VEGA IS FURY 2.0

>>60437691
>Zero IPC improvements
Source?

>>60437691
It's not even coming out until June. Q2 my fucking ass.

AMD has failed completely with Vega. By the time it's even out for consumers, nVidia will probably be less than 6 months away from release Volta.

I've been very impressed with what they've done with Ryzen so far, but holy shit, their graphics division is barely functional right now.

>>60437814
Volta = Q2 2018

They'll be fine.

>>60437737
maths?
Im guessing bassed on clock speed 50% increase leading to TFLOPS 50% increase

>>60437814
June IS Q2 you fucking retard.

>By the time it's even out for consumers, nVidia will probably be less than 6 months away from release Volta.
This is how it worked for years. AMD and Nvidia release cycles were always offset by ~half a year.
Also, Volta won't be available for consumers until at least Q2 18

File: 1494930210457.jpg (501KB, 2094x1242px) Image search: [Google]

501KB, 2094x1242px

>>60437691
IPC for a GPU

Good goy

>>60437885
>June IS Q2 you fucking retard.
June is the release for a product not aimed at consumers. At this rate the consumer version of Vega will be released after Q2.

That's what I'm saying is fucking bullshit. It's the Radeon Pro Duo all over again.

>>60437938
Well we don't know the release date for consumer Vega but hopefully it's not far off.

>>60437862
TFLOPS are a theoretical figure, mostly, that don't translate to real world performance at all.
A huge issue of GCN has always been keeping the cores fed in real world workloads, which is apparent in its lower performance compared to an Nvidia card with similar TFLOPS. Vega changes exactly that with its dynamic CU size or whatever it's called.
While you still have around the same maximum theoretical throughout per core (the cores themselves are totally fine and were not redesigned), during real world workloads you get a far higher GPU utilization with Vega than with GCN.

>>60437896
Why would you post this shit, at all, ever?

>>60437691
POO in GPU

>>60437814

buying volta next year means buying a intel cpu because they run like ass on ryzen and if you buy a i9 you must have brain damage

>>60437896
Disingenuous in the extreme. The pictured difference is about 1ms of additional frametime. Well above the frame rate of any normal monitor. Big deal.

>>60437691
Literally nothing in that idea says anything at all related to IPC. It could have a negative IPC change from Vega for all anyone knows so far.

>>60438480
Maybe Nvidia will fix it by then. lol

>>60437691
It has more advanced ROPs, so it may be much faster in games, as Fiji was ROP starved.

>>60438013
>Vega changes exactly that with its dynamic CU size or whatever it's called.

All that "dynamic CU size" does is allow the card to split/join cores so it either runs 1x fp64, 2x fp32 or 4x fp16 operations.

completely pointless for graphics, it's only useful for compute.

>>60438775
You sure? because console ports rely on fp16 quite a lot.

>>60438790
they showed a chart where vega improves in fp16, guess it's right

>>60438775
>Fiji was ROP starved.

Fiji choked virtually everywhere BUT the back-end.
64 GP/s is 130x overdraw for UHD@60fps.

Mostly it needed better geometry/rasterization setup and a faster memory subsystem.

>>60437691
graphics architecutre doesnt have IPC dipshit they dont work that way.

The Zero IPC meme you are pushing is from FP32 only being about 1.5 times higher from the 1.5X higher clocks. Hoever the FP16 (actual ops used for most videoame rendering) is WAY higher due to the rapid math function.

The real power is jumping from 8.6 TFLOPS of FP16 on the fury which was the same as its FP32 capability to a new arch which is pushing FP16 at pascal levels.

This is a 4000 shader card thats going to run more efficiently and push FLops-FPS performance at pascal levels.

Your looking at a GPU thats running hot on the heels of the 1080ti.

The engineering sample that went against a gtz 1080 OC'ed to 1900Mhz beat it out by 10% and that was in the 1400-mhz range with barely functional drivers.

This final products is over 1600mhz on the core.

>>60441655
desu I think it's gonna flop

AMD basically has to both beat nvidia on performance AND offer that performance for a lower price to actually succeed in the market

they didn't manage it, again

>>60441740
Incorrect. The cheapest vega card is called CORE and its priced at 399$

Thats at the current gtx 1070 price and based on the specs it will beat the GTX 1080 at 4Kwhich makes it the perfect GPU to drive a 4K TV pc game box.

The advantage nvidia has is in the neckbeard tier 144-165hz refresh rates. Normies dont give a fuck about that stuff and want the best product to drive 60hz.

The only way amd can win if vega can overclock to 2000mhz on air. but this wont even possible, maybe in liquid nitrogen

>>60437691
>Zero IPC
>8,6 tflops vs 12.5

>45% jump

>>60441836
Wgy is AMD naming so fucking faggot these days? CORE, EPYC, THREADRIPPER etc etc

>>60442041

The desperate shill having no recourse desperately thrashes about trying to lamd something that sticks.

It's amusing watching them desperately srruggle against the changing tide.

>>60441836
Fuck you need this is a Reddit thread with no basis whatsoever. Stop spreading lies cunt

>>60441740
Best on price.

Nope, AMD just needs to blow NVIDIA out of the water with a similar price. NVIDIA fans don't give a fuck about the price whatsoever

>>60442059
It's just as cringy as everything else on the market. Core i9, bigger numbers Lmao.

>>60438775
>All that "dynamic CU size" does is allow the card to split/join cores so it either runs 1x fp64, 2x fp32 or 4x fp16 operations
That's something different.

>>60442059
Vega Core and Vega Frontier are awesome names desu sempai. EPYC is a bit lame, but Threadripper isn't any more cringy than Bulldozer or ClawHammer.

>>60442613
EPYC is AMD spitting on Itanium's grave for that one last time.

>>60437691
Vega is a meme anyway
It wasn't good enough to release last year, so I can't see anything good coming out of it desu

>>60442624
Spitting on a harmless retard's grave is pretty damned tasteless tbqh.

>>60442041
>>60442260
>1050 Mhz vs 1525 Mhz

He is 100% right, Vega is a overclocked Fury at 14nm with no IPC improvment

>>60442667
That harmless retard would've killed usual PC market if not for amd64.

>>60442692
Then where the fuck all the transistors went?

>>60442692
With 16GB of RAM. That part's kind of important.

>>60442692
I guess if a Fury X on 1500Mhz delivers it'll be alright. Gaymer RX chips might even be clocked a bit higher.

>>60442260
Idk man, If I was able to run an original Fury X at 1525mhz without even overclocking, I think it would perform pretty damn well. It was trading punches with a 980ti in most cases, at it's regular clock speed. The 4gb limitation was pretty stinging too.

>>60438715
The question isn't maybe they'll fix it but why would they ever fix it. I look at it logically if you fix the driver for AMD that just gives them more money and more money allows them to put more money into R&D where they'd be competing with you meanwhile you can completely fuck over AMD side with Intel and since Intel doesn't have a goddamn GPU that's worth a damn and they aren't really competing with you you get no competition. It be smart for Nvidia not to fix anything.

>>60437691
>Frontier Edition
They aren't copying Nvidia's horrible cash grab known as:
>Founders Edition

As a GTX1080 owner I'm both looking forward to seeing how this performs and REALLY NOT FUCKING LOOKING FORWARD to the shitty threads on /g/ when it debuts.

But realistically, they better make it just as good as the 1180/1180ti or they'll be SOL.

>>60437691
You can't determine IPC at a GPU until you have the actual product in your hand. You flop equalize them to a prior generation card and see if there's a difference. In Nvidia's case there is no difference between Pascal and Maxwell. However there seems to be a significant difference between net Polaris and the 290 X

>>60442697
Please. x86 was never in danger. Backwards compatibility is always the most important thing, and 64 bit adds practically nothing over 32 bit and PAE.

>>60442713
Transistor counts are for the package, the package includes the HBM stack.

>>60442797
No you fucking moron, Vega is 520mm^2 die with only 4096 ALU. Where all the die space went?

>>60442628
They didn't have high-bandwidth memory two last year, this likely let them spin the GPU a bit more and get more done. Let's be 100% honest here AMD doesn't have consumer graphics cards that people buy, so AMD not releasing Vega and just putting in a stopgap gddr5x in it and called it a day. AMD could easily back burner it than just increase performance more more over half a years time and then release it when it's fully ready. Remember even when AMD has better graphics cards than Nvidia for cheaper than Nvidia people still buy Nvidia because they only think of Nvidia, so we are honestly lucky that AMD even bothers to release consumer graphics cards anymore.

>TFW I pre-ordered a Fury X and was the first one to post a pic of theirs on /g/
>Was waiting to see what Vega brings, wondering if it would be worth the upgrade costs - along with getting a new custom waterblock and everything else that goes with it
>Its not worth it
Relief.
Guess I'll just wait to see what Navi brings.

>13TFLOPS 520mm2 Vega
>15TFLOPS 830mm2 Volta

VEGA IS FAILURE NO IPC IMPROVEMENT!

>>60442972
NVIDIA used less dense designs for power saving reasons.

So theoretically, if we pretend it's just an overclocked fury and that performance scaled 1-1 with clock, what kind of numbers can we expect?

>>60442988
no, they crammed everything they could to compete with google tensor asics and going to get 1.7 yields again

>>60442997
Who the fuck cares about yields in enterprise where margins are stellar anyway?

>>60443007
you can't sell what you don't have

>>60442995
~1080ti, no less. But Vega has new front-end and some other shiny stuff so who knows?

>>60443018
Even with 1.7 yields they'll manage to stockpile at least something to sell.

>>60443040
you know how long it take to make a wafer "stage 1"? about 4 weeks there is at least 5-7 stages
that's on tested refined node
if something fails on new node - start over

don't take me wrong, it's great they make this giant chip, but it's not what volta was supposed to be
volta was supposed to be what ryzen is - MCMs that system recognizes as single GPU, they backpedaled on that
it's another trash rebrand of maxwell with refinements but essentially same thing

>>60442995
1080 Ti equivalent at 1080p that gets better relative to the competition at 4k instead of significantly worse.

>>60442972
Nvidia TFLOPS =/= AMD TFLOPS

>>60443104
not other way around? HBM is a leverage for 4k not 1080p

>>60442988
That's the most genuinely retarded shit I've read here for a very long time.

>>60443074
Volta was never about MCMed GPUs, even on the oldest roadmaps. NVIDIA simply has no interconnect suitable for that.

>>60442838
It's worth it for people sitting on RX480 / 390 level for sure. Your Fury X is obviously still very strong on 1440p. Not sure what you expected from Vega to be worth it. Performance above the Titan Xp?

>>60442838
Same here. Vega won't be a worthy upgrade for me.
I currently own a 980ti and was thinking about buying another used for ~$300 but one 1080ti is pretty much equivalent to 980ti sli. I don't think I'll ever buy nvidia again. Had so many driver problems with nvidia. It even broke my windows once. Had to format it.

>>60443119
I think that's what he said anon

File: 1494427425715.jpg (53KB, 406x402px) Image search: [Google]

53KB, 406x402px

Kepler, Maxwell, Pascal and Vega all have the same IPC, too.

File: 1493593406522.jpg (98KB, 680x709px) Image search: [Google]

98KB, 680x709px

>>60437691
>Zero IPC improvements

>>60438580
At 1080fps you get a new update every line with vsync off. 500fps is a lot less visible tearing than a mere 300.

>>60443137
no, Mega Titan Goliath Edge Meister Pro Gaymur Extreme Edition XPs in SLI.

retarded jokes aside, slightly higher than 1080TI level performance to achieve solid 60fps at 4K in recent titles would suffice.

File: insomnia-patrick-star-bloodshot.jpg (27KB, 576x384px) Image search: [Google]

27KB, 576x384px

>Been sitting with money in my hand for over two months now just waiting for AMD to make it worth my while
>Finally thought they were going to announce something yesterday
>They just announce workstation cards and tell the people that want a new GPU for games to wait

AAAAAAAAAAAA
Just give me the details you fucking pajeet so that I can make a decision to get a 1080ti or the highest-end Vega.

File: 0617101756801.png (211KB, 371x377px) Image search: [Google]

211KB, 371x377px

>>60443216
Get a 1080ti brainlet. This wasn't computex aimed at gaymer kids.

>>60443216
Goog goy, get 1080ti.

>>60443173
>Kepler, Maxwell
>same IPC

File: Vega murder.png (690KB, 1152x648px) Image search: [Google]

690KB, 1152x648px

>>60443208

>>60443301
That's 4k?

>>60443127
Why not use the NVLink interface?

>>60439210


As someone who knows a little thing or two about GPU architecture, Vega has plenty of improvements to IPC, the issue is to what degree of improvement. I think the biggest improvement will be tile-based rasterization, I think that coupled with HBM2 might allow for Vega to match or perhaps beat Pascal in perf/watt. For the record, I think GP102 might still have better overall performance, but you get my drift

>>60443337
NVLink is beefier SLI. It's not intended for die to die communication.

>>60443320

4k with Vega Pro drivers.

File: 1430850699549.jpg (13KB, 250x240px) Image search: [Google]

13KB, 250x240px

>>60443292
GTX 760 (1152 cores): 2258GFLOPS @ 980Mhz = 2.304GFLOPS/MHz
GTX 1060 3GB (1152 cores) : 3470GFLOPS @ 1506Mhz = 2.304GFLOPS/MHz


Same IPC since Kepler

>>60443372
We don't really know if it's Vega Pro drivers. But HBCC seems nice in practice.

>>60443301
I remember browsing here back during the 390X wait.
>Titan slayer, sources confirm, just wait
>DX12 performance off the charts, sources confirm, just wait
>Will bring world peace, just wait
Ended up being a 290X refresh that could barely contend with a 980 to this day.
Sure Fury was announced but good fucking luck getting a hold of one with such a limited production run, meanwhile 980Ti was stocked everywhere.
I sense the same shit will happen with Fury2.0, and here I am waiting to be disappointed yet again.

>>60443275
I am aware, but I expected at least something, some type of information that would give me a reason to wait.

Currently the only reason I see to wait is HBM2 because of the possible performance improvements and the fact that the Vega cards might work better at 4K than Nvidias cards.

>>60443290
I'd prefer not to, but if AMD keeps this up, I will.

>>60443387

History revisionists are fun.

At the time people thought the 390X was the Fury X. It wasn't until E3 we saw them using the Fury name.

And Fiji chips are still largely underutilized. They did change the world with the Nano.

>>60437691
>Zero IPC improvements
So it's like Volta?

>>60443320
>>60443301

Yes, with only 2GB of HBM2 memory

>>60443385

It will be REALLY interesting if HBCC is only available on Ryzen CPUs.

>>60442809
It's been explained multiple times, even by AMD themselves, where that transistor budget went.

>cache controller
>redesigned CUs targeting higher clocks at same power
which by definition means a less dense, larger chip
>tiled rasterizer
>additional (new for GCN) cache lines to the RBE

>>60443515
That would just kill their sales, people aren't desperate enough to use Radeon GPUs to buy a Ryzen CPU for that purpose. Only NVIDIA could pull something like that and even they aren't that stupid.

>>60443514

They said they accommodate up to 2 terabytes of cache with Highbandwidth Coherent Cache.

>>60443535

If the performance is better than any alternative and you're really concerned with 4k gaming, why wouldn't you invest in AMDs ecosystem like that?

>>60443556
>If the performance is better than any alternative
AMD don't have the ability to do this. They can never seem to downright obliterate the competition, simply catch up to them for a much better price or compete in a different market segment.

Even the highlights of AMD's history have always consisted of similar perf for lower cost, the only possible exception being the 4000/5000 series

>>60443579
>the only possible exception being the 4000/5000 series
And expanding on this, even when this happened NVIDIA were selling a ton more cards

>>60443579

You're seeing it though.

If they can leverage both CPU and GPU to make a faster, more coherent experience, that beats out all the other competition. And at these price points.

Why wouldn't you buy an all AMD system for a better gaming experience?

The numbers in that picture are better than Titan XP and Intel.

>>60443635
See >>60443588

Even if AMD's products destroy Intel/NVIDIA, people will still buy the latter. The number of informed consumers who base their purchases on actual research into the better buy are a minority in the consumer market

>>60443651
they need more effective marketing

>>60443651

Times have changed considerably since the 4000 released.

Even in an anti-AMD bias world, the information is available and could not be ignored if you wanted 4k gaming.

>>60443671
They need products that consistently perform better than the competition so their reputation grows.

Right now people still think their drivers are bad even though they've been infinitely superior to NVIDIA's drivers for two years. I can say that with confidence after coming from a Fury X to a 1080 Ti and missing Crimson sorely.

Out of curiosity, would the Vega Frontier card be any good for games at all?

Do workstation cards in general perform well in games, or are they just complete garbage for games?

>>60443791
they are meh because of completely different driver suit, look at radeon pro or quadro

>>60443791
They perform the same, they're just not worth the incredibly high prices relative to their consumer counterparts and are sometimes clocked lower.

That blue card looks so cool. Why can't us gamers get such good looking graphics cards?

>>60443382
Hmm. Wonder why GPU IPC doesn't make as big of a difference compared to CPU IPC.

>>60443913
You just increase clockspeed and add moar coarz, better drivers done

More cores and more parallel workloads

>>60437691
>HBM Stacks: 2
>Memory Capacity: 16GB
I thought 8GB stacks aren't ready yet?

>>60443936
No, they were ready, but limited production.

>>60443913
Because IPC is the only way to improve CPU performance anymore.

Eventually GPUs are going to hit a clockspeed wall and die shrinks are going to stop and performance scaling is going to be over.

>>60437691
>IPC
>gpu
go away /v/irgin, you have no idea what you are talking about

>>60444263

>>60443382

>>60444272
>floating point operations per second == instructions per cycle
kys

>>60443913
Because raw compute is only a fraction of what they're doing.

>>60437814
Vega was obviously built to get into machine learning because thats where the money is

Why do you think the top end has 16GBs of Vram? Why do you think they hardly showed gaming results? It's slightly faster than a 1080 at best.

>>60443967
Thats why AMD is working on the infinity fabric and multiple dies per package, Na'Vi should end up like Ryzen, multiple compute units per package, like 4 efficient RX 480s strapped together at the top end

>>60437691
>that memory bus
>that memory bandwidth
They cannot be serious. Why gimp the card like that?

>>60437814
So 2 weeks away. How is that not what was expected?

>>60442692
EBen if what you say is true that would mean that vega is a 15-20% improvement over a 580, which is huge.

>>60446570
Still almost twice as much as the Fury X

>>60443791
Workstation cards perform nearly identically to consumer cards. The difference is in the tightness of spec that allows for few to no errors. These errors are imperceptible in gaming. But for scientific analysis or simulations even minor errors in calculations can seriously fuck with the results. Hence the cost difference.

>>60446570
>takes 4 years for Nvidia to catch up to Fury memory bandwidth
>why is Vega only double!?

>>60446712
So the chips with imperfections are getting binned for the consumer RX variants that'll be released later on.

Interesting..

>>60447459
>average /g/tard doesn't know this
>welcome to fucking /g/

>>60442750
Intel hates Nvidia for many reasons, they are much closer with AMD (at least RTG) at this point. Besides, neither Intel nor Nvidia truly benefit from having a literal monopoly. They have spent hard cash on marketing, resulting in a public perception that puts AMD at the bottom of the totem pole regardless of how good their products manage to be.

It doesn't matter if the most heavily involved tech enthusiasts prefer AMD if Vega ends up destroying Nvidias stuff. The average consumer sees Nvidia as the best and only option for a GPU and Intel as the best and only option for a CPU. Market share doesn't just come from having the best (or only) current high-end products, because the average consumer isn't buying those. Changing public perception to favour the GTX 1060 over the RX 480/580 does way more for them than having the 1070/1080/ti at the top.

>>60442972
AMD BTFO AMDRONES ON SUICIDE WATCH VEGA DOA RYZEN DOA ATHLON DOA RAJA SUICIDE WATCH LISA SU BTFO

>>60443506
He said zero, not negative

>>60446492
>like 4 efficient RX 480s strapped together at the top end
My dick. I'm thinking Volta will destroy Vega even if it can somehow best the 1080Ti. AMD will be an entire year without a new lineup, that's just scary. Vega sounds nice, but what I really want is Navi. Having the ability to make a fuck huge GPU at lower costs is amazing for them and us.

>>60443913
think about the clock speed. 980 to 1506 that is 65% increase. if i7-6700 turbo boosted to 4.2GHz now imagine if the i7-7700K with the same IPC turbo boosted to 6.9GHz