Nvidia CEO says Radeon VII is "underwhelming"

https://www.google.com/amp/s/www.techradar.com/amp/news/amds-radeon-vii-graphics-card-dismissed-as-underwhelming-by-nvidia

Nvidia CEO has said the performance is lousy and brings nothing new, is he right? This might be the first time in a long time that I agree with Nvidia but he is right. It is a glorified GTX 1080 ti but 2 years late. It's priced too high and AMD should no better. For $700 like the 2080 but has no RT or Tensor cores. Its manufactured on a state of the art 7nm process but still absolutely power hungry and it seems AMD still can't figure out how to get past the 64CU wall with GCN. Even though Nvidia is a greedy company with overpriced products, I expect when they move to 7nm geforce the results will be very impressive.

Thoughts?

That's gonna be one awkward family reunion.

Eh, just more trash talking. People who watched the industry for a long time know that there is real bad blood between NVIDIA vs ATI/AMD. Marvel vs DC, this is not. Just look at the history of the things they tried to do to shut each other out.

@madrocketeer: that's weird that there us such bad blood even though the CEO's are related.

Sometimes family can be worst.

I really hope there are lawyers and investigators looking into this family. The shit don't seem right...

Well the VII looks like an awesome card, but not for the price. Even still, I can't imagine 16GB of HBM not making the price insane.

Didn't expect him to view it positively but he is technically right in that it doesn't bring anything new.

Well, apparently Lisa Su denies it. Either way, the real rivalry is at the engineers level. And since ATI was said to be an engineers' company...

pretty much. It looks like a great GPU ( RTX 2080 performance ) but imo considering AMD products in the mid range are known for their value this GPU needed to be about 100 less. hopefully they will release an 8 or 12 gb version for around 500

Would have been great at 600$. Not at 700$.

People is less concern with tech so long as it does what it need. Like phones. Phone sales is way down. Apple is dumb to think iPhone sales to increase if prior model they sell for $1000. No way is general consumer see the needs to upgrade to same content for $1200!

Same is going for computers.

it doesn't look that interesting for gaming. by the looks of it its just a vega 60 with a high clock on a 7nm process. its not terrible or anything just...bleh. its probably a compute beast though...an absolute monster in that department.

i think this is just biding time for navi to be honest but i wouldnt expect to see that until after either the PS5 or X2 release (all sorts of rumours around that).

they need to improve their efficiency. they need to take a vega 64 and, without a die shrink, an increase in clock, an increase in power usage or an increase in stream processors, find 30% more performance (or as much as they can) in the most demanding games today. whatever nvidia did with maxwell...its bloody effective.

Being 7nm card he ain't wrong

Going pass 64 CU wouldn't solve the problem since Hawaii era quad rasterization setup is a bottleneck. AMD can easily add more CUs.

This Hawaii GPU's four Shader Engine layout was used from Hawaii 44 to Vega 64 and AMD added 20 CUs without increasing horizontal rasterization power i.e. more rasterizer units.

Vega 56 at 1710 Mhz with 12 TFLOPS beats Vega 64 with 1590Mhz at 13 TFLOPS. Hint: Vega 56 at higher clock speed has higher rasterization power.

Vega 64's quad rasterizer units with 64 ROPS design is like GP104 (quad GPC with 64 ROPS) class Pascal design which bottlenecks extra TFLOPS.

VII has proper support for 64bit floating point which is extra hardware over Vega 64.

Vega 64's high bandwidth cache is useless for gaming.

VII's proper support for 64bit floating point is useless for gaming.

I'm just not confident in their 5th gen GCN architecture. I think if they weren't using HBM, their power draw and heat would be completely unsuitable for any use.

They keep using these mammoth dies and praying that the node shrink will be enough to make it a usable card.

Wasn't this unconfirmed though?

On topic: Jensen may just be salty as Lisa wore that leather jacket attire better than Jensen did.

RTX 2080's chip area size is 545 mm2 at 12 nm process

VII's chip area size is 331 mm2 at 7 nm process

NVIDIA has the larger dies.

For example

RTX 2080 Ti's chip size is 754 mm2 at 12 nm process

RTX 2080's chip area size is 545 mm2 at 12 nm process. Garbage support for FP64.

VII's chip area size is 331 mm2 at 7 nm process. Half rate FP64 i.e. 6.9 TFLOPS.

nvidia owned amd again

Actually, AMD wins with superior "die per wafer" with the smaller 331 mm2 GPU.

But I was told AMD was going to save us all from NVIDIA!

he also talked trash about intels upcoming graphics cards at the end. Sounds like hes shtting his pants

If it was 500 dollars or less, it wouldn't be underwhelming at all. But for 700 dollars? Nah.

Intel's gfx gen 11 is a tile based render similar to Maxwell/Pascal/Turing ( immediate mode tile cache render) or PowerVR (deferred mode tile render). Intel sells more 2-in-1 tablets when compared to Nintendo's Switch or NVIDIA Tegra X1.

He's not wrong, just sounds butthurt.

Radeon VII is not what (or many other people) I expected, but Jensen Huang should not talk such trash. It's still 2080 performance for 100$ less. Sure, there's no raytracing or DLSS, but those are gimmicks for now that are used in like two games. So TBH the Radeon VII is a good competitor in the med/high-end segment because as of now raytracing and DLSS aren't ready for prime time & for all games.

I don’t like Nvidia and their shady business practices, but with that Radeon VII ridiculous price tag is even less appealing and I'll still use Nvidia GPU only because they have what I really need and I ain't got nowhere to look but Nvidia so can you blame me? Sink or Swim, pick one.

DLSS is just sample based pixel reconstruction (more "compressed pixels" from PC NVIDIA master race) method while VII has 128 ROPS and 1TB/s for improved MSAA processing. In the future, AMD's sponsored games could be bias towards maxed MSAA at 4K.

VII is still a reasonable GPU since it has RTX 2080 level performance.

Vega 56 Mod at 1710 Mhz already shown the way for VII.

You fukoff

This entire generation of graphic cards has been very underwhelming and ridiculous priced...

@ronvalencia: Nobody cares. Shut up.

@Juub1990: I stop paying attention to Ronbot after his false on Xbox One X specs.

I remember the nonsensical bloated posts with Ron arguing with me that the Xbox One X would have Zen CPUs and not Jaguar. Those were the days.

Zero market for this.

Should have released it for 399 and maybe people cared. for 700 just buy nvidia no need to bother with the shit tier driver support of AMD.

Nvidia aint wrong, they didn't even try.

So much wrong in your post.

They'd lose so much money at 399. 600 would have been a better sweet spot. They'd corner the mid-grade GPU market if they lowered the price just a tad bit.

OMG Nvidia claims AMD sucks now i know is true.

@organic_machine: i think vega was certainly a missfire from a gaming performance point of view (it is a compute beast and its telling that the biggest leap from vega to vega2 is in compute according to AMDs graphs). there are all sorts of stories around its development...though they are only rumours. apparently it wasnt smooth.

did they get the tile based rendering to work in the end? i heard that was also borked at launch.

but yeah they are heavily relying on die shrinks to just eek out any performance they can. it's not sustainable.

hopefully navi is back to the drawing board a bit. they must know why they are losing so much performance and it's just a case of the fix being expensive and difficult to implement.

Due to 16GB, it's price $700 because of it. $600 would have indeed be a much better spot but $700 for an AMD GPU is really pushing it and I can get a better cheaper GPU then that. AMD sure wasn't even trying but I do understand this is just for content creation GPU due to 16GB. At least AMD is doing a much better job with the CPU though.

@davillain-: Lol with Xbox One X with the power to rival a 1080 Ti.

How long has it been since AMD did anything amazing? lulz.

they're just always playing catch up, why Nvidia can release bullshit like RTX and nap.

@dxmcat: Ryzen has been pretty amazing.

He's just shit scared as AMD's new cards are going to trash RTX......... Nvidia wasted their transistor budget on RT circuitry for 1 game which is why a 2080ti performance jump isn't as big as it should of been.

AMD are building a monster and using GPUCompute for their RT which is what DirectXR uses so they can spend more transistors on beefing up every aspect of the GPU which will deliver massive boosts in every game.

1. I was correct on X1X GPU's near Vega ROPS stupid bot. Tonga IP and Polaris IP has horizontal FP16 instructions not Vega IP's vertical FP16 instructions. Furthermore, XBO GPU has additional support for FP10 formats.

2. You can't read my IF statements.

2 games that I want. Yes! ;)

You can't read my IF statement stupid bot and that's just subsystem comparison with an IF statement NOT game comparison when I create GTX 980 Ti vs PS4 Pro's Mantis Burn Racing with RPM topic.

Furthermore, Pascal has problems with mixing integer and floating point.

https://devblogs.nvidia.com/nvidia-turing-architecture-in-depth/

Turing Tensor Cores add new INT8 and INT4 precision modes for inferencing workloads that can tolerate quantization and don’t require FP16 precision. Turing Tensor Cores bring new deep learning- based AI capabilities to GeForce gaming PCs and Quadro-based workstations for the first time. A new technique called Deep Learning Super Sampling (DLSS) is powered by Tensor Cores. DLSS leverages a deep neural network to extract multidimensional features of the rendered scene and intelligently combine details from multiple frames to construct a high-quality final image

VII has added support for INT8 and INT4 for inferencing workloads within it's CU. VII has higher memory bandwidth and 128 ROPS(Anandtech claim) for heavy MSAA at 4K which reduce the need for DLSS based pixel reconstruction AA

Vega 64 has support for INT8 for inferencing workloads within it's CU.

----------

RT Cores accelerate Bounding Volume Hierarchy (BVH) traversal and ray/triangle intersection testing (ray casting) functions.

https://32ipi028l5q82yhj72224m8j-wpengine.netdna-ssl.com/wp-content/uploads/2016/08/169798-A_AMD_RadeonRays_Intro_FNL.pdf

Acceleration strategies The default acceleration structure is Bounding Volume Hierarchy (BVH) using spatial median splits. It maintains fast build times and provides decent intersection performance. You can enable SAH builder using the global option and trade off construction time for better intersection performance

AMD is aware of BVH issue.

https://worldoffries.wordpress.com/tag/deferred-shading/

BVH code on R9-280X era.

https://wccftech.com/titan-v-high-fps-battlefield-v-rtx/

Enabling RT function RTX coreless on Titan V. Artificial software block on older GPUs.

https://devblogs.nvidia.com/nvidia-turing-architecture-in-depth/

Turing introduces a new processor architecture, the Turing SM, that delivers a dramatic boost in shading efficiency, achieving 50% improvement in delivered performance per CUDA Core compared to the Pascal generation. These improvements are enabled by two key architectural changes. First, the Turing SM adds a new independent integer datapath that can execute instructions concurrently with the floating-point math datapath. In previous generations, executing these instructions would have blocked floating-point instructions from issuing

Pascal architecture landmine with integer instructions blocking floating point instruction issuing. This Pascal landmine doesn't occur on Vega.

There's are reasons for Nvdia's Gameworks existence to minimize landmines like this. Deep learning integer workloads would be landmines on gaming Pascal GPUs.

Try again clown.

@ronvalencia: