2016 nMP

[Mago]

I was thinking a similar solution that some better RAID cards have had for ages; for host OS it is just an i/o card with one or more logical discs. The card itself is run on an independent processor with its own ram that takes care of the RAID stripping etc.

As detailed up, a desktop computer you may think it's only CPU is the one you care of (a Xeon a i7 etc), there are numerous SLAVE subsystems that actually have their own dedicated CPU, just one as smaller as it need to be: Lan, Storage, RTC, TPM and any other smart controller has one or other kind of dedicated CPU or logic circuit offloading processing or functions otherwise will account on CPU cycles.

Apple's security enclave its an good example.

To put ARM processor just for TouchID would be a waste of a good CPU... So why not give it other tasks. It could be a full feature security processor with touchID, encryption and disc i/o all in one. IT could even compress files on the fly without consuming any time from main CPU. And because it would be OS independent, it would be difficult to hack.

In fact touchID on the iPhone relies on its very own CPU and is not the same as security enclave.

Not all ARM processors are aimed at complex devices, in fact most of them barely are processors, an good example it's the very popular maybe unknown Atmel SAM3X8E Cortex-M3 the one that powers every Arduino Due, yes that simple controllers runs on ARM, you have an ARM CPU in every flavor for every need from slave controllers (even clocks) to super computer hexascale clusters.

 

[pat500000]

Hmm shouldn't we be hearing some kind of rumors for Mac Pro ?

 

[koyoot]

http://vrworld.com/2016/04/25/16nm-nvidia-geforce-gtx-1080-leak-ahead-computex-taipei-2016/

While the specs of GP104 are still hidden, we can now say that the chip packs 1920 CUDA Cores and a 256-bit controller supports both GDDR5 and GDDR5X (MSI uses GDDR5X) memory.

Hmmm. 3 GPC, with 640 cores in each. 7.14 TFLOPs of compute power.

 

[Stacc]

Hmmm. 1.86 GHz Pascal GPU is 4% faster than the fastest GTX 980 Ti. It would indicate that it is 2048 CUDA core GPU, not 2560. Also the compute performance delta between the two GPUs appears to be correct(Waterforce GTX 980 Ti - 7.558 TFLOPs, GTX 1080 - 7.618 TFLOPs).

Interesting. I bet they price it fairly aggressively. $500 for the 1080 and $400 for the 1080. Just like the 980 and the 970 when they launched.

Benchmarks are starting to leak out for the upcoming Broadwell-E processors. I found some up on geekbench which gives us a first glimpse at the expected performance.

Intel Core i7-6850K, 6 cores at 3.6 Ghz: Single core: 3896, Multi-core: 22576
Mac Pro, 6 cores at 3.5 Ghz: Single core: 3158, Multi-core: 18495

Intel Core i7-6950X, 10 cores at 3 Ghz: Single core: 3320, Multi-core: 30467
Mac Pro, 12 cores at 2.7 Ghz: Single core: 3216, Multi-core: 33148

And just for reference

Late 2015 iMac, 4 cores at 4 Ghz: Single core: 3942, Multi-core: 15615

What will we get with Broadwell-E(P)? Should be a modest bump in single threaded and multithreaded performance across the board.

 

[Mago]

What will we get with Broadwell-E(P)? Should be a modest bump in single threaded and multithreaded performance across the board.

From Ivi bridge EP to Broadwell-EP raw performance gains are minimal,specially in single thread, actual big gains comes from increased cores availability at same TDP, much higher memory bandwidth (some apps benefits more on this than on raw processing speed ), but the most important performance related gain comes from AVX 2 extensions not available on IviBridge.

Don't care in raw performance the new Xeon B-ep beat by long the former AS LONG THE APP ITS OPTIMIZED FOR AVX 2 OR BENEFITS FROM FASTER RAM OR MULTIPLE CORES ALSO ASUMING APPLE SELL US THE XEONS ACCORDING ITS COST AND TDP (MORE CORES AT SAME PRICE).

 

[fuchsdh]

Hmm shouldn't we be hearing some kind of rumors for Mac Pro ?

Mac Pro isn't that interesting to the wider world, so the chances are someone leaks something decreases (I would be surprised if one of those analysts would even ask their "trusted sources" about anything beyond "new Macs" that would involve the Pro.)

Not to mention that with the Mac Pro being assembled in the USA, there's no rich Chinese pipeline of leakers to funnel information to the press.

 

[Mago]

On Ivi-Bridge-EP vs Broadwell-EP performance gains:

As I said on my previos post, it depebnds on the Apps being optimized to take advantage of AVX2.

YCruncher its an PI calculation torture test, on AVX2 outperfoms AVX 2:1 thats not trivial

source http://www.numberworld.org/y-cruncher/internals/arch-optimizations.html

 

[t0mat0]

Hmm shouldn't we be hearing some kind of rumors for Mac Pro ?

Nope - see the rumours for the nMP - Basically Apple themselves fanned those flames to show that a refresh was happening at all. Pretty much nothing known till the keynote.

 

[ManuelGomes]

Tomorrow we'll know all I guess, it's confirmed that they'll detail the cards.
I wonder if AMD will also make some sort of announcement? I guess not, I believe they'll wait.
Whoever is about to buy a new card will want to wait to see what AMD has up their sleeves.
The new NVidia coolers seem a little flashy to me, but maybe it pleases those who like to pimp their rigs I guess.
Don't forget to tune in tomorrow...
[doublepost=1462479277][/doublepost]I guess NVidia employees must have been working overtime to beat AMD to the punch with earlier launches.
It seems clear they wanted to come out first with every new stuff, be it HBM2 cards (paper launch) and now mainstream cards probably with GDDR5X.
Do I sense some respect/fear for what's coming from AMD?
AMD might even not come through but at least it seems they hit a nerve somewhere.

 

[tralfaz]

Nope - see the rumours for the nMP - Basically Apple themselves fanned those flames to show that a refresh was happening at all. Pretty much nothing known till the keynote.

And possibly no news even after WWDC. I give it a 50% chance that a Mac Pro update even gets a mention. I think a MBP update has an 80% chance for an update. Any other Mac less than 30%.

 

[Mago]

And possibly no news even after WWDC. I give it a 50% chance that a Mac Pro update even gets a mention. I think a MBP update has an 80% chance for an update. Any other Mac less than 30%.

Actually we have an solid leak about the nMP update and will come along an VR-Focused FCP (check the 89 pages to find it), so we are 90% sure there is an nMP at WWDC, 50% on new Macbook retina at 14/16"and actually only cues about new Macbook pro is from AMD Polaris supposedly selected by apple for some notebooks and desktops.
0 cues the Mac mini, mac server or ARM macs

 

[wallysb01]

Ah, I have a new fav phrase "next time I have an Idea like this, punch me in my face - Tyrion Lannister"

I'm totally stealing a different Tyrion Lannister quote, "That's what I do: I drink and I know things."

 

[koyoot]

Tomorrow we'll know all I guess, it's confirmed that they'll detail the cards.
I wonder if AMD will also make some sort of announcement? I guess not, I believe they'll wait.
Whoever is about to buy a new card will want to wait to see what AMD has up their sleeves.
The new NVidia coolers seem a little flashy to me, but maybe it pleases those who like to pimp their rigs I guess.
Don't forget to tune in tomorrow...
I guess NVidia employees must have been working overtime to beat AMD to the punch with earlier launches.
It seems clear they wanted to come out first with every new stuff, be it HBM2 cards (paper launch) and now mainstream cards probably with GDDR5X.
Do I sense some respect/fear for what's coming from AMD?
AMD might even not come through but at least it seems they hit a nerve somewhere.

In previous years, Nvidia was able to control a bit what AMD offered, and was able to bring more powerful GPUs after some time, because there is something called "technology spying". AMD and Nvidia both used TSMC, as a manufacturer of their hardware. They both were spying each other, to know the performance, and to counter it(GTX 980 Ti vs Fury X, or GTX 680 vs HD 7970).

Currently, because AMD uses GloFo technology to manufacture their hardware, Nvidia is not able to spy it. That is why they went all in with Pascal. What is even funnier in the context is that AMD did not revealed the setting from presentation demo of P10 at Capsaicin Event to counter calculating the performance of the GPU. Both of demos were showing efficiency of the GPUs, rather than raw performance.

 

[ManuelGomes]

Just a few hours to go...

Big day today!!!

 

[p.l]

Just a few hours to go...

Big day today!!!

Where can i follow along?

 

[Zarniwoop]

I am not gonna stay awake tonight... Nvidia is showing the finger to Europeans. Live broadcast starts on Saturday morning at 3AM CET?? What kind of business time is that!?

Good timing for the vast audience of the Pacific in Hawaii and Tahiti. Those who don't live somewhere in the American timezones, it'll be early morning. Let's say, to 85% of the population.

http://www.timeanddate.com/worldclock/fixedtime.html?p1=224&iso=20160506T18&msg=NVIDIA Livestream on Twitch.TV/NVIDIA

I think Nvidia choose this time, because all the Stock markets of the world will be closed for next two days.. what are they afraid of? Short selling anyone?

 

[Mago]

I'm at eastcoast tz, i'll try to see it.

The time is good for stock markets (most closed, except few electronic se that exchange 24x7x365, this gives time to nVidia to patch any misunderstood before their stock paid.

Which means the launch won't say anything, or they offerings will be non traditional.

 

[koyoot]

Well, they may be afraid of the reaction to paper launching of their GPUs.

Most likely Nvidia will announce that they will go on sale in middle of June, but nobody will be able to buy them, until early July.

 

[Zarniwoop]

According to this rumor, Zen die is going to be 8-core only with or without "HT". Depending on the quality, all 6- and 4-cores are based on this die.

http://www.bitsandchips.it/52-english-news/6955-no-dual-and-four-cores-zen-cpus-at-least-initially

 

[MacVidCards]

Well, they may be afraid of the reaction to paper launching of their GPUs.

Most likely Nvidia will announce that they will go on sale in middle of June, but nobody will be able to buy them, until early July.

Should they have used the AMD "over promise, under deliver" method?

Just check the stock, up so far. Guess the market isn't as easily rattled as the AMD PR department.

But nice try.

 

[koyoot]

According to this rumor, Zen die is going to be 8-core only with or without "HT". Depending on the quality, all 6- and 4-cores are based on this die.

http://www.bitsandchips.it/52-english-news/6955-no-dual-and-four-cores-zen-cpus-at-least-initially

Time will only tell. So far Forni was right about Pascal launch schedule, was right about two versions of the same die(GDDR5X, and GDDR5), was wrong however about the naming both GP104 chips. Was also right about the die size of GP100.
http://www.bitsandchips.it/english/...ill-have-2-pcbs-base-gddr5-and-premium-gddr5x
http://www.bitsandchips.it/english/52-english-news/6620-rumor-nvidia-pascal-timeline
http://www.bitsandchips.it/english/...r-tsmc-will-produce-the-huge-gp100-interposer
It was also him who said that the GP104 GPUs will come in early Q3. So far, even Qualification Silicon of that die dated for 14th week of 2016 shows this. http://www.bitsandchips.it/52-engli...egy-reveals-something-about-pascal-and-parker

Who knows, maybe he is also right about AMD Zen?

 

[ManuelGomes]

Man, this guy only steps in when someone makes a remark about NVidia.
Are you for real?

 

[linuxcooldude]

Should they have used the AMD "over promise, under deliver" method?

Just check the stock, up so far. Guess the market isn't as easily rattled as the AMD PR department.

But nice try.

Most likely relying on their market share to pull them through. Since people using Nvidia & third party software thats stuck on using CUDA and Gameworks.

 

[ManuelGomes]

I'm curious but I ain't waiting to 2am for this.
Tomorrow I'll catch up when all details are out.
After that we'll all wait for details on Polaris to see how they stack up against each other. What a rush...
Are we feeling like kids again waiting for the next big thing? Too bad if it's not that big :-(

 

[Mago]

I'm curious but I ain't waiting to 2am for this.
Tomorrow I'll catch up when all details are out.
After that we'll all wait for details on Polaris to see how they stack up against each other. What a rush...
Are we feeling like kids again waiting for the next big thing? Too bad if it's not that big :-(

P100 give us all the juice, GP106 & GP104 are just scaled down versions, it's like Chevy launching a Camaro V8 first then V6 versions, only intriguing on gp106 & 104 is about its fp32:fp64 ratio and if it will offer same SLI or an nvLink souped up SLI, ahh and it's pricing, E.E. was revealed on P100 launch with enough details as to expect any surprises.