Slashdot Mirror
Nvidia CEO Trashes AMD's New GPU: 'The Performance Is Lousy' (gizmodo.com)
An anonymous reader shares a report: Yesterday, AMD announced a new graphics card, the $700 Radeon VII, based on its second-generation Vega architecture. The GPU is the first one available to consumers based on the 7nm process. Smaller processes tend to be faster and more energy efficient, which means it could theoretically be faster than GPUs with larger processes, like the first generation Vega GPU (14nm) or Nvidia's RTX 20-series (12nm). I say "could," because so far Nvidia's RTX 20-series has been speedy in our benchmarks. From the $1,000+ 2080 Ti down to $350 2060 announced Sunday, support ray tracing. This complex technology allows you to trace a point of light from a source to a surface in a digital environment. What it means in practice is video games with hyperrealistic reflections and shadows.
It's impressive technology, and Nvidia has touted it as the primary reason to upgrade from previous generation GPUs. AMD's GPUs, notably, do not support it. And at a round table Gizmodo attended with Nvidia CEO Jensen Huang he jokingly dismissed AMD's Tuesday announcement, claiming the announcement itself was "underwhelming" and that his company's 2080 would "crush" the Radeon VII in benchmarks. "The performance is lousy," he said of the rival product. When asked to comment about these slights, AMD CEO Lisa Su told a collection of reporters, "I would probably suggest he hasn't seen it." When pressed about his comments, especially his touting of ray tracing she said, "I'm not gonna get into it tit for tat that's just not my style."
It's impressive technology, and Nvidia has touted it as the primary reason to upgrade from previous generation GPUs. AMD's GPUs, notably, do not support it. And at a round table Gizmodo attended with Nvidia CEO Jensen Huang he jokingly dismissed AMD's Tuesday announcement, claiming the announcement itself was "underwhelming" and that his company's 2080 would "crush" the Radeon VII in benchmarks. "The performance is lousy," he said of the rival product. When asked to comment about these slights, AMD CEO Lisa Su told a collection of reporters, "I would probably suggest he hasn't seen it." When pressed about his comments, especially his touting of ray tracing she said, "I'm not gonna get into it tit for tat that's just not my style."
Or is hearsay and copy-paste shit from the internet count as an article link now?
/. was getting bad, this is a hot garbage example of that.
I knew
This summary makes little sense.
I'm not planning on upgrading my video card until I can get 4K video running at 60 FPS for new games. Neither nVidia nor AMD can pull that off, with nVidia coming the closes in its video cards priced above $1000.
Maybe the next video card generation...
So no article to read, or link to
No benchmark
Summary can't decide if Radeon VII is 7nm or not. At the least not from first glance
WTF anon
When the competition bashes your products, you know they are scared. AMD has a winner GPU.
"Yesterday, AMD announced a new graphics card, the $700 Radeon VII, based on its second-generation Vega architecture. The GPU is the first one available to consumers based on the 7nm process. It's impressive technology, and Nvidia has touted it as the primary reason to upgrade from previous generation GPUs. AMD's GPUs, notably, do not support it."
So AMD made a GPU that NVIDIA thinks is the primary reason to upgrade, yet AMD doesn't support it. DOES ANYONE ACTUALLY READ WHAT THEY APPROVE TO BE POSTED?
The GPU is the first one available to consumers based on the 7nm process. It's impressive technology, and Nvidia has touted it as the primary reason to upgrade from previous generation GPUs. AMD's GPUs, notably, do not support it.
This makes no sense until you realize later in the summary that they're talking about ray tracing, not 7nm manufacturing
"goodbye and hello, as always" ~Prince Corwin, from Zelazny's Amber series
The article contradicts itself... "the $700 Radeon VII, based on its second-generation Vega architecture. The GPU is the first one available to consumers based on the 7nm process"... "Nvidia has touted it as the primary reason to upgrade from previous generation GPUs" ... "AMD's GPUs, notably, do not support it" Is it based on 7nm or not. The extract is unclear and there are no links currently in the text to the references. Moderation needed. This almost sounds like an ad for Nvidia.
"Their 700USD product is SHIT compared to our twice as expensive product!" - nVidia PR dude.
Seriously, $700 for a video card? After taxes that represents a full week salary for me :(. Not to mention I have to pay for my studio apartment and my bus pass.
Really it's very sad. I'm 50 years old and I still care about video game performance :(.... I'd also use the video card to help render video for my Youtube channel. But I can't help but think it's a waste. I put so much time and effort into the channel, but it's never really taken off and I don't think it ever will. I don't really have friends and so these sort of activities are supposed to be both my way of communicating with the world, and a little sideline where I can make a few extra dollars. But nothing ever seems to work out for me :(...
Author needs "help."
Maw! Fire up the karma burner!
Ask a CEO of a public company about a corporate competitor and they are going to trash talk them without any substance.
Even TFA opens with this admission:
Yesterday I spent two hours listening to the CEOs of rival companies talk trash about each other.
And they ask for details about his trash talk it all fizzled out.
When pressed about his comments, especially his touting of ray tracing she said, “I’m not gonna get into it tit for tat that’s just not my style.”
This is the kind of crap you would read in a Hollywood gossip rag with a twist.
Anons need not reply. Questions end with a question mark.
Seems like just a publicity thing...Jensen is Lisa's uncle after all...
You need to offer snark or a car analogy, other options just don't fly on modern /.
And even car analogies are suspect, since if the current "editors" drove like they "edited" we'd all be dead by now.
nVidia spouting "alternative facts" like an angry, petulant child who didn't get what they wanted for Generic Winter Gift-giving Holiday...
"Their performance, which is better than ours at the same price point, is lousy"
FIFY
I have installed NVidia binary drivers in quite a few different Linux systems, for the most part without any problems. I have been to install AMD binary drivers successfully in a Linux system exactly once, out of many attempts. I have no allegiance whatsoever to either company, but that is my experience.
To think, AMD's competitor would suggest that his own product is better than AMD'd product.
I mean, before this, who would have ever thought of doing that?
"I do not agree with what you say, but I will defend to the death your right to say it"
https://babeltechreviews.com/n... Thankgivings must be interesting for their family.
I think they should settle the issue with a good old-fashioned wrestling cage match. Donald Trump and Nancy Pelosi could be the warm-up act, and the main event could be Lisa Su vs Jensen Huang. I know the girls are smaller than the guys, but I think the gals are fiercer. All in all, the card looks pretty even. A good time would be had by all, and we could pay down the national debit a little by the vigorish from the parimutuel betting pool.
"Man is nothing without the works of man" -- Helvetius
One of Nvidia's engineers, on leaving the company, had his work on this 'feature' revealed on sites like Beyond3d- and he explained exactly what so-called RT is on Nvidia's TURING.
-the 'programmable' maths on almost all modern GPUs is done on what are refered to as 'SHADERS'. Shaders usually run an assembly language that actually allows for general coding. Most of you here will not know this.
-Nvidia's 'new' turing GPUs have only the same general purpose shaders as the earlier Pascal parts (well there are real 'tensor' cores- but these were included for new crypto currency mining algorithms which never happened due to the crash- they have nothing to do with graphics- and NO they are not even used for the so-called DLSS, a 'smart' upscaling tech).
-a smart cookie at Nvidia noticed that doing ray calculations using shader instructions was slower than the theoretical perfromance of the shaders. Shader assembly has an 'overhead' due to its focus on general purpose code. He suggested that a small ASIC (app-specific circuit) could be added that would directly deed geometry data from triangle geometry units to the shader ALUs. This was intended for NON-REAL-TIME back end ray traced output from industry standard tools, where a ray speed up of maybe 3-5 would be helpful.
-the DOWNSIDE of the ASIC is that it requires the insanely complex software machine that drives a modern GPU to be reconfigured from DX12 or OpenGL/Vulkan mode to be completely recoded for the ray function. This is a massive overhead for real time games- for which the ray calculation HACK was never intended.
-the best current Nvidia GPU does around ONE ray per pix per frame at low rez and low refresh. And that ray has but one LIMITED collision. Actual ray tracing requires TENS OF THOUSANDS of rays per pixel, each of which may spawn similar numbers of rays when they hit matte surfaces. Needless to say, not Nvidia GPU can do what the industry understands as ray tracing in realtime- not now and not in 20 years time either.
-the ray calculations Nvidia does do are hence limited to the speed of the chips ordinary shaders. Like I said, this makes for one ray with one collision today at 1080P at 60Hz. The coming 7nm parts will maybe double this. Chip shrinks are almost over- so maybe two more doublings to come. That makes maybe EIGHT rays per pixel, each with just a single limited collision from Nvidia's best (a two thousand dollar plus card) in maybe 2032. At 1080P. at 60Hz. Laughable, just laughable.
-ray tracing was developed BEFORE standard triangle rasterisation and eventually rejected for inherent inefficiecy, especially in regard to memory access. This is just a function of maths.
-today, if the run time wasted on ray calculations on Nvidia's GPU were used for rasterisation algorithms, they would achieve HIGHER visual quality with far faster performance. Nvidia RT is another Gameworks cancer on the gaming industry.
PS Nvidia's DLSS (their so-called AI-based smart upscaling) is no different from the last two decades of Photoshop plug-in CONS that claimed to magically recover detail when upscaling your photos. In tests each of these EXPENSIVE plug-ins prove no better than known simple upscaling algorithms. Nvidia's AI-claim (and the use of 'tensor' cores- but tensor cores at Nvidia HQ on their super-computer) is based on Nvidia using game footage at the highest rez to generate a giant look-up table that gamers can apply to their lower rez play to upscale. However, any game with a signoificant number of high rez textures overwhelms the data base, leading to the textures you care about being blurred. This 'feature' encourages game devs paid off by Nvidia to massively limit the use of high rez detail (signs, posters etc) in game to make the game DLSS 'friendly'. More trashing of game quality to suit an Nvidia sales agenda.
PPS the tensor cores will be missing from Nvidia's new 7nm GPUs. The ASIC that allows more efficient use of shaders for line collision with triangle maths will remain since it is a tiny number of transistors. But it will fall into disuse when a general ray API is added to DX12 because the ASIC is a hard coded limited solution.
Intel is the dominant CPU maker but GPUs are increasing the market sectors they are irreplacable in. First Gaming, now AI, and soon drone and embedded vision systems, and soon servers will use them for many mundane things like flight schedule planning. Once they break into tablets then We'll see some applications we havent thought about yet. TVs also need them evidently though some say they are better off without the smoothing.
Nvida doesn't make decent CPUs and Intel doesn't make decent GPUs. AMD makes both.
Yet NVIDIA GPUs and Intel CPUs still are much much better in actual use. Why? it's not the hardware it's the software. Intel MKL crushes AMD for anything numerical. And CUDA-writ-large crushes AMDs offering. And consequently everyone write APIs that depend on INTEL and NVIDIA first.
Even Meltdown and related bugs that come from Intel's inability to sign threads during context switches (AMD supposedly can but I really don't understand this) isn't enough to kill intel's out- perfrormance on AMD on numerical calcls.
But now that AMD has finally equaled Intel and Nidia you'd think an integrated CPU/GPU could be unbeatable. The achilles heel of GPU's is the lack of memory shared with the CPU and this could potentially solve that
Some drink at the fountain of knowledge. Others just gargle.
When can I get a P1 166Mhz on a laptop motherboard with a Voodoo 2, Realtek 8129 LAN port and Soundblaster 16 Audio. That's all I want. Give me a nice 13" DOS/Win98SE gaming laptop for under $500 AUD.
4K at 60fps for most games is just fine on AMD's new Vega 2, or a 1080TI (2080). You simply have to dial down a couple of NVIDIA sponsored settings that kill performance while making little to no visual difference. Indeed some of these settings (especially Nvidia 'anti-aliasing' and Nvidia 'shadows' can actually make the game look WORSE).
Nvidia PAYS reviews to always review games with all settings dialed to max. This can reduce performance by almost 50% (HALF) over great sensible settings.
It gets worse. The bigger issue- framepacing (stuttering, hitching etc) gets vastly exponentially worse, the more Nvidia gameworks features you activate. Nvidia gameworks has been so ruinious, it forced WB to refund any PC player of batman Arkham Knight, and the PC version of that game had FEWER graphical features than the console versions to boot. It tooks months before thios game was patched to remove constant stutter introduced by Nvida code that was forced to run on AMD GPUs as well.
The problem with PC gaming is NOT the lack of 4K gaming, but the lack of good games worthy of 4K (like God of War and Red Dead Redemption 2). Most PC games are not 'flattered' when played at 4K.
If Nvidia is talking trash about a product which isn't even released yet, it clearly have them scared for some reason. Why would they otherwise advertise their competition?
Huang might feel all cool crapping over AMD, but he should consider the message he's actually sending. This is about him making himself feeling better, not AMD.
I think the ground that Huang is on is shaky here.
I mean, if his point is that Vega has been, and continues to be underwhelming, OK, yeah. True enough.
OTOH he needs to be careful. The Nvidia RTX 20-series itself can be seen as underwhelming. I think the ray tracing support is cool and I've advocated mainstream support of RT for years. But really now, this is an alpha or beta-level capability, a true first generation product. Real world software support is vanishingly thin and will take years to build.
OK, but what about the traditional features and performance? Those are great, right? Well, they are better, but not by any great margin, and the card prices are high enough to discourage buying for that alone.
Thus you are left with a situation where the Nvidia RTX 20-series has a "barely there" justification for it's cutting edge ray tracing support. And it's equally "barely there" justification for traditional rendering features. And for all this you get to pay a premium price.
Everything you said about the AMD Radeon VII, Mr. Huang, would appear to apply to your flagship product as well! Would you care to state that for the record? I didn't think so.
For the last 12 years I've been a diehard ATI/AMD fan. There products have always been "good enough" and far cheaper than Nvidia. I just purchased the Nvidia 1080ti, because "ray-tracing" is garbage and the new 2070+ lines are way too expensive. None of the new features are really that ground breaking or likely to be widely necessary before these new cards will be obsolete. The only good thing about Nvidia's new line is that they caused the 1080ti card pricing to plummet and now I can have near top-of-the-line graphic power for very reasonable pricing. Who knows where we will be in three to four years, but unless a major upset happens, once this card is no longer cutting it, I'll go back to AMD and get a great card for a great price.
I agree and I hope you are correct.
"I'm not gonna get into it tit for tat that's just not my style."
Lisa Su being the bigger man. Bigger woman? Whatever.
Corruption is convincing someone that the selfless ideal is the same as their selfish ideal.
All of that tech is out of patent now (although the requisite JEDEC standards, etc are not, making a full implementation difficult for an open hardware developer.)
As an interesting aside, the Pentium supported up to 64GB thanks to PSE, and had support for dual processor with the right APIC configuration (three APICs if I remember correctly.) meaning that with some work you could have a dual pentium system software compatible with everything back to the 1980s, but with the capability of addressing 3.5-63.5G of RAM, in addition to the PCI/ISA IO space.
As an added bonus, there are PCI to PCIe adapters that can plug in a modern PCIe video card. Unfortunately many of those cards, even the ones with BIOS support (instead of UEFI) require a 686+ processor/chipset to properly run the option rom on (I haven't verified pci ids showing up on that particular system yet.)
The potential is there for a modern application capable Pentium class computer using entirely libre hardware designs based around iCE40 or related FPGAs, allowing the full design and toolchain to be available.
At 64 bit core width an iCE40 should be able to run at ~66-133 mhz (its maximum is ~533 at 16 bit width) which means it could handle PC100-133 SDRAM, plus the 66mhz bus of the pentium (or the 83mhz/100mhz SS7 added bus speeds.)
Done correctly this would provide a slower but still capable platform for open hardware development and user-trusted security.
Should have redactdot.org for /. articles which are little more than click bait.
Coming soon to /. "We asked 4 dogs and 75% of them said cats are evil." news at 10. /. and tech media problem is that the large tech companies are well past great grandfather in tech cycle age and produce mainly sequels to existing products with minimal new features worth reporting.
Open source projects are either very old with the same incrementalism of big tech companies; or likely to never get much traction and die in a year or two.
Leads to /. having minimal good material to report on. There are lots of good interesting articles out there which don't fit the /. template.
I was told (by people that I trust in that area) that Nvidia's cards are an expensive flop because they support ray-tracing which is slow, not used by any games, and doesn't look much better unless you look very closely, and which nobody wants, while all other performance aspects haven't improved.
Gee they say that all the time...
It's clear why Nvidia is such a disgusting company. Rotten head, rotten snake.
When all you have is a hammer, every problem starts to look like a thumb.
And cram them up your ass.
The other day i was building a new PC for a small business, where the PC will just be used for general office stuff (Email/Internet/MS Office);
As i built the machine with a AMD Ryzen 5 CPU, it needed a graphics card;
My choice was between a Geforce 710 1Gb for $45 or Radeon RX 560 4Gb for $165 (basically these were the cheapest nVidia & AMD cards available near me).
Obviously i chose the Geforce 710, cause why would you spend more than 3 times that amount for Graphics in a very basic PC??
And that's why i have no understanding of the company's current approach to Gfx cards...
Except it's the same price.
Nvidia CEO Trashes AMD's New GPU: 'The Performance Is Lousy'
To which the AMD CEO responded "Well at least I'm not gay"
1. better OSS partner then nvidia
2. card is cheaper then nvidia
3. has no raytracing, but the tech is new and doesn't work properly for games at this point anyway, AMD card will get it when it's more 'mature'.
4. performance might be a bit worse, but probably not too much, which is OK for me.
give me some good arguments why i should pick nvidia above amd again?
On a long enough timeline, the survival rate for everyone drops to zero.
Breaking news!
AMD's CEO says Nvidia's CEO's mom is so fat...
"Their 700USD product is SHIT compared to our twice as expensive product!" - nVidia PR dude.
They literally compared a $700USD product to the RTX 2080 ... which has an RRP of $700USD.
I'm more inclined to question their game. I'm sure Nvidia will be doing Intel style benchmarking here.
As long as it has a FREE driver, it will be better than nvidia.
Using nvidia means so much pain that as long as it costs less than 30% more for the same performance, it will still be worth it. That's not the case, as AMD usually has a better performance/cost ratio.