Vega flawed

>>61191977
>Source: My Ass

File: poo.png (312KB, 595x654px) Image search: [Google]

312KB, 595x654px

>>61191977
sir pls delet

File: 1420834220908.jpg (43KB, 699x637px) Image search: [Google]

43KB, 699x637px

>>61191977
guaranteed replies

>>61191977
And the source is your ass?

File: hsw136.gif (22KB, 620x147px) Image search: [Google]

22KB, 620x147px

this hasn't happened with other hardware
can't believe how incompetent AMD is

File: 1490838034305.gif (3MB, 200x200px) Image search: [Google]

3MB, 200x200px

its bullshit but i believe it

>>61191977
I actually wouldn't rule out hardware bugs for Vega's terrible performance, but
>source: my ass

File: amd-vega-2.png (242KB, 616x330px) Image search: [Google]

242KB, 616x330px

Considering it manages 0.68 throughput per clock compared to GCN 3.0(fury) its very clear something broken on the FE model of the card

File: explosion demon.png (521KB, 639x557px) Image search: [Google]

521KB, 639x557px

>>61191977
>Phenom TLB bug 2.0 housefire boogaloo

>>61192063
>programmable
Means you need to program it, which means drivers and FE has no drivers.
Who the fuck thought Vega FE was a good idea?

A new silicon spin was talked about earlier than day 1

>>61192063
Are you saying that it's actually WORSE than GCN?

>>61192263
Yes! Vega is a total failure on every level!
And you know what you need to do.

File: 1446230336543.png (128KB, 2000x2710px) Image search: [Google]

128KB, 2000x2710px

Wait™ for Vega 2.0.

Pic related in who to fire.

>>61191989
>tfw poo in gpu backfires

Or the drivers aren't ready because TBR isn't enabled and obviously the new geometry engine isn't as well(did anyone even do benchmarks? techreport?)

>>61192366
I only remember someone doing TBR testing and it was not working.
Vega FE rasterized the same way Fiji does it.

>>61192383
I know, I'm talking about the geometry pipeline.
If that's not working either, obviously the chip will perform like Fiji.

>>61192408
Yes, primitive shaders are not doing their job. In all honesty we know nothing about Vega as uarch.

I just wait for RX benchmarks. Speculation is useless.

File: R9FuryX1400mhz.png (646KB, 1414x1095px) Image search: [Google]

646KB, 1414x1095px

>>61192263
Yes
A 1250mhz clocked fury which has the same core count as vega(4096) gets a better firestrike GPU score vs the 1600mhz vega
Here is a 1400mhz fury scoring nearly 2,000 points higher than the vega card

>>61192513
>Speculation is useless.
But it's funny. And since we know fuckall about Vega, it makes it even funnier.

>>61192433
We know a lot about vegas architecture
The flaw is in the new geometry pipeline its currently operating at one third rate as the other two lines are disabled

>>61192513
#TeamRed

#JustWait

>>61192544
We have full reviews on vega at this point with benchmarks and teardowns
What do you mean we know full all about vega?

http://www.gamersnexus.net/hwreviews/2973-amd-vega-frontier-edition-reviewed-too-soon-to-call

>>61192564
>We know a lot about vegas architecture
Only some high-level details and absolutely nothing about they way RTG implements them.

File: fury4790kbestresults.png (55KB, 946x957px) Image search: [Google]

55KB, 946x957px

>>61192538
>This Fury retard again
You can stop posting fake benches at any time now. You're not on list in validated scores and the highest Graphics score with your setup is like 3000 points lower so please fuck off.

>>61192614
is the vega FE validated?

If you are using a non standard bios the score becomes invalid

>>61192614
>amdtards lying on the internet
shocking

I'm going with the theory that shit won't be ready until late August, feature wise and optimization wise

>>61194053
>until late August
which year

>>61194082
This decade.

>>61194099
this century

>>61194156
It'll be called an app later on in the century, not a GPU

>>61194156
this millennium

File: feels wojak.jpg (12KB, 236x236px) Image search: [Google]

12KB, 236x236px

>>61191977
lets be real amd ROYALY fucked up, vega will be between 1070 and 1080 shit dosent have enough power to beat the 1080 ti im betting on amd goes with the "budget" gamer thing selling as good performance with 2 cards instead for 1 powerfull card -AMD Shill

>mfw gaymer rx vega actually BTFOd the 1080ti
>/g/ eternally BTFO

>>61194177
this billennium

>>61194232
>amd goes with the "budget" gamer thing selling as good performance with 2 cards
It was possible with polaris which is cheap to produce. Vega production costs more than 1080Ti based on die size and HBMeme.

I honestly think that rx vega will beat 1080ti when they activate everything in the drivers.

>>61194300
still wont even beat the 1080

>>61194319
>amd drivers are cancer anon

>>61194344
Will without a doubt beat 1080

>>61194368
I have no issues with amd drivers senpai. Its just a stupid meme

>>61194255
trillennium

>>61192280
Kek

File: 2fast2furyx.jpg (562KB, 2000x1500px) Image search: [Google]

562KB, 2000x1500px

>>61194434
anon we all want the rx vega to beat the 1080 but deep down we know that isnt happening

>>61194232
If the price is right like it was for the 470/480, I'm not complaining

>>61194497
We already know it beats the 1080 tho

File: 1496947363658.jpg (255KB, 1536x1536px) Image search: [Google]

255KB, 1536x1536px

>>61191977
Everyone's calling bullshit on this and rightly so, but can someone explain why it's wrong? Or just unverifiable?

>>61192061
I'm going more driver in fallback mode till the consumer version comes out.

>>61194531
regardless of price
300w for 1070 ti performance is terrible on 14nm

Card has a flawed design or bottleneck somewhere and I bet that is why it has been delayed so long

File: Sheeniemagic.jpg (619KB, 2000x1000px) Image search: [Google]

619KB, 2000x1000px

>>61191977
Nice slide thread, goldstein

>>61194756
>Card has a flawed design or bottleneck somewhere and I bet that is why it has been delayed so long
but amdfriends were telling me how much of a revolutonally new design would be for months completely leaving nvidia in the dust with async and HBM2

>>61194801
I don't see how those are mutually exclusive
>AMD has brilliant design
>Keeps fucking it up

>>61194801
You can have a good design but still have hardware bugs.

File: 1435707570126.jpg (36KB, 297x365px) Image search: [Google]

36KB, 297x365px

>>61194796
delete now you dirty goyim

>>61191977
Op is a faggot

I wish I never fell for the Freesync meme.

>>61194815
You're not funny.

File: 1461262274864.jpg (41KB, 688x456px) Image search: [Google]

41KB, 688x456px

>>61194843
we meet again mr shekelstein

File: 35225C4CFF744502BFE90D0DE1B77CFE1916700001AEC53287019.jpg (203KB, 381x424px) Image search: [Google]

203KB, 381x424px

>>61194843
Funny is subjective, jewish turbonigger

>>61194843
oy vey

>>61191977
Hello kinds sirs pls do the needful and kindly delet this tread

thank u and have a grate day

File: fear the samurai.jpg (48KB, 600x329px) Image search: [Google]

48KB, 600x329px

>>61194843
>goyim

>>61194830
checks out

>ayymdpoojets trying to derail the thread

>>61194942
shilled

File: Screenshot_20170702-213550.png (2MB, 1440x2560px) Image search: [Google]

2MB, 1440x2560px

>>61191977
I believe it.

AMD delayed the 6970 because they had to spend time sanding the corners off otherwise the heatsink wouldn't fit.

>>61194969
>ayymd literally cutting corners

File: 1467497478790.png (41KB, 802x799px) Image search: [Google]

41KB, 802x799px

>>61191977
Makes sense. The entire existence of Vega FE is kind of a mystery so I actually buy it especially amid claims that Vega FE has been unable to perform tile based rendering at all. It would also explain the crazy delays in getting RX Vega out.

I think they just wanted to get some version of Vega out H1 2017. Part of me wonders if certain higher ups get huge incentive bonuses for meeting release deadlines in regardless of the performance on release. Can't believe that everyone at RTG is happy with the state of Vega right now.

>>61195106
Tbh Vega FE is a good buy if literally the only thing you do is 16 bit floating point math

>>61191977
ye not like the exact same post you just made was on the amd reddit like yesterday made from a amd fanboy...
try again

>>61195355
>knowing what happens on plebbit
go back

I can't find any compute benchmarks for Vega.
I don't give a shit for gay men.

>>61195476
PCPer review and Gamersnexus review have them.

>>61195476
for mining?
Its bad like 10% above a 580

>bios to not use that part of the chips
And still draw 375 watts, just fired that pajeet

>>61191977
http://vocaroo.com/i/s0Wu9hb3JEeU

File: 1466384162977.gif (2MB, 263x200px) Image search: [Google]

2MB, 263x200px

>>61196517

>>61194434
>I have no issues with amd drivers senpai. Its just a stupid meme
This is the actual meme.

>>61194831
Why it's available for cheaper than gsync, and the latencies are pretty close. I have my gripes about amd, but freesync isn't one of them.

>>61195106
Look at how rushed Ryzen was. It's no surprise that RTG had a lot of pressure to release quickly as well. AMD has been hemorrhaging money for quite some time now and it makes perfect sense that there was a ton of pressure to make every deadline. That type of thing is one of the advantages of being a relatively small company, so AMD MUST take advantage of it. The downside is that if you make mistakes, it can be catastrophic. I fully believe this "leak" since it doesn't make a lot of sense to me that a card that, based on clocks alone, should be at least 50% faster than Fury is this bad at gaming.

I'm calling it. By RX release day, they release new drivers, for the FE too.
And these enable the actual new geom hardware, and benchmarks reflect so.
AMD are just hiding actual gaming performance until they have the gaming cards ready in volume.

>>61196494
>And still draw 375 watts, just fired that pajeet
Most likely doing geom using general purpose hardware, rather than the specialised, dedicated hardware.
The actual Vega RX release is going to be so much fun.

>>61196651
>AMD has been hemorrhaging money for quite some time now and it makes perfect sense that there was a ton of pressure to make every deadline.
AMD isn't hemorrhaging money at all, what are you talking about?

>>61191977
Why are you trying to push this lie so hard?

>>61196678
>123123D chess

isn't it a good thing? meaning that FE is in no way indicative of arch performance?

>>61196703
funny thing that people don't seem to understand when company has to grow perfect balance is when you lose a little bit of money meaning everything goes for development

if company starts to get profits without new product is when they can close doors next season

>>61194658
It is such a high level fuck up that no one associated with RTG would escape the stench of failure. It would literally be the bulldozer of GPUs.

So the people who could verify it, don't want to say and lose their job, and the only way we would know for sure is if someone from Global Foundries or RTG leaked this.

>>61192614
Thanks for posting this, I was seriously confused seeing clocks that high because my Fury X can only reach about 1150mhz before shitting the bed. 1400mhz would be impossibly good binning.

>>61196517
I like you anon

but aren't we getting actual gaming, non-workstation Vega cards?

>>61196875
it would be if there was more than 50k of FE cards
initial rumor that only 20k vegas will be available probably true
so all depends if RX Vega is same fuck up or not

>>61196949
Yes. July 30th.

>>61196960
...so what's the big deal then

>>61196900
As far as I'm aware, the only fury that has ever clocked that high was a one time LN2 overclock.

>>61196608
I'm not saying Freesync is bad. It's honestly served me well along with my 390 but had I a Gsync monitor I could've been rocking a 1080 or Ti for a while now and not having to Wait(to) for AMD to get its shit together.

File: 1497527945109.jpg (297KB, 1280x870px) Image search: [Google]

297KB, 1280x870px

>>61191977
How the hell they didn't catch it at the poototype level?

-Pajeet, did the vega passed all the test?
-No, fuck em, its perfect, burn the wafers Raja!

or they gone "Just fuck my shit up senpai" road?

>>61197019
or nvidia could simply include OPEN AND FREE standard into the driver tomorrow

>>61196975
Well, this is a really weird situation. Vega FE's release is enigmatic to begin with, but on top of that, it's entering the market at the worst possible time. People are starved for new GPU tech, or just fucking GPUs on the shelves in general due to the mining craze, and they've been highly anticipating a true high end release from AMD. It all adds up to be the perfect shitstorm of overhype.

From a marketing standpoint, AMD didn't position this card far away enough from the very sensitive PC enthusiast/gamer segment and aimed it at a very nebulous group of people that probably account for like 0.1-1% of PC enthusiasts, if they exist at all. Not truly professional, not truly for gaming, but not cheap either. Add it all up with some lackluster performance benchmarks in current games and you've got a recipe for a PR disaster that we see now.

This card was cooked up just so they could claim that they released Vega on target in the first half of 2017. It wouldn't surprise me in the slightest if the driver is half-baked.

>>61196517
> Wait for Navi
> Vega is very efficient guys (it's not)
> Wait for the "endgame"

>>61196703
>>61196842
AMD's earnings reports are public, you know. You can go look at their yearly losses any time you want. They'll probably be profitable again this year.

>>61197037
The fact that Nvidia doesn't support Freesync on their desktop GPUs is probably best for AMD because that sunken cost fallacy keeps a lot of people from switching to green I imagine.

>>61197099
>M-m-muh Nvidia shekels goy!
Just wait(tm), you silly goy!

>>61197140
If you though cash4poop applies to gpu business, you're wrong, Raja. Back to india.

>>61197032
russians aren't funny and your milk looks like semen

>>61197180
that's not milk

>>61197177
I'm either going to have to support Jews or Poo-in-the loos. Fuck that. I'm waiting, goy! I have no need for you now.

>>61197108
yes, they are public, all you seem to see is how much in the red they are every quarter not how much of that red went into R&D

>>61197180
>>61197204
Its condensed milk
tastes exactly like semen
yummy

>>61191977
the source is inside of his ass

File: 1498468951928.jpg (58KB, 240x241px) Image search: [Google]

58KB, 240x241px

>>61197032
>How the hell they didn't catch it at the poototype level?

Who knows.
AFAIK it was iriginally runnng on Fiji drivers .
Vega actually does have a lot of new hardware featueres.
Perhaps they tried to make their tiled rasterization, or that geometry shader work all that time until they fugured out its broken or some shit.

At least that would make sense why the fuck doesnt Vega FE work.
And yes it really isnt good enough for what it is, with that die, those CUs and that frequency it should run faster.

>>61191977
Sure fix monumental error in one month, maybe launch would be paper launch.

File: new_2017_imac-pro_full.jpg.og.jpg (155KB, 1200x630px) Image search: [Google]

155KB, 1200x630px

>>61197448
>Sure fix monumental error in one month, maybe launch would be paper launch.
Im fairly certain they were aware of that long ago and have started doing something about it.

This has to be the reason they releasing FE and RX long after they have relased Instinct mi25.

>btw App Lel kids gib fugged
>buying defective chips in a shiny case
for a premium price
>cant ever upgrade it
Lke pottery.

YASS, MAKE MACFAGS CRY BASED RAJA

>>61194497
now plug it with and x299 and roast some BBQ

>>61197219
>how in the red they are
>how much of that went into RnD

Which is what motivated them to push their products out the door as fast as possible. What point are you trying to make? It seems like you're just defending AMD against what you perceive to be an attack, where I'm just stating what AMD did and why. Their business strategy was sound, but as I said, there are drawbacks to every strategy.

>>61191977
Nvidia can steal the ideas and make it work properly.

Really need another company to compete with Nvidia/Intel. And is useless

>>61197747
Yes, shlomo, we should totally buy Xeon Scalable Family processors. AMD simply can't compete with THE MOST SCALABLE processors ever!

DUMP AYYYMD STOCKS NOW! I'LL BUY THEM AS CHARITY! DUMP NOW!

>>61191977
Prove it. Hint: you can't.

File: 1478862552900.png (490KB, 1024x732px) Image search: [Google]

490KB, 1024x732px

>It's another episode of /g/ has no patience and claims the end of the GPU world despite all reason
Great. A month or more of this bullshit before we actually get game-focused drivers and the consumer product.

Woo hoo...

File: 1281469062717.jpg (2KB, 126x103px) Image search: [Google]

2KB, 126x103px

>>>61191977

>>61197076
To add fuel to fire, even its mining performance is lackluster. It has the same performance as Fury when it comes to ETH mining.
AMD fucked up big time here. I don't see the point of Vega FE at all.

>>61191977
>the die changed

Do you know how unbelievably fucking expensive that is? It can and has bankrupt companies.

how fast is this vega fe mining ethereum?

>>61199106
AMD brought it upon themselves.
Release a GPU without drivers then expect no one to make jokes - you're fucking dreaming.

>>61199385
At least hardware itself exists and seems working.

>>61199203
30MH/s or something. Google it.

It needs to come within 10% of the 1080Ti at 70% the cost to be even interesting. It also needs to be uninteresting for miners.

I've heard that Vega FE performs better in games when in pro mode rather than gaming mode. Does anyone have any info on that?

>>61199149
RX480 has the hashrate of a TitanXpPp. How will nvidia ever recover?

>>61200152
Game mode makes no difference compared to pro mode. FE performs like shit anywhere.

>>61192614
go away retard. it is real.
http://forum.hwbot.org/showthread.php?t=142320

>>61200247
Why bother having modes then?

>>61201647
Gaming mode enables different driver features.

>>61196900
Not shit posting but before I sold my fury X with helluva voltage I could hit 1250, almost 1300 in some tests. That was with bios mods though and the fan speed up pretty high, but still. Check the ocn thread on Fiji bios modding, 1200 isn't too hard and close to 1300 isn't unheard of

File: 30321565043213312.png (327KB, 1491x1614px) Image search: [Google]

327KB, 1491x1614px

>>61191977
Is Nvidia will release this before Navi?

>>61200177
>>61199435
>>61199203
>>61199149
Fiji was trash for eth and everyone knew Vega would be too, I want to see the equihash results
>mfw the 3 fury tri-x I bough are now worth more than I paid, and make more than double per day what one 480 does for 50 bucks more

File: op-8.jpg (31KB, 272x400px) Image search: [Google]

31KB, 272x400px

>>61191977
Given the time it takes to verify a chip design and get all the production assets of it ready, and the time between the release of the Vega FE and RX Vega, they would literally have had to have the fix complete for a long time, and then they would have never put the Vega FE into production to begin with. OP is a faggot.

>>61202124
No, they lack interconnect.

>>61202272
Nvidia has a 300GB/s interconnect

>>61202427
That is point to point PCI-E substitute, while IF scales up to 512GB/s.
The paper itself is interesting. Same idea as AMD's "chiplets".

>>61202205
Then why the hell is Vega FE not using tile based rendering like AMD said "Vega" would be able to do?

>>61202527
Drivers aren't finished/stable enough for release.

>>61202527
Drivers with TBR are probably sorta unstable and they wanted to shit something that at least stable enough to run ROCm and meme learning frameworks?

>>61202540
>>61202550
tiling should be pure hardware feature though

https://www.youtube.com/watch?v=Nc6R1hwXhL8

>>61202569
It's a hardware feature, but it still needs driver implementation.
Guess why Maxwell2 was delayed.

Also nVidia has like zero experience and dealing with MCM designs while AMD made both Magny-Cours/Interlagos and EPYC, both being a shining examples of MCM done right.

>>61202492
http://research.nvidia.com/sites/default/files/pubs/2017-06_MCM-GPU%3A-Multi-Chip-Module-GPUs//p320-Arunkumar.pdf

It sure sounds like Nvidia already tested and has it in the works for future gpus.

http://research.nvidia.com/publication/2017-06_MCM-GPU%3A-Multi-Chip-Module-GPUs
Link to whitepaper.

>>61202663
And Intel used mesh topology in Teraflops Research Chip yet only know they reused it for KNL/SKL-EP.
AMD has something that already works wonders in EPYC. nVidia is yet to show their interconnect.

>>61202527
It's not like it has been officially confirmed or anything that Vega isn't using tile-based rendering. It's just a /g/ rumor.

>>61202767
Sorry, I meant reddit rumor, actually. So even worse, in fact.

>>61202540
>>61202550
Yeah I realize there could be a driver issue. I'm replying rhetorically to the guy saying that AMD couldn't possibly "fix" Vega by the time RX Vega is due for release later this month.

>>61202767
Afaik PCPer did the same testing Kanter did. Vega behaved like Fiji.

>>61202786
Even if that's true, that still doesn't explain why it performs worse than Fiji clock for clock.

>>61197032
I remember when raja was going on about how great this usb interface was to check for bugs on the chip and help speed along driver development

and yet here we are

>>61202803
Wider bibeline, also programmable geometry bibeline that's not programmed properly. Fuck I can pull fuckton of reasons outta my ass since we know nothing besides some high-level details about Vega.

>>61202804
I'll give you a few minutes to think why the fuck was Maxwell2 delayed for several quarters.

>>61202835
?

what's the secret, anon?

>>61202804
Writing software for hardware is one of the most arduous tasks a software developer can do and it's why computer engineering is it's own discipline

>>61202875
TBR requires drivers. And guess what was nVidia implementing in Maxwell2?

File: 137423794241.png (132KB, 436x857px) Image search: [Google]

132KB, 436x857px

>>61202668
Lower power consumption

>>61202881
I'm not discounting said development of those drivers at all, and he certainly made it clear that the hardest part for RTG was their development.

>>61202907
gimped memory :^)

>>61191980
Well it's not like people cannot check it once it becomes available... (not a fanboy of any card, just saying)

>>61202920
And easier cooling. Threadripper will be piss easy to cool.
Anyway, it's a race now.

>>61202966
Intel offers heterogeneous and homogeneous MCM designs such as the Iris Pro and the Xeon X5365 processors respectively.

>>61203032
One is simply a slab of eDRAM and the other is long deprecated.

>>61202920
>Lower power consumption
As compared to PCB interconnect, not compared to larger monolithic GPUs.

>>61203181
You can simply clock the silicon lower, since you have retarded amount of shaders anyway. Just look at EPYC: yes, GMI takes die space and eats power but the ability to have 128 lanes and 8 memory channels is well worth it.

>>61203262
is EPYC actually in use somewhere already? anyway to know?

>>61203282
We don't really know.

There is so much pent up demand in every single segment of the graphics card market right now (except for maybe low end discrete GPUs) that if RX Vega is priced appropriately it will still sell out anyway. It may not be very profitable for AMD but it might also not be as big a loss as one would believe based on these disappointing benches.

>>61203282
They're being tested in several data centers, but not deployed yet. I bet they only received evaluation samples for now because production of motherboards hasn't started yet.

>>61203321
AMD has been selling better GPUs for cheaper for several generations and it barely helped.

>>61203323
No, that's totally too late for evaluation samples. At least for 7+1 for sure.

>>61203282
Not for sure, but about 50% of the big silicon companies have taken up Epyc big time.

>>61203359
all of the money made by graphics division went into zen
if RTG was independent who know how much profit they would make

polaris pulled them to almost out of red last year, that's with cpu department which is bigger than rtg

>>61203489
Only Evergreen was really profitable and boosted their market share. The rest were meh.

>>61203498
market share =/= profit
it's all about margins and r&d returns

>>61203359
They lack the pizazz of Nvidia. Proprietary physics and gpu recording software and all that jazz really do work for marketing purposes. AMD only ever comes up with poor clones that they kill after a few months. They should invest more money into software, continue improving freedync, and bring back XGP

>>61203516
Radeon always had relatively slim margins. Heck, EPYC is going to be AMD's first high margin product since forever.

Whoever thought it would be a good idea to release Vega FE first should be fired, seriously.

>>61203535
>PHYSX
Is fucking dead and no one markets it.
>recording software
ReLive is bretty fucking good.
It's just nVidia has cult following and Radeon lost it ages ago. Catchy slogans sell the products, not products themselves.
>>61203557
Don't worry, mommy Su probably did it already.

>>61203569
I doubt anyone got fired. The decision to release Vega FE came straight from the top.

File: images-2.jpg (7KB, 299x168px) Image search: [Google]

7KB, 299x168px

>>61191977
So your saying an over-hyped card is actually incapable of performing what it's advertised to be?
>>M E N T A L

>>61203662
Hello /v/eddit. You need to go back.

>>61203662
Wait for proper drivers
Wait for Poor Volta

File: aragorn.jpg (96KB, 506x506px) Image search: [Google]

96KB, 506x506px

>>61203662
Insane in the membrane.

>>61203282

https://www.servethehome.com/amd-epyc-7601-dual-socket-early-power-consumption-observations/
> Bios isn't final yet.

There are a lot of review samples and test systems in the wild, OEM have been using design samples for a few months. The launch event had a bunch of prototypes servers from launch partners for show.

Testing a server platform takes a lot more time and thought and people doing it usually don't go posting on youtube.

If power efficiency and scalability really is good, big-cloud will adopt it. It's a matter of cost and hard numbers, not subjective reputation and marketing.

>>61204917
I mostly wonder about the retooling of all these server racks and if the costs will be worth it in the long run. This kind of business shit is beyond me, tbdesu.

My coworkers friend works at AMD and has done work on VEGA.

Apparently VEGA will not bail him out of his purchase of AMD stock at $15.

The reason it has been delayed is because it has problems.

Beware.

>>61204963
>I mostly wonder about the retooling of all these server racks and if the costs will be worth it in the long run.
>more memory
>more I/O
>more cores
>less power
Also the possibility of 1S system with 128lanes is beyond majestic.
It alone makes EPYC worth it.

>>61204976
>Source: my dad works at Nintendo

>>61204986
the only thing I understand from running a business is that constant money must be spent on growth in order to grow. I suppose this would be a wise investment, but are there major caveats in moving from a 100% or large majority xeon ecosystem to a zeppelin ecosystem? What are the drawbacks?

>>61204963

The fact tat AMD does not intend to change the platform for 4 years and that there is no segmentation of features makes it super versatile and possibly candidate for CPU upgrade in 4 years. Something Intel would never allow nowadays (new cpu = new platform).

>>61205019
>What are the drawbacks?
Some cache latency sensetive software. That's about it.

>>61205042
no quirks in compatability? Platform issues with integrated bridges? I assume if you're working in big datacenters even if there were issues you'd be smart enough to work around them at a large scale very quickly

https://www.youtube.com/watch?v=mskFUpnU8pI
Well that definetly IS how you make a good presentation.

>>61205028
To be fair Intel isn't changing sockets in the server space like the consumer stuff. Especially in the 4-8S systems.

And then theres Itanium's-oh-god-why-did-we-sign-this-contract-with-HPE drop in replacements from two or so months back.

Sooo hold on lads
if vega has IF
couldn't they already make a dual GPU act as a single GPU?

>the world's biggest sandbag

>>61205699
That's Navi.
>>61205741
That was EPYC. AES-128 engine inside of memory controller? It can encrypt separate VMs? It's fully transparent?
THE FUCK.

>>61205699
Wait for Navi(tm).

>>61205768
>>61205778
yea but if vega already has IF can't they do it now?

>>61205961
No, it requires another pretty time consuming hardware redesign. The one RTG hardware team is probably doing right now.

>>61205961
>16 core RX Vega card
>Powerplantripper

>>61205961
>>61205699
>>61206141

It was stated many times that the next gen APU will use Vega-based graphics, connecting with IF.

The early ZEN announcements were clear about the APU being a prety powerfull, enough that AMD would not have to create low end dedicated GPU...
I remember something about 1080p 60fps for "E-sports" games.


I think that The Vega delay forced them to create a RX550, something that have no equivalent on the previous gen RX400 series. so they could cover their ass in the low end.

>wait for zen they said
>wait for polaris they said
>wait for vega they said

Thoughts on Vega as an APU component?

From some things i've read it looks like all their APUs going forwards will be based on Vega.

>>61206633
just wait. what part don't you understand?

>>61207401
By the time APUs are out, either the drivers will be mature or Vega will be buried under 15 feet of concrete in AMD HQ.

>>61207401
>APU

Useless meme trash only used in poorfag builds and consoles that hold back entire industry in which we have GPUs powerful enough to run games 10 years into future if they were actually optimized.

>>61209762
Oh god it's the same autismking anon that shat up the other Vega thread I can already tell. Hit the deck. Major league autism incoming.

>>61206633
Nvidia is going to beat them at their own game with scalability.

>>61210612
With what interconnect?

>>61210750
One that isn't designed by poo in the loo.

>>61210835
Raja had nothing to do with IF, polshitter.

File: 1468152943828.jpg (827KB, 515x1046px) Image search: [Google]

827KB, 515x1046px

let's be clear about something. Right now, with Vega, the following does not work:

- Rapid Packed Match
- 2 Ops per shader per clock cycle
- Tile Based Rasterization
- HBCC
- Primitive Shaders

The 2 Ops per shader per clock alone should make it outperform Fiji and put it on par with a 1080. TBR should put it on par to in between 1080 and Ti. Rapid Packed Math & Primitive Shaders (if implemented at the driver level), should put the card into Ti territory straight up, and the HBCC if implemented at the driver level, should put the card at min Ti and between Ti and Titan Xp.

But currently, not only is none of that implemented yet in the driver, I will go as far as to say that close to half those features will be "they are there for the developers to pick and choose." Which means all the benefit and value of those features will be useless, and they'll consume power and offer fucking nothing in return.

This card, based on features alone, should be going toe to toe with a 1080Ti and in some cases, exceeding it. But because of zero drivers, and AMD's half assed "we'll let the devs choose" approach, its going to basically be 1080 or worse.

Its amazing how much they invest into feature R&D and throw it straight into the trash, and as a result, is so utterly infuriating to watch in real-time.

It's like Ferrari built an amazing car and then half assed the ECU so it runs on random cylinders and 1 wheel spins.

>>61210972
welcome to poo in loo project management

>>61210972
So is it hardware or software?

>>61210972
Pic shows 4x Ops per clock and it clocks at 1.6ghz. Why isn't it 6x Fury X performance?

>>61211261
yes

>>61210972
>>61211269
Even half this would destroy the Titan Xp at 2200mhz. Is this how they based their Poor Volta claim?

>>61211269
Theoretically it should be
Clock speeds and core count alone it should be 240% faster than a RX480 that is not factoring any optimization they have claimed in the media postings they made in Jan
Something is not working correctly on Vega this is VERY clear at this point. Simple math using wattage vs compute shows its worse off than Terrascale 2.0 which is 7 generations of architecture ago on AMD GPUs
I am not sure what happened but FE is most certainly an emergency card and Vega being delayed 8 months~ might resolve the issue but who knows.
Raja said the cards should be competitive with the 1080 ti in 4k gaming but so far we have seen 1070-1080 level of performance

>>61211261

The physical chip, with all its bells and whistles, if it can be used as intended, will absolutely and thoroughly brutalize the Pascal architecture like the 9800 Pro did, the 7970 did, and the 290x did for its respective competing uArch.

But what's holding AMD back, is drivers and most importantly, this idea of "freedom of choice." Nvidia does offer freedom of choice via its <name>Works software suites; but all its hardware level features are implemented and handled at the driver level. They leave absolutely nothing up to the developer, because they know that most developers are fucking hacks who will take the easiest way out.

So all AMD has to do, is invest more money into a competent driver team that can implement all the features it has in hardware, at the driver level, for handling such that all the game developer has to do is write standard code; and the driver can go: "oh, this is good for RPM, send it to that. This is good for primitive shaders, send it to that; and the rest is good for Async operations; I'll let the hardware scheduler handle that."

>>61211310

Poor Volta claim is based on a theoretical option, where a developer will take advantage of each and every feature in Vega, to design a game where when compared to 1080Ti/Xp or even 2080Ti/Xv (Volta), wouldn't matter; as the card would utterly triumph over.

AMD's methodology requires the developer to give a shit, and take their time with the software and optimize it properly for the hardware. All five features I listed in Vega, don't exist in consoles at all. Further, only two of the five, TBR & 2 ops per shader per clock, exists in Nvidia's uArch.

So from a developer's perspective, there's zero reason to use RPM, HBCC, & Primitive Shaders in their code; even if it can potentially offer 100-150% more performance per clock. Because remember that not everyone is on Vega. Many people are on 780s, 980s, 1060s, 7970s, 290s, 480 and 580s, etc.

If its not implemented @ driver level, trash.

>>61211347
>worse than terrascale 2.0

hahahaha OH MY FUCK

>if my old 6950 had more CU's and was clocked equally it would be faster

fucking embarassing. What the hell happened here

File: amd_vs__nvidia.jpg (95KB, 311x837px) Image search: [Google]

95KB, 311x837px

>>61211494

Its pic related, except, the final pic is AMD and there should be no driver, and the picture should be of a car on paper, on fire.

Literally. Beyond embarrassing.

>>61211449
So AMD should create their own games so it'll be 100% optimized for AMD products?

File: Vega_vs_Pascal.png (215KB, 311x837px) Image search: [Google]

215KB, 311x837px

>>61191977
>>61211494
>>61211347
>>61211310
>>61211279
>>61211269
>>61211261
>>61211194

>>61211688
this hurts me a little inside
>select all squares that match label: CANOE

ok google

File: amdvega.jpg (156KB, 633x758px) Image search: [Google]

156KB, 633x758px

>>61211688
DELET

>>61211449
>Poor Volta claim is based on a theoretical option, where a developer will take advantage of each and every feature in Vega

The only feature of Vega that has to be implemended by the devs is Primitive Shader.
Every single other feature is at the driver level and has nothing at all to do with third parties.

>>61211766
The other thing is that it really depends on how easy/not easy it is to implement the feature. ATI had hardware tessellation long before Nvidia did, but it never got used because there were no dev tools for it. Years later, we got DX11 with APIs for tessellation and suddenly devs are tessellating fucking everything, including objects that can't even be seen by the player.

To be honest, I have no idea what that anon is talking about by saying everything needs to be implemented in driver. That's completely false. There just need to be proper dev tools. AMD has shown they're committed to good dev tools through their release of Mantle and support of Vulkan. I have no doubt that they will provide dev tools for Vulkan and for a DX patch, should Microsoft want the patch (let's be real, they already cannibalized Mantle for many DX12 features).

>>61192263
NCU is still GCN

>>61211811
>- Rapid Packed Match
>- 2 Ops per shader per clock cycle
>- Tile Based Rasterization
>- HBCC
>- Primitive Shaders

Out of these features only primitive shader requires developer awareness, others are low level shit that devs dont even have to know about, AMD simply has to make them work.

>>61211766

Right now, not a single one of those features exists at driver level, given Vega is running Fiji drivers. On top of that, Siggraph is another 27 days out; and Vega FE is showing a performance regression to Fiji--and in some cases.

A card that's supposed to compete with the 1080Ti is performing worse than its predecessor, and sits between a 1070 and a 1080.

A fucking 1920 shader card is only 10-15% weaker than a 4096 shader card. Yes, Nvidia/AMD shaders aren't equitable, but in the eyes of the common consumer who only looks at graphs and sees numbers and doesn't understand the architecture; sees a card less than half as powerful, being on par with it.

TWENTY SEVEN FUCKING DAYS of pure freedom for Nvidia to do nothing, and let the entire PC gaming community shit all over AMD.

Vega FE was AMD's one chance to make a good impression and they failed through the fucking floor. Unless the highest performing RX Vega that's supposed to compete with a 1080Ti is priced at $550 or less, in severely undercutting the Ti and even the 1080, everyone's gonna look at the RX Vega and go: "$600? 300W? 1080 performance? Fuck that, I'll spend $450 and buy a 1080 that outperforms it."

Vega is Bulldozer & Skylake-X rolled into one hot chip right now.

>>61211841

Yes, but none of those features are active in the driver that Vega FE is using. Consider the market implementations: releasing a card with brand new features, giving your direct competitor a talking point about how garbage your card is and how incompetent not only your company is but your entire leadership to release a high end product with NO DRIVER.

Not even Jensen, if he was drunk off his ass and snorting cocaine off a hooker's tits would do something that fucking stupid.

>>61211698
https://www.youtube.com/watch?v=TVeASyYtRCc

95avg vs 70avg
7970GHZ edition 2048 / 1050 vs 1536 / 880 and only avg 25 higher fps

>>61211850
>Right now, not a single one of those features exists at driver level, given Vega is running Fiji drivers.

Thats the point.
Everything else is irrelevant.
Vega FE gaming benchmarks are entirely irrelevant until then.

Considering their provded footage they do have those features in their internal drivers. They probably dont release them bcause they are unstable as fuck.

They have shown sniper elite 60fps@4k, HBCC Tomb Raider and other crap a month ago, they have a month more to go.

If they can partually half ass their shit together till the release its gonna be a sucks ass.

>>61211911

>everything is irrelevant

No its not. Because AMD put out a new uArch with no driver support, expected people to pay $1000 for it; and added an option to game on it.

If it was strictly a prosumer card with no gaming benefit, it would be a different story. But they released a card and said you can game on it by the way. So the consumer went and did just that, and found that the card is pure garbage.

You're moving goal posts.

>>61211911
Two Vegas doing 60fps

>>61211964
In fact they told you not to fcking game on it.
They did say its good for solidworks and blender, it actually is true.

>>61211972
>I-It's just to show they can feed threadripper guise

>>61211991
How come you can game on the P6000 without losing FPS vs the Titan Xp?

>>61211972
Two vegas were doing Prey with tripple screen tearing.

No fps counter but tripple tearing indicates FPS about 3 times the refresh rate.

File: ae0.png (529KB, 1280x800px) Image search: [Google]

529KB, 1280x800px

>>61211991

>don't game on this
>include a gaming mode in the driver for it

Who the fuck runs this shop.

>>61212041
90fps?

A single 1080 Ti is 60+, no?

>>61212050
was the monitor 30fps or 60fps?
180fps?

>>61212025
Because p6000 works entirely and vega functions partually for now. Its like an alpha version of a product.
Until the driver updates, indeed.

>>61212173
>>61212050
Prey is a potato game, 160-180 @4k sounds labout right for two high end gpus.

>>61211766
>Every single other feature is at the driver level and has nothing at all to do with third parties.
So knowing AMD those features will be available in 3 years.

File: amd2.jpg (52KB, 568x612px) Image search: [Google]

52KB, 568x612px

>>61212192
We'll just have to wait and see.

>>61212173
Projector confirmed 30hz

File: 1493096553759.png (433KB, 600x562px) Image search: [Google]

433KB, 600x562px

>>61212041
>They were using Freesync.
Can't make this shit up.

>>61212211
>gaming projectors with synch exist

>>61212230
It's software based , not hardware.

>>61212247
You still need a DP-capable display that can change refresh rates. Projectors aren't exactly the type of display that can change refresh rates on the fly,

>>61212046
>this is a weed whacker
>it is for lawn detail work
>you could mow your entire lawn with it, but that is not recommended

fucking weed whacker manufacturers utter fucking retards, am i right

>>61212261
>>61212247
>>61212211
They explicitly stated that there was no sync enabled, adaptive or not.

File: me.jpg (37KB, 600x600px) Image search: [Google]

37KB, 600x600px

After reading the youtube comments section on an r9 fury review I am completely convinced that gamers should be banned from discussing computer hardware in any form by threat of immediate confiscation of their gpus.

and retards like this >>61212211
have convinced me that /g/ is almost on the same level as those retards.

>>61212324
/g/ is indeed filled with retards. remember what site you're on. You are required to sift through shit posts.

File: why amd can't properly supply.jpg (139KB, 1024x576px) Image search: [Google]

139KB, 1024x576px

>>61212336
I hate the internet

>>61199152
If it actually happened, raja should be fired

Frankly, the shitshow as it is means raja should be fired. As well as the middle managers under him who helped enable this shitshow.

Keep the engineers though. They're the important guys. It's not their fuckups that matter, it's management.

File: hyperspurdo.jpg (311KB, 560x560px) Image search: [Google]

311KB, 560x560px

>>61212418
>instead of building a normal big polaris, we're skipping straight to vega
>yes HBM2
>of course I want tile based rendering
>what you mean I have to hire engineers to make drivers?
>of course I can deliver by the end of H1
>we'll just call it frontier edition and promise that rx is going to be better
>who cares that it has worse compute ratios than GCN 1, it has higher clocks!

Raja you dumb fuck

>>61212418
pretty sure Su made FE decision
I still hope we are being rused
hope dies last

also you are overestimating how expensive it is to change the die, it's expensive if product gone full blown production of 100k chips a month, but for small series it should be under 5m

>>61204976
>Apparently VEGA will not bail him out of his purchase of AMD stock at $15.

Can you speak fucking english?

Holy shit that isn't even a sentence that makes any fucking sense. Goddamn fucking immigrants

>>61212465
I speak finance-istani

He's saying that the idiot bought AMD stock at 15 dollars and he was banking on vega to raise the stock even more. Traders and their ilk are notoriously shit at speaking english, mostly to keep normal human beings far away

>>61212494
native english speakers just suck at non analytic sentence structure
you can't understand a thing if a word is in wrong position
synthetic and semi synthetic languages are far superior to your half assed germanics

>>61204976
>my dad works at nintendo
>The reason it has been delayed is because it has problems.

lmao

>>61212522
are you fucking /lit/ shitposting at me right now because you better fucking not be you john green loving fuck

>>61212522
Can you give me examples

pls don't be esperanto or some shit

>>61212522
Loose sentence structure is OLEV
In other hand you will need a massive grammatical monstrocity to make it work.

>>61212459
Su's a woman and is just going to go along with whatever the majority board members/execs say, because that's how women work.

Su's gotta go full dominatrix on the RTG management team, but she probably doesn't have the balls. Raja will keep his job because, once again, women are nice and won't fire people for fuckups.

File: amdloo.png (316KB, 882x758px) Image search: [Google]

316KB, 882x758px

>>61191977

>>61212557
Russian.
No matter how you translate Master Yoda to Russian it still makes sense and isnt even incorrect.

File: 4 minutes in paint.jpg (64KB, 311x838px) Image search: [Google]

64KB, 311x838px

>>61211688

>>61212557
just an observation that I can understand almost anything
and native speakers sometimes have issues

my native language is russian, so structure means nothing to me
russian can write same sentence dozen different ways and still be grammatically correct all that changes is style
german is somewhat the same somewhere in between english and russian

want to know more read up on language families(analytical, synthetic, fusional etc.) it doesn't mean one is worse or better they all achieve same things but with different methods

>>61212595
this guy is correct, I was confused at first why everyone makes jokes about yoda, took me a while to understand why

>>61212628
It might be correct but is it crystal clear? You can get the meaning from one of yoda's sentences but it just doesn't flow correctly. English has a specific way of speaking that kind of enforces this.
For instance, we don't say clop clip. Ever. We will always say clip clop. Or click clack. Never clack click. Cant' remember the name of this little quirk but if you go around breaking this conventions people will think you're charles fucking manson.

>>61212580
that's what any CEO who doesn't own more than 20% of the stock does.

Source: I work in silicon valley and have worked at plenty of company headquarters

>>61212646
right side and left side definition if I remember the terms right

>It might be correct but is it crystal clear?
yup, style and emotional coloring changes but meaning stays the same
it's hard to explain, it's more of a feeling than "rules"
that's why it's called synthetic language, you can build it on the fly, sort of

https://en.wikipedia.org/wiki/Morphological_typology

>>61212646
Lets say it does make perfect sense, but isn't exactly how people normally converse.

>>61212628
English can be written in many different orders, you are just getting them wrong.

>>61212706
>>61212695
no wonder policy with the russians always takes so god damned long

>>61212628
German is a language specifically designed for war, I am positive of this.

You can get the gist of a sentence without hearing the full sentence. Everything is implied in the conjugations of every word. Although they likewise can say absolutely nothing for an entire paragraph and then throw in the last final word that suddenly makes the entire previous paragraph finally make sense.

>"In the daybeforeyesterdayshortlyaftereleveno'clock Night, the inthistownstandingtavern called `The Wagoner' was downburnt. When the fire to the onthedownburninghouseresting Stork's Nest reached, flew the parent Storks away. But when the bytheraging, firesurrounded Nest itself caught Fire, straightway plunged the quickreturning Mother-stork into the Flames and died, her Wings over her young ones outspread."

>"The trunks being now ready, he DE- after kissing his mother and sisters, and once more pressing to his bosom his adored Gretchen, who, dressed in simple white muslin, with a single tuberose in the ample folds of her rich brown hair, had tottered feebly down the stairs, still pale from the terror and excitement of the past evening, but longing to lay her poor aching head yet once again upon the breast of him whom she loved more dearly than life itself, PARTED."

>Thread derailed by grammar Nazi's.

Every fucking time.

>>61213400
The plural of Nazi is Nazis, not Nazi's.

>>61213420
it's navis

>>61195613
So the Vega cards will soon be cheaper than the 580?

>>61213534
yes, and it will get sold out ragrdless

>>61196822
Except if it means that the regular cards have this flaw as well, or that they will push back their launch a lot, or quantities will be limited due to having to fix production issues

https://www.youtube.com/watch?v=b_ZMOn0X6jw
literally

>>61214156
I know Nvidiots are loving this but Jesus Christ, this is hardly AMD's fault, all they promised was an improvement, not THE best GPU on the market. Blame shitty clickbait YouTube channels for making out AMD was releasing a 1080ti killer, all they're good for is shilling, fuck them.

>>61196651
the only reason vega is out now is to stop the idiots from shilling
we do know already from Buildzoid that amd has disabled almost all of the ncu features the card normaly has
plus we do know for a fact that its using fury kind of level drivers
this is all the biggest sandbagging even from when amd was lying to everyone about the capabilities of ryzen chips
people on beyond3d already got 2 on their hands they tested every single bit of it if they concluded that the card is literally held back by amd you know this shit is going to analy rape everything

>>61214425
>it will rape everything
Depends on how good AMD's TBR implementation is. And how well primitive shaders work when wrapped around usual vertex code.

>>61211850
>>61212025
This is a niche prosumer card after all; the WX9100 competes with the P6000, not FE. Vega FE was released in conjunction with ROCm 1.6's MIOpen and prefaced with Instinct, so it shouldn't be a surprise that drivers are incomplete. Only primitive shader macros are present in disassembly, not too many other feature names are visible, DSB/PDA/TBDR are also all disabled/missing per B3D despite primitive discard being present in Polaris.

RTG and AMD needed to put something out for the investors as they promised Vega H1, it isn't incompetence, it's just typical damage control when you can't meet a deadline while you have momentum. This is indeed a horrid image to project on AMD, but they don't really care. Vega accomplishes three major objectives, penetrating the HPC/ML market, getting into the professional market starting with the Mac Pro win (in fact OS X drivers work better than Windows drivers at this point) and they also apparently think it's good enough for laptop APUs.

>>61211819
Not sure about that. It's called GCN5 but it's all semantics. Pipeline was changed for higher clocks and there's likely a laundry list of SIMD changes before we even consider the CU itself. It is definitely still inspired by GCN though.

>>61214707
And Fermi was *technically* a g80 successor. Yet Fermi was drastically different to it.

>>61210750
See >>61202668, it refers to a single-ended signalling technology that nVidia has been developing.

>>61215009
That's not even in silicon yet. Come on we all know that nvidia will never be the first to achieve MCM GPUs.

it's highly likely that SOMETHING is wrong with Vega FE.

meme all you want about AMD, they're not the company to accomplish absolutely fucking nothing in two years. they're not Matrox, for fuck's sake.

>>61210972
>let's be clear about something. Right now, with Vega, the following does not work:
Source: My uncle works at Nintendo.

>>61214744
Yes, that's my point. GCN5 already outlined a lot of things which are in Vega. Conservative Rasterization, TBDR, HBCC, larger L2 pool, immediate framebuffer, DSBR+PDA, longer pipeline (thus 1.6GHz), native FP16, a better HBM controller, IF, plus multiple SIMD level and RBE changes. A lot of these technologies are playing catch up with nVidia, still lacking is better color compression and the level of voltage gating nVidia has.

>>61215025
We know precisely what is wrong with Vega FE.
>>61215042
He even said B3D, anon.
People already tested what does and what doesn't work in Vega FE. Not everyone is retarded /v/edditor like you.

>>61215045
In all honesty better DCC did fuckall in Maxwell, it was already good enough in Kekler. And Polaris had some improvements to DCC afaik.

>>61215042
>My uncle works at Nintendo.
>>>/v/

>>61211449
>take their time with the software and optimize it properly for the hardware
>hours and hours of testing, just for one architecture
>"giving a shit"
I have complete and utter understanding for a developer that doesn't spend his life optimizing for a single architecture when there are dozens and dozens out there, and most work perfectly well without such specific and time-consuming optimization.

>>61212522
>germanics
>non-synthetic
Nah brah, that's just English.

>>61211766
You don't even need devs to directly implement primitive shader code. You can wrap them around usual vertex code in the drivers.

>>61215132
english is germanic family language.

>>61215066
I think so too, but at the moment it isn't enough still, Polaris was 256-bit and 1060 was 192-bit? nVidia hasn't really needed to up their memory game until recently when they started being more aggressive in the HPC sector again. Shame that AMD missed their chance to penetrate the GPGPU market as they had the superior computer architecture this time around.
>>61215024
I don't know about that. I think nVidia is too far ahead in terms of R&D to let this go. However, nVidia also has significantly less scalable designs albeit more modular than AMD's. In addition at the moment AMD can't simultaneously offer HWS/ACE while doing IF without latencies which would put a whole shader engine out of sync. We'll see if this comes together in Navi though.

>>61215153
Yes, what I said was that other Germanic languages aren't as non-synthetic as English is.

>>61215142
what we know is that nvidia could do it, who the heck knows when amd will
it took them 3 years to make good drivers

>>61215169
They have working interconnect working in retail silicon.
Nvidia has some simulations. Also AMD was historically good at making good hardware and NVIDIA was good at making drivers.
>>61215163
We'll see it Vega brings any improvements to DCC.

>>61215163
>nVidia is too far ahead in terms of R&D to let this go
intel spent several billions to make ringbus into meshringbus
sure nvidia is better at tech than intel and sees trends a bit further, but amd already got it working

Also nvidia has ~zero experience in dealing with complicated NUMA setups, something that AMD was doing since forever.

>>61215196
amd had zero experience in SMT, see how that turned out

I seriously don't care who will be first, I want GPUs to just stop being such clusterfucks.

>>61215212
SMT isn't much different from CMT, anon.
Both put unused execution units to work

>>61215212
AMD had Keller, THE Alpha dude.

>>61215212
>amd had zero experience in SMT
Keller was on the team that essentially invented SMT.

>>61212294
You are of below average intelligence to say the least.

File: aa.png (3MB, 1920x1200px) Image search: [Google]

3MB, 1920x1200px

>>61210972
>Right now, with Vega, the following does not work:
>- Rapid Packed Match
>- 2 Ops per shader per clock cycle
>- Tile Based Rasterization
>- HBCC
>- Primitive Shaders
aside from the tile based rasterizing, I assume you have source for these, right?

>>61215554
Nvidia will be the first to successfully implement all of AMD's ideas.

>>61211850
AMD had a shareholder commitment to meet, and so they did it. Sucks, but that's how it goes. Angry shareholders > angry nerds.

>>61215554
>>- 2 Ops per shader per clock cycle
are you retarded? every GPU ALU in decades has been 2 ops/clock, since they can all do a multiply and an add separately.

e.g., Polaris 10: 2304 ALUs * ~1.25 Ghz * 2 ops/clock = ~5.8 TFLOPS.

Vega will not be any different in this regard, just in its ability to do 2x fp16 and 4x int8 throughput.

GDDR has a wait after an operation, i.e. high latency

HBM doesn't, which plays a large role in how HBCC is actually able to work. HBCC will play a large role in a MCM configuration

>>61216133
Why does their chart in the OP show 2x the clock speed and 2x the Ops? Is it that much better than Polaris 10 for real?

>>61216133
not sure why you're telling me this, he was claiming that was disabled and I'm just asking for sources.
also off topic, but to consider FMA as 2 ops is stupid, it's still a single instruction IMO. Just because you can do it in 2 separate ops doesn't mean that the ALU does 2 ops per cycle.

File: Vega-Final-Presentation-29.jpg (873KB, 2666x1500px) Image search: [Google]

873KB, 2666x1500px

>>61216308
please don't tell me you actually believe that this thing was some sort of fucking quantitative graph

>>61216351
sorry, I replied to wrong post without realizing.
other anon is the retarded one.

>>61216359
2x might be believable at 7nm, right?

>>61216484
No, 7nm won't clock anywhere near 2x for comparable designs, nor will a FMA unit ever do more than 1 multiple and 1 add of the maximum data widths per clock. The closest you will ever see is chips with double-clocked ALUs, but it's been a while since the last ones for good reasons (i.e., poor perf/W, which is the king of metrics nowadays).

>>61216351
>to consider FMA as 2 ops is stupid
Shit's been that way since forever, like the old Cray supercompute days and earlier. Blame marketers.

File: GF7nm.jpg (95KB, 1260x643px) Image search: [Google]

95KB, 1260x643px

>>61216541
Is this accurate?

>>61216351
>to consider FMA as 2 ops is stupid
that's the way it is tho

>>61216796
A hot and dense part like a GPU won't see a 40% boost, which is a theoretical ideal anyway. Maybe like 10-15%. Ryzen 2 ought to see a nice boost though.

>>61216834
nvidia reached around 30% uplift from maxwell 28nm to pascal 16nm. 30-35% can be done 14 to 7.

>>61194300
So, basically $5.

>>61216960
More like 40%