[a / b / c / d / e / f / g / gif / h / hr / k / m / o / p / r / s / t / u / v / vg / vr / w / wg] [i / ic] [r9k / s4s / vip / qa] [cm / hm / lgbt / y] [3 / aco / adv / an / asp / bant / biz / cgl / ck / co / diy / fa / fit / gd / hc / his / int / jp / lit / mlp / mu / n / news / out / po / pol / qst / sci / soc / sp / tg / toy / trv / tv / vp / wsg / wsr / x] [Settings] [Search] [Home]
Board
Settings Home
/g/ - Technology



Thread archived.
You cannot reply anymore.




File: 2nd gen R9 slides.png (450 KB, 640x720)
450 KB
450 KB PNG
friend who works at AMD sent me this. have fun Intlel...
>>
Legit, my dad works at Nintendo.
>>
Looks fake desu. Why would anyone want to buy ++ CPU (muffled 14++++++ in the distance), if 7nm are only 1/2 year away?
Although I see 2 possibilities
1. 7nm aren't for consumers, only for enterprises so far.
2. AMD just wants that halo of being top dog.
3. AMD wants to force Intel to do some another stupid move.
>>
I want a robe made out of infinity fabric
>>
>>68013526
\threads
>>
File: amdvega.jpg (156 KB, 633x758)
156 KB
156 KB JPG
>12-core slower in gaming than intel 8-core
>>
>>68013372
i can't see them releasing 2 cpus now and 7nm a few months later
>>
>>68013526
i don't think it matters that 7nm is close, it's still the same socket

amd can release whatever they like as long as that's true desu
>>
>>68013372
>Ryzen 9
>When they already have Threadripper.
Fake.
>>
>>68013372
>400/500 bucks
Nice try you falseflagging shintelkike. Go kys.
>>
>>68013798
that pic's obviously fake you tard
besides if you could read the chart, you'd realise that the supposed 10 core is 10% slower and 20% less expensive than the 9900k, not to mention that it's not pozzed
>>
>>68013372
>6 core ccx
Fake
>>
Glofo 7nm? I don't think that is the case. They are going TSMC aren't they?
>>
>>68014156
it says glofo 12lp, but yeah they are going tsmc for 7nm
>>
>AMD shills thinking that a 10/12-core Ryzen will work on old motherboards without throttling like hell
>there are AMD shills who actually believe they can run 10+ core CPUs on their garbage B350/B450 boards
(lol
>>
File: paidoing@urlife.jpg (69 KB, 839x471)
69 KB
69 KB JPG
>>68014209
>>68013984
>>
Anybody care to tell the price difference between 4cpu vs 2cpu vs 1cpu system price difference if you have the same core count?
>>
>>68014236
>>
>>68014209
>being this mad that AMD doesn't force mobo changes on people
lol
>>
>still not 5ghz

If amd released a 5ghz 8c proc I'll buy it immediately.

If I wanted shittons of slow ass cores I'll buy an off lease xeon off ebay.
>>
>>68013526
>1. 7nm aren't for consumers, only for enterprises so far.
This is obviously going to be the case for at least Q1 and Q2 of 2019, and will probably extend into Q3, and maybe even Q4.

But we COULD see consumer 7nm chips late 2019. Expect GPUs in spring 2020 after getting delayed from the current 2019 rumors.
>>
File: 1538664799932.png (1.1 MB, 2224x1369)
1.1 MB
1.1 MB PNG
>>68014396
>moar niggahurtz
t. 2002
>>
File: 1539275662201.jpg (423 KB, 641x2895)
423 KB
423 KB JPG
>>68014396
amd has higher ipc than intel now
>>
>>68014419
This, I got a solid 10-20% better fps in most games when I upgraded from 2133MHz to 3200MHz on my 2600, haven't messed with the timings yet though.
>>
>>68014593
based, intlel stuttershitters are seething
https://www.youtube.com/watch?v=YS8h7znjNPA
>>
File: 1517794023822.png (169 KB, 341x340)
169 KB
169 KB PNG
>>68013372
>STILL ON 12NM
>DOING THE SAME ++ SHIT AS INTEL
>EVEN MORE GLUED TOGETHER CCX

>b-but based AMD will bring us 7nm

AHAHAHAHAHAHAH

Never trust the eternal Poo
>>
>>68013960
No, this is what Jim Anderson wanted to release but Lisa Su said no
>>
>>68013372
Fake and gay, although 7nm zen will rape this shit coffeelake rebrand
>>
>>68013372
Cache numbers are extra wrong for your bait. 10cores would have 5MB of l2 so result would be an odd number on the 2800x. 2900x looks like it is 22MB of L3 which isn't going to divide by 12 or 2 easily, especially if it keeps Zen1 layouts.
>>
>>68015905
>intel stays on 14nm for 4 years noone bats an eye
>AMD stays on 12nm for 10 months AYYMMMD FINISHED AND BANKRUPT
>>
>>68013372
>Intlel-tier naming scheme
>Zen++
>12nm
>Intlel-tier pricing
Very well done Rakesh. Mr. Schlomo will be happy to pay you double the rupees for your out of the box shill mentality.
>>
>>68014410
Hasn't Navi been in production for PS5 for a few months already? They'd have to order a few more each month and that's it, basically no delay to develop it or anything
>>
>>68014410

Lolno.

Zen is a single design for the entire stack. Its the entire desktop and enterprise market is based entirely on binning practices. The absolute BEST binned dies go into all models of EPYC, then the 2nd best go into TR, third best into R7s, fourth best into R5s, fifth best into R3s. All dead dies act as balancers for the TR and EPYC socket because of the way IF works.

If AMD is already sampling 7nm EPYC to to their most trusted customers RIGHT NOW, it means that Q1 we'll see them talking about the Zen2 architecture changes and perf uplifts in IPC, cache optimizations, etc. Q2 we'll see them start talking about consumer release and latest Q2 end to early into Q3 we'd see Zen2 formally release into the consumer market.

There's no fucking way AMD is going to let Intel carve out market share with the 9900X/K series chips when they're 14nm+++++++++ nuclear reactors.

AMD is in the absolute best position to completely slaughter Intel, and they're going to capitalize on it. That's the beauty of making a CCX based single-die design that can scale up to the entire stack courtesy of the Infinity Fabric.

R9 looks fake though, even if the in the deepest of my heart I want it to be true.
>>
File: 1497779849086.jpg (863 KB, 1906x1536)
863 KB
863 KB JPG
>>68013372
>friend
>works
on /g/, obviously fake
>>
>>68013372
>6 core ccx
will never happen. there is a logical reason why amd went 4 core. 4 core per ccx is absolute max. 4 cores means 6 total links. 8 cores is 28 links. way to many on a practical scale and more importantly cost for production. its the WHOLE REASON why amd made ccx. unless you really want amd to toss in a ring bus to add more latency to the game. the point of ccx was not to make big monolithic cores.
>>
File: 1497878953615.jpg (71 KB, 552x661)
71 KB
71 KB JPG
>>68013372
P-PLEASE BE REAL!!!!
>>
>>68017528
number of links per core
>1 core = 0 links total
>2 core = 1 links total
>3 core = 3 links total
>4 core = 6 links total
>5 core = 10 links total
>6 core = 15 links total
>7 core = 21 links total
>8 core = 28 links total
this is why intel adopted ring bus and now mesh. this is why amd went faildozer and now ryzen with ccx. ways to make multi-core chips without needing large dies and complex manufacturing. we will never see more than 4 cores on a zeppelin die. no in ryzen nor second generation ryzen coming out next year in 2019 on 7nm.

8 core will be the max for mainstream. 32 core will be the max for epyc. unless amd can fit another whole die or two on a single substrate. meaning, more than 4 dies on a substrate. amd will never go beyond two dies on a single substrate for mainstream. if you want that they have x399 and its successor. threadripper if you need more than 8 cores.
>>
>>68013372
>unlocked infinity fabric multiplier
Great, no more need to sell a kidney for RAM that won't bottleneck ryzen.
>>
>>68017612
Or you know, more than 2 CCXes per die.
>>
>>68017612
>this is why intel adopted ring bus and now mesh. this is why amd went faildozer and now ryzen with ccx. ways to make multi-core chips without needing large dies and complex manufacturing.
Just slap on more than 2 CCX per die if you want more cores per die, or simply cram more dies into a chip like on Threadripper/EPYC.

>8 core will be the max for mainstream. 32 core will be the max for epyc.
What a load of bullshit.

https://wccftech.com/amd-epyc-rome-7nm-64-core-cpu-performance-benchmark-leak/
https://www.notebookcheck.net/Chiphell-leaks-another-apparent-AMD-Rome-Cinebench-score-this-time-with-some-proof.337356.0.html

Rome is 64 cores.
>>
File: 1539077767489.jpg (37 KB, 460x608)
37 KB
37 KB JPG
>>68013372
other sources have been saying they will be rereleasing a upgraded 2800.
makes sense to me.
i believe it.
>>
>>68015905
It's a fake, retard.
>>
>>68017690
unless amd is able to achieve equally as high yields as they do with two ccx's on a single die that won't happen for mainstream. 1 die on a substrate with two ccx's on that single die for mainstream will be max most likely. amd needs high yields for max profit. adding more ccx's will do nothing but add more to cost and lower yields when 8 cores 16 threads is competitive enough for the next foreseeable future. let software developers catch up first. especially when there is threadripper available for those pro-consumers in the mainstream market that need more.

amd might do it for epyc if there is enough demand as that's worth the extra cost as epyc chips have much higher profit margin. but mainstream? even for threadripper i can't see them doing that. especially considering amd has no issue simply slapping on another die to a substrate to make 12 - 16 - 24 - 32 core threadrippers for that market.
>>
File: npc.png (45 KB, 330x342)
45 KB
45 KB PNG
>>68015905
>falling for low quality bait
>>
>>68014371
They where hitting 6Ghz and above on that shit though.
>>
>>68018144
I agree with you that 4 core CCX and 2x CCX per die will say. However, we know that this is easily scalable by simply adding more dies to a chip.

While Intel's mesh architecture scales far better than the ring bus, it is still problematic when you begin to scale to very high core counts, because core communication induces higher and higher latencies the more hops a packet has to make. That makes latencies both unpredictable and potentially very high for a large amount of cores.

The AMD Infinity Fabric doesn't need to make hops. Each core is directly connected to each other within a CCX, and each CCX on the same die is directly connected to each other. This induces complexity if you were to try and put 8 cores in a single CCX, or more CCX on a single die, but it also allows you to not do that at all and simply add more dies on the same chip. On multi-die chips (like all of their chips currently) each die is then directly connected to each other. This makes latencies completely predictable and is extremely scalable.
>>
>>68013372
>loses at gaming, the only one that matters and sells units
So close yet so far AMD.
>>
File: intlel btfo.png (21 KB, 600x230)
21 KB
21 KB PNG
>>68018463
>ONLY GAMING MATTERS
lmaoing @ your life

Look at this EPYC obliterating TWO Incel chips that have an 8 core advantage and 16 more threads over the 32c/64t EPYC.
>>
>>68017612
Yes and no, There a great reseach paper by AMD on the the performance impact of adding a active substrate over a passive substrate.
>>
File: 6850kmemory1le.png (169 KB, 539x538)
169 KB
169 KB PNG
>>68018463
>loses at gaming, the only one that matters and sells units
you know what grinds my gears? is when people say ryzen sucks at gaming they don't realize they're telling people who have broadwell that their cpu's suck at gaming. since you know, clock for clock, ryzen and broadwell are virtually identical. hell it even applies first gen skylake sucks cock since skylake didn't have high clocks like kaby and beyond.

plus that also implies all those haslel's like the 4790k suck too since it isn't to far behind broadlel.
>>
>>68013372
Fake. The power usage would require an HEDT board.

These are thread ripper core counts, they wouldn't work on an AM4 socket mobo and AMD already committed to supporting AM4 socket motherboards until at least 2020.
>>
>>68019364
Even moreso when you consider that most consumers on steam for example have quad cores.
Most likely those are haswell area of chips.

Also now after all the dust has settled AMD is within 5-10% of single core performance, thats not going to be a gigantic difference in games.
Hell with the new ram speed tests with faster timings and whatnot its even closer.

I have a 1600x at 4gz and I did see a noticeable difference between that and my previous 4670k at stock speed, sure i could have overclocked my intel core but me personally, im 100% done with intels bullshit and i wanted to vote with my wallet.
>>
>>68014236
Why doesn't it show 2700X with 2666 memory?

Oh wait, that's the one in orange...
>>
>>68014417
>GPU bottleneck
>>
>>68014593
This applies to first generation ryzen too, well into the 15%+ IPC gain on my chip when i went from 2133 to 3200, havent tested timings tho.
>>
>>68013372
>infinity fabric
>gaming
lol
>>
>>68014037
It's not fake. Chips with infinity fabric will be slower in gaming as shown by 1950X/2950X/2990WX
>>
>>68013372
You still haven't fixed the strange cache and clock speeds I recommended when you posted this fake yesterday. You know, you're not under any requirement to make your shit fake AND gay - it's just a saying.
>>
File: 98797987987.png (8 KB, 363x364)
8 KB
8 KB PNG
>>68019583
>267MHz results in a 20% performance bump
>>
>>68019583
The 2666/2933 benchmarks seem to be intended to show performance when using stock specs - 2933 is the supported maximum RAM speed for the 2700X (and other Pinnacle Ridge CPUs) and 2666 is the supported max for the 8700K (and other Covfefe Lake i5/i7 CPUs).





Delete Post: [File Only] Style:
[Disable Mobile View / Use Desktop Site]

[Enable Mobile View / Use Mobile Site]

All trademarks and copyrights on this page are owned by their respective parties. Images uploaded are the responsibility of the Poster. Comments are owned by the Poster.