NVIDIA discussion [2024]

  • Thread starter Deleted member 2197
  • Start date
NVIDIA secured a 500 million $ orders from India for tens of thousands of H100 and H200 GPUs.

So the total order from this single Indian data center company will likely exceed $1 billion, and there are a lot of DC companies out there to sell to.
Will be interesting to see if this trend continues for DC to procure very high numbers of AI GPU's.
 
Expreview managed to get hold of the GALAX GeForce RTX 4090D Metal Master graphics card which is based on the new AD102-250 GPU silicon and is designed to comply with US policies since the original RTX 4090 had been restricted due to its TPP (Total Processing Performance) exceeding the 4800 points limit.
...
The tech outlet tested the NVIDIA GeForce RTX 4090D graphics card against the RTX 4090 in various games at 4K resolution and found it to be around 5.8% slower and with frame generation enabled, the difference falls to just 5%. The NVIDIA GeForce RTX 4090D was also compared against the 4090 & 4080 GPUs in AI and productivity benchmarks and once again ended up 6% slower.
Following is the performance chart which shows the RTX 4090D easily exceeding the AMD Radeon RX 7900 XTX in rasterized performance:
NVIDIA-GeForce-RTX-4090D-GPU-Performance-Benchmarks-vs-RTX-4090.png
 
This mentions the 4070 Super having only 36MB of L2. That's a lot less than the previously reported 48MB. Will be interesting to see how much this matters vs the full 48MB 4070Ti.

Seems like it was an error. Nvidia's changed the product page to 48MB for L2, the same as the 4070ti/full AD104.


4070 ti Super though is still a cut down 48MB vs the 64MB for full AD103.
 
I can't really argue this point, but my GTX970 consumed around 200W after OC (probably more like 175W) and it wasn't nearly as long and wasn't very loud. It was also way cheaper but I guess that ship has sailed.
The two fan GTX 970 I have (I'm actually using this paired with a 12700k in my main gaming PC for "reasons") - https://www.techpowerup.com/review/msi-gtx-970-gaming/

I find loud if running full tilt without dynamic v-sync (so 60 fps at the time) if hit close to 2k rpm or higher. Noticeably audible if ramping past 1.5k as well. And I just ran it stock (so close to 165w).

But I should mention I use open earphones with the PC just under the desk so there's barely an noise mitigation.
 
The two fan GTX 970 I have (I'm actually using this paired with a 12700k in my main gaming PC for "reasons") - https://www.techpowerup.com/review/msi-gtx-970-gaming/

I find loud if running full tilt without dynamic v-sync (so 60 fps at the time) if hit close to 2k rpm or higher. Noticeably audible if ramping past 1.5k as well. And I just ran it stock (so close to 165w).

But I should mention I use open earphones with the PC just under the desk so there's barely an noise mitigation.
That's the exact same 970 I have. I can't hear it when I'm gaming but I use a normal headset.

Oh and don't feel bad, for a bit I had it paired with my 13600K for "reasons" as well :LOL:
 
Last edited:
When there's limited CoWoS capacity they can't just expand DC stuff and gaming still brings healthy profit, not to mention synergies in R&D, why would you shoot a milking cow just because another cow milks more and you can keep both?
Lots of companies do that though. If they have a target and a segment is less profitable it may disappear. It seems like a bad idea p.r.wise but it happens
 
January 22, 2024
This development is no surprise to industry observers who recall similar adaptations with the RTX 20 and 30 series. These companies are leveraging the lower cost of mobile GPUs, combined with budget cooling solutions and simpler PCB designs, to offer more affordable desktop GPU options. The mobile GPUs, which are capped at a power consumption of 175 Watts, are being repurposed without official sanction, with NVIDIA seemingly disregarding this practice. Despite the lack of official endorsement, these modified GPUs are finding their way into the market, providing gamers a cost-effective alternative to the more expensive desktop versions.
...
While not officially supported by NVIDIA, these cards utilize the mobile GPU dies paired with custom cooling solutions and PCBs to work in desktop PCs. According to reports, the RTX 4080M desktop variant offers 7424 CUDA cores and 12 GB GDDR6 memory, representing a 24% reduction in cores and 4 GB less memory versus the desktop RTX 4080. The desktop RTX 4090M is even more cut-down, with 9728 cores and 16 GB memory—a 40% drop in cores and 8 GB less memory than the flagship RTX 4090 desktop card. Pricing falls between $420 and $560 for the RTX 4080M and exceeds that of even the desktop RTX 4090 for the 4090M variant. Performance and longevity still need to be determined for these unofficial cards. While they present a cheaper RTX 40-series option for Chinese gamers, the reduced specifications come with tradeoffs.
 
Cards should be smaller across the board. My baseline MSI RTX4070 Ventus 3x is gigantic. It barely fits in my case. There's no reason for this.
Totally agree with this. The Gigabyte 4070 Ti Gaming OC I brought home actually didn't fit in my case. Totally dwarfed the previous 2080 Ti which only had 35W less power draw. Ended up having to put it in another PC.

Then I got an Asus Proart 4080 on a Black Friday deal. This card is only 30 cm long, less than a Founder's Edition even. It is still possible to make cards with halfway sane dimensions.
 
January 22, 2024
I'd be ok with this if they were supported with normal Nvidia driver releases. But they seem to require hacked driver kits. Ugh.
 
Totally agree with this. The Gigabyte 4070 Ti Gaming OC I brought home actually didn't fit in my case. Totally dwarfed the previous 2080 Ti which only had 35W less power draw. Ended up having to put it in another PC.

Then I got an Asus Proart 4080 on a Black Friday deal. This card is only 30 cm long, less than a Founder's Edition even. It is still possible to make cards with halfway sane dimensions.
Damn that 4080 is really nice looking. I didn't know anyone made cards like that.
 
Totally agree with this. The Gigabyte 4070 Ti Gaming OC I brought home actually didn't fit in my case. Totally dwarfed the previous 2080 Ti which only had 35W less power draw. Ended up having to put it in another PC.

Then I got an Asus Proart 4080 on a Black Friday deal. This card is only 30 cm long, less than a Founder's Edition even. It is still possible to make cards with halfway sane dimensions.

Damn that 4080 is really nice looking. I didn't know anyone made cards like that.

The line is smaller than the 4080 because the 4080 FE itself is really large sharing the same heatsink as the 4090.

Unfortunately for you Homerdog the Proart line shares the same heatsink down the 4060ti, which means it's actually really large for a 4070 class card, even more so for a 4060ti class card (this was unfortunate for me as there was wicked sale for a 16GB 4060ti, for ~$360 USD, but it was the ProArt and the case I wanted to put it only takes 29cm long cards).

In terms of aesthetics if you're interested, MSI is launching it's "expert" series which goes for design ques closer to Nvidia's FE line. The downside is it'll probably a have a significant premium, as typical for AiBs, over the FE as opposed to MSRP.
 
The line is smaller than the 4080 because the 4080 FE itself is really large sharing the same heatsink as the 4090.

Unfortunately for you Homerdog the Proart line shares the same heatsink down the 4060ti, which means it's actually really large for a 4070 class card, even more so for a 4060ti class card (this was unfortunate for me as there was wicked sale for a 16GB 4060ti, for ~$360 USD, but it was the ProArt and the case I wanted to put it only takes 29cm long cards).
Yep you nailed it. The 4080 FE is a pretty sizeable card, and just about every OEM seems to have taken that as a cue to bring cards that are at least 2 cm longer still, with more than a few gaining as much as 6. And while thickness is likely less often the main issue when it comes to fitting in people's cases, the FE is already quite chunky at 3 slots. So of course they need to one up that too and shoot for at least 3.5. And then take those designs down to their 4070s as well. It's all really silly if you ask me.

Still, for a 4080 level card I'm liking this form factor. Noise isn't much of an issue with full fan stop when not under load, and max RPM is easily tolerable.

ProArt-GeForce-RTX-4080-OC-Edition.jpg
 
Last edited:
As per the UDN report, NVIDIA will add Intel as a provider of advanced packaging services to help ease the constraints. Intel is expected to start supplying NVIDIA with a monthly advanced packaging capacity of about 5,000 units in Q2 at the earliest. While TSMC will remain NVIDIA's primary packaging partner, Intel's participation significantly boosts NVIDIA's total production capacity by nearly 10%. Even after Intel comes online, TSMC will still account for the lion's share—about 90% of NVIDIA's advanced packaging needs.
 
Back
Top