nVidia Ampere

[Winjer]

NVIDIA Ditches Intel Xeon, Goes All Onboard With AMD’s EPYC CPUs With Next-Gen Ampere GPUs

In a surprise unveil, NVIDIA has decided to ditch Intel's Xeon CPUs and go all onboard with AMD's EPYC Rome processors to power its next-generation Ampere GPU powered DGX systems.
This will be the first full outing of two next-generation 7nm server products in the same platform featuring insane hardware specifications.

NVIDIA 7nm Ampere GPUs Paired With AMD's 7nm EPYC Rome CPUs In Next-Generation DGX A100 Supercomputing System
The news announcement comes on the launch of NVIDIA's Ampere GPU which is by far the biggest 7nm GPU ever built.
In the presser, NVIDIA revealed full technical specifications of its latest DGX-A100 system that will be utilizing NVIDIA's 7nm Ampere GPUs and AMD's 7nm EPYC Rome CPUs,
marking the first time that green team has opted for AMD's server chips for use in its flagship HPC platform.

CPUs da AMD com GPU da nVidia. Parece que voltamos à super equipa de 2003

[LPC]

NVIDIA Ditches Intel Xeon, Goes All Onboard With AMD’s EPYC CPUs With Next-Gen Ampere GPUs



CPUs da AMD com GPU da nVidia. Parece que voltamos à super equipa de 2003

NForce 2 powa!

A Nvidia sabe que a AMD está mais forte nos CPU´s...

No entanto do que já vi penso que vai ser mais uma ronda de sucesso para a Nvidia com o Ampere...
Visto que agora o Gsync é meio Agnóstico, devo meter uma 3080ti (se não custar os olhos da cara).
Quero garantir os 4k 60fps com tudo no prego...

Cumprimentos,

LPC

[Enzo]

NVIDIA Ditches Intel Xeon, Goes All Onboard With AMD’s EPYC CPUs With Next-Gen Ampere GPUs



CPUs da AMD com GPU da nVidia. Parece que voltamos à super equipa de 2003

Neste caso pode ficar batizado desde já como Killer 7 🤓😎

[Winjer]

NVIDIA Ampere Powered GeForce RTX 30 Series Reportedly Launching With RTX 3080 & RTX 3090 In September

[dblaster]

Na 3090 não vamos outra vez ter problemas de multi gpu ou isto tende a resolver-se com updates de api. Ou será que é para aumentar a performance em Ray tracing?

[LPC]

Na 3090 não vamos outra vez ter problemas de multi gpu ou isto tende a resolver-se com updates de api. Ou será que é para aumentar a performance em Ray tracing?

Boas!
Não é garantido que a 3090 seja de facto multi gpu. A meu ver será a nova designação das "TI", sendo que o Super passa a ser as "especiais" vamos chamar de refresh de meia vida...

Cumprimentos,

LPC

[Winjer]

Mais uns rumores para deitar sal:

350 watts for NVIDIA’s new top-of-the-line GeForce RTX “3090” Ampere model explained, chip area calculated and boards compared

Well, meanwhile there are several leaks of “pre-release” models of the upcoming GeForce RTX 3080, but I don’t really trust the roast published here, because I just assume design validation samples. And before I think about how to explain the 350 watts to you, I have created a simple super projection (and cut it a bit on the right side for clarity), which of course has no artistic value whatsoever like several render images, but raises very legitimate questions. I had reported that there are two boards, of which the PG133 is used exclusively for the Founders Edition (FE) and the PG132, which will be the reference for the board partners.
Leaked cooler vs. PG132 and the chip size of the GA102
If one now takes the experiences of the last few years as a guide, places the memory closer to the BGA according to the usual memory manufacturer’s specifications for the tracks and also groups the memory around the BGA for the first time in the only plausible 4-3-4(+1) scheme, then there would hardly be any space left for the display connections on the cooler shown on the Internet, let alone for the necessary voltage converters. The large Turing cards already use a double-sided voltage converter layout, which makes absolute sense considering the hotspots.
If one assumes that for electrical reasons certain minimum distances between the boards and even the individual components must be kept, the distribution based on the leaked cooler would not be possible at all, unless the PG133 of the Founders Edition would completely deviate from the “alternative” reference for the first time. This is of course quite possible and would also give the FE a certain exclusivity, but it would also be an extreme cost driver. All this is very well possible.


My projection, which I have created from memory, should, however, in terms of PG132 (and only this one!) after deducting all possible inaccuracies, be pretty much the same as what Nvidia has already provided to the board partners as a Thermal Design Guide. These things almost always resemble each other, and also the circulated 21.3 cm length falls exactly into this scheme. But I would never publish such originals anyway, as traffic-bringing as that may be. It’s just not appropriate, even if you haven’t signed an NDA for it.
I have placed the chip in the drawing correctly, at least in relation to the board size. TechPowerUp writes on the homepage in the database entry for the GA102 about 700 mm² and is not even very wrong. If you add up all the facts you already know, then with a little bit of calculator and additional information from beloved sources you get a chip area of just over 627 mm², which should be correct.
It is also a fact that there is enough space for the two power supply connections I predicted, because everything that needs to be cooled is first of all supplied in form of electrical energy. Which brings us to the power consumption. And this is exactly where we want to calculate together.
How 230 watts become 350 watts

The professional has two terms at hand when it comes to specifying the total power consumption of a graphics card. Besides TBP as Total Board Power, there is also TGP as Total Graphics Power from Nvidia’s point of view. It deviates a bit in small details (TGP without 3.3 Volt), but at least it still comes up quite well in terms of the orders of magnitude. Do you remember my article “Power Consumption: TDP, TBP and TGP for Nvidia and AMD graphics cards calculated with destruction of a PR foil” ? You should read it again for better understanding.
Some people in forums have been hostile to my estimation (ok, it was rather a nice bird’s twitter) with the 350 watts for the large and the 320 watts power dissipation for the two smaller cards, but I don’t want to comment on the reasons for the lack of understanding. But power comes from fuel and even without “songbirds” the power consumption could be estimated quite well.
So I’ll stick with the 350 to 320 watts and want to calculate down how much could still be allotted to the chip. And this is where it gets interesting, because you can extrapolate voltage transformer losses, low voltages and other things quite well based on the last generations. Well, I asked explicitly for the memory, I am not a clairvoyant.

[LPC]

Boas!
Vão vir mesmo em boa altura então para o Inverno!

Cumprimentos,

LPC

[Winjer]

Nvidia passes Intel as the most valuable US chipmaker

Thanks to a recent surge in its stock price, Nvidia has overtaken Intel to become the most valuable chipmaker in the US. Following a 2.3 percent jump yesterday afternoon, team green’s shares have now reached $408, putting its market cap at $251.3 billion, ahead of Intel’s $248.1 billion.

[Winjer]

GeForce RTX 2070 Super discontinued, GeForce Ampere RTX 300 Series Launch details

Asia based mydrivers is relaying word about a number of discontinued GeForce RTX card, as well as sharing some launch date information on the upcoming Ampere, GeForce RTX 3000 series graphics cards.
Ampere is approaching soon. NVIDIA is stimulating the discontinuation of the RTX 20 series. The high-end RTX 2080 Ti and RTX 2080 Super had been discontinued before, and now the RTX 2000 range is further extended downwards. According to the latest news, the RTX 2070 Super series has been completely discontinued, and the remaining last batch of stock chips has been dispatched. It would not be restocked, and no new orders will be accepted. The market demand for the RTX 2060 series and GTX 1660 series in the mid-end mainstream market is still strong, especially the recent mining market has clearly risen. It is said that the shipment prices of RTX 2060 and GTX 1660 Super chips have increased by 15-20 US dollars and 5-8 yuan respectively, and the factory prices of various models of the RTX 2060 and GTX 1660 series will generally increase by 30-40 yuan.

Ampere-based NVIDIA RTX 3080 Ti and RTX 3080 graphics cards would officially be released on September 17, the RTX 3070 and RTX 3060 series will follow up in October and November.

[LPC]

Boas!
Vamos lá a ver o que vêm por ai...

Espero que seja um bom salto em performance...

Cumprimentos,

LPC

[SleepyFilipy]

Eu não espero nada a mais do que já vem sendo regra. Novo topo de gama com 25% de performance acima do topo anterior... Se fizerem mais que isso vou ficar positivamente surpreendido, mas para já é com o que conto.

[Winjer]

O que eles têm de melhorar é o desempenho com rtx. É muito mau ligar o rtx num jogo e o desempenho cair para metade.
O desempenho em renderização tradicional até pode aumentar apenas em 25%, mas o desempenho em rtx tem de duplicar.

[LPC]

O que eles têm de melhorar é o desempenho com rtx. É muito mau ligar o rtx num jogo e o desempenho cair para metade.
O desempenho em renderização tradicional até pode aumentar apenas em 25%, mas o desempenho em rtx tem de duplicar.

No mínimo...

Cumprimentos,

LPC

[Winjer]

No mínimo...