mining card from 470 on up to RX Vega 64 (not counting VegaFE, and not counting RX 590 since it's too new to have been used in the mining rush). Outside of that one horrible Gigabyte Vega56 and one stray MSI Vega56, everything that can hang with a 1660Ti (or anywhere near it) is above the $300 mark. 580s and weaker are as cheap as $160. That's a huge price hole. You could drive a truck through that. I'm sure there's a market for a $250-$300 Navi that beats a Vega56 and 1660Ti. That's really where the 5700 should have been, not . . . $379.

Excess mining stock hurt NV more, since people were mining with 1080s and (in some cases) even 1080Tis.

Radeon 5700 is not RTX 2060 competitor...

Did you watch the presentation? AMD compared 5700 directly to 2060. It was at best 10% better across the board on their slides.

Go after that 1660 Ti and kill it.

Exactly! Maybe they didn't have to kill the 1660Ti and 2060 all at once, but the 5700 could have done it. 1660Ti sells for $260 at the low-end right now, and they're all 6GB cards to boot.

 

[JDG1980]

"1 x 8-pin & 1 x 6-pin"
The PCIe slot is good for 75W, the 6-pin connector is good for another 75W, and the 8-pin connector chips in up to 150W. That means we are looking at a TDP of as much as 300 watts on a Pitcairn-sized chip. Looks like all the perf/watt talk was just hot air. Consider that RTX 2070 tops out at a bit over 200W - despite being a full node behind Navi.

AMD needs to hire some GPU engineers who actually know what the hell they're doing. And they need to throw GCN in the trash, since it's clear that as long as any trace of it remains, they will never have any chance of being competitive.

 

[soresu]

That price point is disgusting. It should be a $329 card at most, given the lack of RT/DLSS, proper OpenGL support, and sky-high TDP.

Proper OpenGL support?

AMD's OpenGL performance is admittedly not so great, but it does support the latest version of OpenGL AFAIK.

I suspect that AMD's more compute focused uArch was designed to be more flexible/versatile than nVidia's - at the cost of making it more difficult to implement a high level API efficiently on it over a low level API, hence why DX12 and Vulkan runs better for AMD, while OpenGL and DX11 doesn't run so great.

 

[linkgoron]

What did you expected? That they will undercut Nvidia massively and hurt themselves at the same time?

Again. What did you guys expected?

I was hoping for something relatively good at the ~$300 level, which hasn't seen a real performance increase in YEARS. It's still at ~1070 performance (released more than three years ago now)! The $300 range has gone up by "one" tier in 3 years (1060->1070), and that's just horrible.

 

[JDG1980]

Proper OpenGL support?

CEMU and pcsx2 can't run worth a damn on AMD cards. Their developers have said this is because AMD's OpenGL support is completely broken.

 

[Glo.]

We still haven't seen any power draw numbers, and everybody already have said it is high. Interesting.

 

[soresu]

CEMU and pcsx2 can't run worth a damn on AMD cards. Their developers have said this is because AMD's OpenGL support is completely broken.

I have seen these comments before, I do wonder why the AMD Windows team don't take a feather out of the Linux open driver book and learn from them, their driver has massively improved in the last 2-3 years.

Having said that, the Dolphin coders seem much more on the ball about the AMD driver problems - I've seen certain comments that indicate CEMU development focus has been towards performance first, and stability later, likewise PCSX2 has lagged for some time in significant development, one of its best coders refractionpcsx2 has moved over to another PS2 emulator called DobieStation.

 

[JDG1980]

They mentioned something about 4k encoding and decoding

Even GCN 1.0's UVD can do 4K encoding/decoding with H.264, VC-1, and MPEG-2. They added HEVC on 14nm, but Raven Ridge is the only AMD product so far that can do full hardware decoding of VP9.

 

[DrMrLordX]

We still haven't seen any power draw numbers, and everybody already have said it is high. Interesting.

The 1x6 + 1x8 pin is just for overclocking, I think. Stock I don't see it taking up that much juice.

 

[Glo.]

The 1x6 + 1x8 pin is just for overclocking, I think. Stock I don't see it taking up that much juice.

Exactly my thoughts.

 

[JDG1980]

We still haven't seen any power draw numbers, and everybody already have said it is high. Interesting.

If it was low enough to be competitive with the 2070, they would have said so. They remained silent on this matter, therefore we can reasonably assume it sucks. And if it was under 225W, then why would they have a 8-pin *and* a 6-pin? The 8-pin alone would suffice for that.

The truth is that if the architecture was as good as Nvidia's, perf/watt should have blown Nvidia out of the water due to the node advantage.

 

[railven]

Welps, nothing really worth reading over. Reception seems bad, but eh what you expect?

Now wait for July 7th. I hope they got acoustics down, feel like AMD's been recycling the same blower/shroud design since 6K series!

 

[crisium]

AMD CEO Edition?

 

[Glo.]

Haha, AMD's Founders Edition.

 

[railven]

AMD CEO Edition?

Well she's now sporting a leather jacket! Anandtech's own recap oddly pointed this out haha. (also some interesting commentary, feel like they weren't impressed just from reading their live blog thingy).

 

[DrMrLordX]

I hope they got acoustics down, feel like AMD's been recycling the same blower/shroud design since 6K series!

I"M SORRY I CAN'T HEAR YOU OVER THE FANS

 

[Glo.]

If it was low enough to be competitive with the 2070, they would have said so.

That is only your assumption, not a fact.

 

[railven]

I"M SORRY I CAN'T HEAR YOU OVER THE FANS

Will always remember (and chuckle) at this: