• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

GTC 2022 Keynote with NVIDIA CEO Jensen Huang

Purgatory

Purgatory

Purgatory

Soldato
Joined
7 Dec 2010
Posts
8,698
Location
Leeds

Today at GTC 2022 NVIDIA CEO Jensen Huang presents the latest breakthrough in AI and data science. He also introduces next-gen 4nm-based Hopper architecture, offering 3X higher performance than A100 in raw compute power.
 
0mnDqYR.jpg


hIM0pOC.jpg


Kuues2n.jpg


kDs47d8.jpg
 
I think the real big news is all the leaks were fake and wrong.. wasn't this meant to be a MCM design by the great leakers :cry:... Proving now the so called respected leakers are BS artists and Lovelace leaked "facts" will be wrong too. BUT of course they will pretend they were right and Nvidia made some changes lol or they will twist their "facts" to make it look like they were right.


Hopper looks good but again the power use and the Nvidia special benchmarks/improvements they showed are well to be taken with a grain of salt till they are in the hands of a reviewers. So far what I see is what was really expected from Nvidia so again same old same old Nvidia. Lovelace ..20-30%+ on high end and lower down the stack as per usual. Just they will again flash big numbers and him screaming out our biggest gains ever as always and in real world gains the usual. Honestly the Hopper chips are great and the extra bandwidth and NVLINK updated too, BUT as expected really so nothing really massive change wise.

Also these will cost a small fortune .. look at the power components on the boards and the amount of phases required, they are double what ampere needed. :eek:
 
Last edited:
Purgatory said:
I think the real big news is all the leaks were fake and wrong.. wasn't this meant to be a MCM design by they great leakers :cry:...
Click to expand...

TBH, the "usual suspects" have been back-peddling from MCM for a few weeks now. It's now supposed to be the Hopper follow-up that's MCM. Guess it will have to be if the current rumours about RDNA3 being MCM are true. Then again, those rumours might also be utter BS..
 
Purgatory said:
I think the real big news is all the leaks were fake and wrong.. wasn't this meant to be a MCM design by they great leakers :cry:... Proving now the so called respected leakers are BS artists and Lovelace leaked "facts" will be wrong too. BUT of course they will pretend they were right and Nvidia made some changes lol or they will twist their "facts" to make it look like they were right.


Hopper looks good but again the power use and the Nvidia special benchmarks/improvements they showed are well to be taken with a grain of salt till they are in the hands of a reviewers. So far what I see is what was really expected from Nvidia so again same old same old Nvidia. Lovelace ..20-30%+ on high end and lower down the stack as per usual. Just they will again flash big numbers and him screaming out our biggest gains ever as always and in real world gains the usual. Honestly the Hopper chips are great and the extra bandwidth and NVLINK updated too, BUT as expected really so nothing really massive change wise.

Also these will cost a small fortune .. look at the power components on the boards and the amount of phases required, they are double what ampere needed. :eek:
Click to expand...

Most likely we'll see it coming before MCM designs become a reality as there will likely be OS updates, etc. to provide feature support a good time in advance.
 
Purgatory said:
I think the real big news is all the leaks were fake and wrong.. wasn't this meant to be a MCM design by the great leakers :cry:... Proving now the so called respected leakers are BS artists and Lovelace leaked "facts" will be wrong too. BUT of course they will pretend they were right and Nvidia made some changes lol or they will twist their "facts" to make it look like they were right.


Hopper looks good but again the power use and the Nvidia special benchmarks/improvements they showed are well to be taken with a grain of salt till they are in the hands of a reviewers. So far what I see is what was really expected from Nvidia so again same old same old Nvidia. Lovelace ..20-30%+ on high end and lower down the stack as per usual. Just they will again flash big numbers and him screaming out our biggest gains ever as always and in real world gains the usual. Honestly the Hopper chips are great and the extra bandwidth and NVLINK updated too, BUT as expected really so nothing really massive change wise.

Also these will cost a small fortune .. look at the power components on the boards and the amount of phases required, they are double what ampere needed. :eek:
Click to expand...


They were not wrong, you just have to look at what was shown and draw the missing links it also helps if you understand how Nvidia works.

When making a new GPU Nvidia has dozens of teams all working on an individual part of the end product and they work in virtual highly controlled environments, these teams know very little of what each other is working on - only management know and have privy to how it pieces together. The leaks come from a lower down employee in a single team, he only has access to what he is doing and then he leaks it, or another leaks theirs etc so all the Twitter leakers have is small tidbits of information- some of thisnis vague and they infer from it

for example the reason they said it's MCM is because an Nvidia employee said he saw two dies on a PCB that's all; now that Nvidia has announced the GPu we know why - one of the HP100 server cards does HAVE TWO DIES ON THE PCB but one of the dies is Nvidias first ever CPU and the employee did not know it's a CPU because that was not his team



this means that the leakers will never be 100% correct when it comes to Nvidia GPUs - being 100% correct would require a person from every engineering team in Nvidia to leak info of the same product
 
Last edited:
Grim5 said:
They were not wrong, you just have to look at what was shown and draw the missing links it also helps if you understand how Nvidia works.

When making a new GPU Nvidia has dozens of teams all working on an individual part of the end product and they work in virtual highly controlled environments, these teams know very little of what each other is working on - only management know and have privy to how it pieces together. The leaks come from a lower down employee in a single team, he only has access to what he is doing and then he leaks it, or another leaks theirs etc so all the Twitter leakers have is small tidbits of information- some of thisnis vague and they infer from it

for example the reason they said it's MCM is because an Nvidia employee said he saw two dies on a PCB that's all; now that Nvidia has announced the GPu we know why - one of the HP100 server cards does HAVE TWO DIES ON THE PCB but one of the dies is Nvidias first ever CPU and the employee did not know it's a CPU because that was not his team



this means that the leakers will never be 100% correct when it comes to Nvidia GPUs - being 100% correct would require a person from every engineering team in Nvidia to leak info of the same product
Click to expand...



That's not how MCM works that's a dual GPU card.
 
NVIDIA RTX A5500 with RTX 3080 Ti specs
Alongside H100 AI accelerator, NVIDIA is today announcing new workstation GPUs for desktops and laptops.

The A5500 is a desktop workstation graphics card equipped with GA102 GPU and 10240 CUDA cores. This configuration matches the RTX 3080 Ti gaming GPU specs, but there is a very major change: the workstation model features GDDR6 memory with EEC whereas the gaming GPU supports GDDR6X. NVIDIA has not confirmed the exact specs, such as clock speeds or TDP for this SKU.



Moreover, NVIDIA is introducing as many as six new laptop workstation GPUs, including RTX A5500 (Laptop) flagship model clearly using the GA103 GPU with 7424 cores and 16GB of GDDR6 memory. This further confirms that NVIDIA is doing an RTX 3080 Ti desktop and laptop relaunch, but for workstations.

Among new parts there is RTX A500, the slowest Ampere laptop GPU released thus far, featuring 2048 CUDA cores and just 4GB of memory. Other models such as A1000, A2000 and A3000 will feature 4GB, 8GB and 12GB respectively, while RTX A4500 will be using 16GB memory and cut-down GA104 GPU with 5888 CUDA cores.


u26d3Si.jpg



OAcC73n.jpg



kDze8W6.jpg


https://videocardz.com/newz/nvidia-...-with-rtx-3080-ti-specs-and-6-new-laptop-skus
 
You must log in or register to reply here.
Back
Top Bottom