wccftech NVIDIA RTX 6000 Ada Workstation Graphics Card With Over 18K Cores & 48 GB Memory Hits Retail For $6800 US

[bonami2]

  Quote

The specifications of the NVIDIA RTX 6000 are also better than the GeForce RTX 4090. It has just 2 SMs disabled at 142 which offer up to 18,176 CUDA cores. That's 11% more cores & SMs than the RTX 4090 GPU. But NVIDIA surely has some room left for an even higher-end variant that will use those 2 remaining SM units. The card comes with a similar boost clock of 2.5 GHz and has double the VRAM capacity at 48 GB (EEC) versus the 24 GB featured on the RTX 4090. The card does use slower 20 Gbps GDDR6X memory dies versus the 21 Gbps dies featured on the gaming graphics card.

Expand  

NVIDIA RTX 6000 Ada Workstation Graphics Card With Over 18K Cores & 48 GB Memory Hits Retail For $6800 US

WCCFTECH.COM

---

Today, NVIDIA announces the retail availability of its flagship workstation graphics card based on the Ada architecture, the RTX...

 

 

https://store.nvidia.com/en-us/nvidia-rtx/store/?page=1&limit=9&locale=en-us

 

Possible 4090ti soon? 

 

 

 

 

Edited January 21, 2023 by UltraMega

[UltraMega]

Hopefully Nvidia can start directly accepting organ donations as payment. 

[Sir Beregond]

  On 20/01/2023 at 23:42, bonami2 said:

NVIDIA RTX 6000 Ada Workstation Graphics Card With Over 18K Cores & 48 GB Memory Hits Retail For $6800 US

WCCFTECH.COM

---

Today, NVIDIA announces the retail availability of its flagship workstation graphics card based on the Ada architecture, the RTX...

 

 

https://store.nvidia.com/en-us/nvidia-rtx/store/?page=1&limit=9&locale=en-us

 

Possible 4090ti soon? 

 

 

 

 

Expand  

Doubt it if these sell. Why allocate AD102 silicon to a ~$2000 product when the same silicon can be used for a $6800 product and people are still buying up any new 4090's that hit the retail space (i.e. Not scalped).

[J7SC_Orion]

  On 20/01/2023 at 23:42, bonami2 said:

NVIDIA RTX 6000 Ada Workstation Graphics Card With Over 18K Cores & 48 GB Memory Hits Retail For $6800 US

WCCFTECH.COM

---

Today, NVIDIA announces the retail availability of its flagship workstation graphics card based on the Ada architecture, the RTX...

 

 

https://store.nvidia.com/en-us/nvidia-rtx/store/?page=1&limit=9&locale=en-us

 

Possible 4090ti soon? 

 

 

 

 

Expand  

 

...NVidia '''launched''' this card > back in September 2022 and I think it was referenced before at EHW, but doesn't hurt a second time for twice the fun  ...given the ongoing shortages of 4090s, NVidia will probably sell a fair bit of the RTX 6000 Ada to workstation users...it has more VRAM which is good for machine learning and rendering, albeit slower GDDR6 rather than GDDR6X VRAM, compared to a 4090.

 

Eventually, NVidia will bring out a 4090 Ti - probably shortly before the 5090s (shades of 3090 Ti)...or may be they'll do two or three: A Titan RTX Ada, a 4090 Ti and a 4090 Ti Super !

Spy pic 4090 Ti:

  Reveal hidden contents

 

 

[bonami2]

  On 21/01/2023 at 00:39, J7SC_Orion said:

 

...NVidia '''launched''' this card > back in September 2022 and I think it was referenced before at EHW, but doesn't hurt a second time for twice the fun  ...given the ongoing shortages of 4090s, NVidia will probably sell a fair bit of the RTX 6000 Ada to workstation users...it has more VRAM which is good for machine learning and rendering, albeit slower GDDR6 rather than GDDR6X VRAM, compared to a 4090.

 

Eventually, NVidia will bring out a 4090 Ti - probably shortly before the 5090s (shades of 3090 Ti)...or may be they'll do two or three: A Titan RTX Ada, a 4090 Ti and a 4090 Ti Super !

Spy pic 4090 Ti:

  Reveal hidden contents

 

 

Expand  

 

 

Yea it was i posted it then wondered about it and checked.  It just they have updated it on WCC because they are now officialy in stock. 

 

@Sir Beregond   Possible but they probably make more money selling defective gpu with a big VRAM capacity. And they could sell the full die with less vram or something and disable most of the workstation/cluster or idk what it called feature.

I don't understand why they would release a full die for server use because no one gonna pays the extra i would think? Maybe am wrong. 

[J7SC_Orion]

  On 21/01/2023 at 00:44, bonami2 said:

 

 

Yea it was i posted it then wondered about it and checked.  It just they have updated it on WCC because they are now officialy in stock. 

 

@Sir Beregond   Possible but they probably make more money selling defective gpu with a big VRAM capacity. And they could sell the full die with less vram or something and disable most of the workstation/cluster or idk what it called feature.

I don't understand why they would release a full die for server use because no one gonna pays the extra i would think? Maybe am wrong. 

Expand  

    

 

NVidia also has the H100 Hopper in play (mostly for the enterprise market, but there is some overlap to high-end workstation and R+D). Anyway, re. NVidia's general pricing (never mind AMD issues), here is an, ahem, somewhat critical view of things...

 

 

[Sir Beregond]

  On 21/01/2023 at 00:44, bonami2 said:

 

 

Yea it was i posted it then wondered about it and checked.  It just they have updated it on WCC because they are now officialy in stock. 

 

@Sir Beregond   Possible but they probably make more money selling defective gpu with a big VRAM capacity. And they could sell the full die with less vram or something and disable most of the workstation/cluster or idk what it called feature.

I don't understand why they would release a full die for server use because no one gonna pays the extra i would think? Maybe am wrong. 

Expand  

It's a large die with a massive increase in transistor density. Not sure what their yields of perfect/full dies are vs defective that can be cut down to 4090 spec.

[J7SC_Orion]

  On 21/01/2023 at 16:31, Sir Beregond said:

It's a large die with a massive increase in transistor density. Not sure what their yields of perfect/full dies are vs defective that can be cut down to 4090 spec.

Expand  

 

...that's the thing with massive monolithic dies that carry 76.3 billion transistors . The alternative, the mGPU chiplets, have instead issues around IF - for now...I expect that to change soon enough as the big three are gaining more experience with their current > mGPU chiplets in the enterprise market.

 

If you look at the Wikipedia-derived tables below, it becomes quite clear that the 4090 > Ti would be very much like the 3090 > Ti on a percentage basis - a minor performance increase compared to the other members of the Ada Lovelace family. Speaking of which, a few days ago, I was at a local outlet of a big chain here to pick up some USB extension cables...while they finally had received a new shipment of 4090s days earlier (like many places in N. America and Europe), all of those had sold out quickly at that location and also mostly online. I know this because I asked just out of habit --- also, no 4090s on the shelves behind the cashier, but four long and double-stacked rows of 4080s and two double-stacked rows of 4070 Tis.

 

Per below, it isn't just the Cuda core count but also the memory bus width - which is critical for Ada Lovelace...the 4090 (and presumably RTX 6000 Ada) can in fact get bottle-necked at 4K by the CPU/RAM in some apps. While many 4090s can reach 24 Gbps (instead of the stock 21 Gbps) on their GDDR6X , still-faster VRAM will be as important as the remaining Cudas, IMO. Finally, there's the power consumption. While 450W for GA102 is technically correct for the lower of two numbers subject to PSU (the other is 600W), I have seen just about 700 W on my 4090 and other folks with an even more aggressive vbios posted 822 W... 

 

Edited January 21, 2023 by J7SC_Orion