Nvidia thinks "Pascal is just unbeatable" and already decided not to move Volta ahead

[Tiwaz]

7 hours ago, lots of unexplainable lag said:

It's perf/watt is way better than Polaris, I mean Vega 56 draws the same power as an OC'd RX580 while being way faster.

 

Vega 64 is past the efficiency curve of the whole architecture. For the performance AMD wants out of it they have to push it way past that curve with high clocks and thus high power draw (also doesn't help AMD pushes a crapload more voltage than is really necessary). Vega 56 seems to be the sweetspot of the entire architecture.

compare the efficiency to NVidia

[Guest]

4 hours ago, silberdrachi said:

Its performance is less of an issue than the fact that i think its space heater qualities is causing it problems for me. The card is slowly giving out, and the face im in a nice hot southern summer isnt helping its lifespan much. Im mostly just crossing my fingers it doesnt give up the ghost before something worth buying comes around. Spending 400 bucks on a used 1070 is just unbelieveable.

In the same boat. Was eagerly awaiting top tier AMD cards to upgrade. But efficiency and noise are more important to me now more than ever. 

 

 

[laminutederire]

Major Dick move from nvidia though. They shouldn't talk like they're entirely out of touch... nobody wants an Intel situation with them sitting on their asses because they think no one can begin to touch them

 

[Jito463]

3 hours ago, i_build_nanosuits said:

we'll just have to disagree then...personally i hate to see the nvidia cards tested at 1650mhz when 95%+ of them can hold 2000mhz stable with the pushing of a single slider in afterburner and 100% of them can do 1900mhz...i think it's miss-representation of actual performance.

If every single card produced can do 1900MHz, why isn't every single card set to run at (or at least boost to) 1900MHz from the factory?  That seems a rather bold claim, to me.

[Marinatall_Ironside]

I'm not surprised at all.

 

Given the current market conditions, Nvidia as of now is in no position to release the GeForce 20 series of GPUs, and would be most likely unnecessary if they have no competition from the Red team.

 

I'll admit, I was hoping that Vega 56 and Vega 64 would be competitive enough to put some pressure on Nvidia, but its performance-per-watt is embarrassing to me. 295 watts of total board power and it can't even beat out a GeForce GTX 1080 which operates at a board power of 180 watts. With similar power draw on the GPU alone, the GTX 1080 Ti (250W reference board power) runs circles all day long. The Vega 56 has a board power of 210 watts, while the GTX 1070 has a board power rating of 150 watts. All figures I've listed are for reference boards (Founders Edition or otherwise) and AIB variations can have higher board power ratings, due to components used and overclocks.

 

It gets close in benchmarks I've seen, and the Vega 56 trades blows with the GTX 1070, but per/watt is just not there at all. I don't think Vega is a total disaster, it's just not what we've all hoped for, which was to blow Nvidia's GTX 1070 and GTX 1080 out of the water. I feel that the Vega 56 will sell better, just remember the power draw and get a bigger (preferably 750 watts minimum), higher-quality power supply with two 8-pin PCIe power connectors before getting one of these puppies. If you have one already, then you can disregard the power supply requirements I've outlined.

Edited August 15, 2017 by JurunceNK

[The Benjamins]

17 minutes ago, Jito463 said:

If every single card produced can do 1900MHz, why isn't every single card set to run at (or at least boost to) 1900MHz from the factory?  That seems a rather bold claim, to me.

I bet it is the same 100% as the 1700 can hit 3.9Ghz.

[Marinatall_Ironside]

3 hours ago, XenosTech said:

I know and the 56 kept up and beat it in some games... think steve used a factory oced 1070 too and at the end of his i think  vega was like a couple % faster

Steve on Gamers Nexus used both an Nvidia GeForce GTX 1070 and an EVGA GeForce GTX 1070 Superclocked for testing (comparing reference to aftermarket versions based on the same chip)

[Brooksie359]

1 hour ago, Bananasplit_00 said:

Yes. You can just change it in the console 

Why would they set the limit to 300 by default then? Why not have it at some arbitrarily large number?

[LAwLz]

18 minutes ago, Jito463 said:

If every single card produced can do 1900MHz, why isn't every single card set to run at (or at least boost to) 1900MHz from the factory?  That seems a rather bold claim, to me.

Maybe because there are other factors involved as well, such as power, heat, and noise?

That's not meant to be a counter argument by the way, I genuinely don't know what the power, heat, and noise numbers are for an overclocked 10XX card is.

 

What I do know however is that Pascal cards overclocks way better than Vega RX cards.

Has anyone compared Vega RX against Pascal when both cards are overclocked with what would be considered an average overclock for both cards?

[mr moose]

That sounds like a marketing call to me,   reading between the lines: "we have to many pascal units on the shelf to release a new unit and risk having so much unsalable stock on the shelf.  So go out and buy a pascal".

[Jito463]

1 minute ago, LAwLz said:

What I do know however is that Pascal cards overclocks way better than Vega RX cards.

I don't doubt that's true, I just questioned the "100%" number.  If the claim was that the majority of cards could do it, I would probably have accepted it without much thought.  It was simply the claim that every single card could do 1900MHz, that gave me pause.

[Marinatall_Ironside]

4 hours ago, M.Yurizaki said:

Besides, performance scales linearly with clock speed. So the stock clocks give you your baseline and whatever % above that you can overclock your part to, you get that same % in performance.

Until you reach a point where overclocking the hardware doesn't yield you any more performance (diminishing returns or even a negative impact on performance), and all you're doing at that point is shortening the effective lifespan.

 

There also exists a phenomenon that occurs when overvolting called electromigration. This is especially the case due to how tightly packed the transistors are in the 16nm FinFET process, and is also why Nvidia defines the maximum safe voltage limit at 1.093 as read on utilities such as MSI Afterburner.

[XenosTech]

23 minutes ago, JurunceNK said:

Steve on Gamers Nexus used both an Nvidia GeForce GTX 1070 and an EVGA GeForce GTX 1070 Superclocked for testing (comparing reference to aftermarket versions based on the same chip)

Oh yeah forgot about that and vega 56 wasn't too far off either in their testing

[dexT]

What a crap time to buy a GPU. You have a dusty-ass old Pascal and limp Vega.. all overpriced and never in stock.

 

I'd be holding off if a gamer and maybe investing in some of the sweet CPU action going on.

 

 

[Tiwaz]

25 minutes ago, rrubberr said:

That's exactly the issue with AMD. Whereas Nvidia, even when the new "Architecture" is just a reordering (like Maxwell was) of current hardware, manages to lower power consumption and raise performance (in games, Maxwell screwed everything else), AMD tries to move technology faster in irrelevant and impractical ways, not to mention Vega MUST be just a reordered GCN, which was garbage from the start.

100% agree

Ryzen are great for rendering but they just suck ass for single threaded performance, the single threaded performance of a ryzen is comparable to an ivy bridge cpu

[Marinatall_Ironside]

1 minute ago, rrubberr said:

Lol, Ryzen is especially unimpressive at rendering. It has the same weak FPU issue that virtually all of AMD's archs have.

Rendering at what? 3D? Ryzen has shown to be impressive at 3D rendering.

[Marinatall_Ironside]

1 minute ago, rrubberr said:

People have a tendency to compare unrelated hardware because of price, that is not a good metric. Moving on from that, no CPU is impressive at path tracing anymore.

Well yeah, why do you think I've GPU accelerated my 3D rendering tasks in AMD Radeon ProRender?

 

Also people like me look at 3D rendering performance on CPUs or GPUs is because that's what we do, and if we're in the market for one of those, it would be something worth looking at, especially if it's the software we use.

[Hereby]

DAMN IT IM WAITING FOR VOLTA SO I CAN REPLACE MY GTX 970!!! *im not mad :>*

[Marinatall_Ironside]

Just now, rrubberr said:

Lol, you say that as if that isn't what I do with my life. You think I need two Xeons and three GPU's for playing games? My garbage old Xeons perform within spitting distance of threadripper, while apparently using about the same amount of power (for two processors) in the software I use, and thus my value judgment is skewed away from threadripper.

 

Don't try to be an elitist among your peers .

Ask the professionals who make a living using the software in the industry. I'll bet you that they'll beg to differ, even if they're using older hardware and have no need to upgrade it like you don't have any need to upgrade it for what you do.

 

Look, I'm not telling you how you should be spending your money, I'm just saying that you shouldn't be dismissing one set of benchmarks just because it doesn't reflect real-world uses or doesn't represent your usage of the hardware in question.

 

The reason why people also compare two "unrelated" hardware on price is on how much value you're getting out of product a vs. product b, especially if the performance in multithreded applications is so close. Why do you think that people are comparing the Intel Core i9-7900X to the Ryzen Threadripper 1950X in the first place? Because both cost USD $999, and one of them is objectively better in multithreaded applications (video encoding, 3D rendering, etc.). One may have better IPC in single-threaded applications, but that's all it has going for it, while the other one has the 64 PCIe lanes, 16 cores with 32 threads.

 

Plus I'm impressed that your workstation has lasted you that long and still performs well within your expectations.

 

If you still disagree with me or just dismissing me, maybe you should become a hardware reviewer. I'll be intrigued to seeing you review computer hardware.

[Curufinwe_wins]

15 minutes ago, JurunceNK said:

Ask the professionals who make a living using the software in the industry. I'll bet you that they'll beg to differ, even if they're using older hardware and have no need to upgrade it like you don't have any need to upgrade it for what you do.

 

Look, I'm not telling you how you should be spending your money, I'm just saying that you shouldn't be dismissing one set of benchmarks just because it doesn't reflect real-world uses or doesn't represent your usage of the hardware in question.

 

The reason why people also compare two "unrelated" hardware on price is on how much value you're getting out of product a vs. product b, especially if the performance in multithreded applications is so close. Why do you think that people are comparing the Intel Core i9-7900X to the Ryzen Threadripper 1950X in the first place? Because both cost USD $999, and one of them is objectively better in multithreaded applications (video encoding, 3D rendering, etc.). One may have better IPC in single-threaded applications, but that's all it has going for it, while the other one has the 64 PCIe lanes, 16 cores with 32 threads.

 

Plus I'm impressed that your workstation has lasted you that long and still performs well within your expectations.

 

If you still disagree with me or just dismissing me, maybe you should become a hardware reviewer. I'll be intrigued to seeing you review computer hardware.

quick statement multithreading instances with low or non-existent AVX implementation.

 

HPC code using avx has shown even i5s out performing Ryzen 7... I guarantee you in those workloads the 7900x obliterates TR. 

 

Not that that matters for basically anyone outside of hpc though) (well x265 encoding uses it, albeit not as heavily).

[Marinatall_Ironside]

Just now, Curufinwe_wins said:

quick statement multithreading instances with low or non-existent AVX implementation.

 

HPC code using avx has shown even i5s out performing Ryzen 7... I guarantee you in those workloads the 7900x obliterates TR. 

 

Not that that matters for basically anyone outside of hpc though) (well x265 encoding uses it, albeit not as heavily).

Yeah considering that the Zen architecture can do AVX workloads in two cycles compared to the one cycle that Intel's architectures can do. Depending on what you do, Intel may be the better solution, but it all boils down to what you're actually doing.

[Marinatall_Ironside]

2 minutes ago, rrubberr said:

I would never recommend using a consumer grade CPU for real work, much less rendering for days or hours on end, as is my use case.

 

Comparing it to a 7900x, I don't quite see what Threadripper is better at, though. Based on the three videos I've seen and two reviews, I don't actually see the performance gains. With 60% more cores and an extra 30 watts, it doesn't seem to show a performance gain.

 

Why wouldn't my workstation perform within my "expectations?" 

Running consumer-grade hardware at stock can last as long and be as reliable as server/workstation hardware. This is nothing new at all, and is a legitimately perfect strategy, and a lot of people do it all the time if you're maximizing your bang for the buck and still getting highly respectable performance.

 

I told you, Threadripper is better at multithreaded workloads compared to Intel's current offerings at the same price point, once you consider that the software fully supports multithreading. It loses out in maybe one or two (maybe three) applications (namely Solidworks CAD and Creo 3D in CPU composite workloads). Even if Intel releases the rest of the Skylake-X lineup, they're going to find themselves in a very hard spot due to their higher asking prices, making the i9-7920X (USD $1199), i9-7940X (USD $1399), i9-7960X (USD $1699), and the i9-7980XE Extreme Edition (USD $1999) a very hard sell, and just not worth it. Even if the Intel Core i9-7960X beats out the Ryzen Threadripper 1950X (comparing core count at this point) by virtue of better IPC than the Zen architecture, you're paying a USD $699 premium just to have better IPC. Sorry, no. Not worth the premium to me when the rest of the i9 stable also has 20 fewer PCIe lanes. All this is foregoing gaming because these CPUs are really meant for multithreaded applications like video editing/encoding, 3D rendering, CAD/CAM, and compute, first and foremost.

 

http://www.gamersnexus.net/hwreviews/3015-amd-threadripper-1950x-1920x-review/page-3

http://www.anandtech.com/show/11697/the-amd-ryzen-threadripper-1950x-and-1920x-review/8

http://www.anandtech.com/show/11697/the-amd-ryzen-threadripper-1950x-and-1920x-review/10

https://www.digitaltrends.com/processor-reviews/amd-ryzen-threadripper-1950x-review/#/2-4

https://www.techspot.com/review/1465-amd-ryzen-threadripper-1950x-1920x/page4.html

http://www.tomshardware.com/reviews/amd-ryzen-threadripper-1950x-cpu,5167-11.html

 

 

Because everyone else has expectations. Why would it be any different for you?

[Jito463]

6 hours ago, rrubberr said:

Taking a smaller margin on a product does not make it more impressive

Except that they're really not.  The reduced costs of their MCM design - coupled with the high yields that producing smaller dies result in - means their costs aren't nearly as great as Intel with their monolithic dies.  Thus, they can sell at a reduced price without cutting into their profit margins.

[Jito463]

Just now, rrubberr said:

Their profit margin has been negative for the past four years virtually constantly. Apparently, it isn't working.

I'm not talking about the FX series, I'm talking about Ryzen/TR/Epyc.  We don't know the results of that yet, as the products only launched 6 months ago.  They couldn't price the FX line too high, as they just didn't have the performance to compete against Intel in anything but the low-mid range.  Now they do.

[Bananasplit_00]

1 hour ago, Brooksie359 said:

Why would they set the limit to 300 by default then? Why not have it at some arbitrarily large number?

Because when the game was originally made 300fps was an arbitrary large number lol

 

1 hour ago, Tiwaz said:

100% agree

Ryzen are great for rendering but they just suck ass for single threaded performance, the single threaded performance of a ryzen is comparable to an ivy bridge cpu

No its closer to Haswell.