D
For Compute-optimized GPUs that are mainly used for cloud and datacenters, NVIDIA controls the whole market with over 300K shipments during Q2 2023 whereas Intel and AMD were around 10-20K units.
For Client GPUs in Q2 2023, NVIDIA sets at 87%, AMD at 10% and Intel at 3%.
Total shipment increased in Q2 compared to Q1, where marketshare numbers for NVIDIA stood at 88% and AMD at 8%. So the market is rebounding.
GPU Market Rebounds In Q2 2023: AMD, NVIDIA & Intel See Increased Shipments, Discrete GPU Up By 12.4%
JPR has posted KJQ2 2023 market analysis, showing increased GPU shipments for AMD, Intel, & NVIDIA thanks to market normalization.wccftech.com
Less and less reasons for people to upgrade their dGPU.
That's incredible. I assume if the GPU doesn't have extra support inside the case (in addition to the support from the IO bracket at the rear) then this can happen in transit. A 4090 is truly titanic. I doubt the standard PCIe mounting system was designed to support cards of that size and weight.dont really know where to put this so here it goes (move it out if this is wrong place)
the precision here is insane
i dont understand how this damage happened, was this card in a motherboard in a fully built computer chassis and then shipped internationally or is he saying it happened inside a retail box with padding etc? i didnt understand when he talked about it
i would never dare do anything like this, yikes scary especially since i just ordered an msi 4090
It is now disclosed that NVIDIA has started to take steps to proceed towards its Blackwell AI GPUs since DigiTimes reports that the tech giant has already contacted for services of TSMC. Blackwell GB100 GPUs were already rumored to feature TSMC's 3nm process & now it looks like a deal has been verified. However, right now, TSMC particularly its 3nm manufacturing is facing hindrances in the form of utilization rates and a lack of advanced manufacturing plants. So, for now, the actual 3nm implementation could be seen deep into 2024, right when Blackwell GB100s are scheduled to be released.
...
AMD's EPYC Instinct MI300 accelerators are expected to launch later this year on the 5nm process node so a follow-up can be expected around 2025-2026 which will mean Blackwell GPUs will have a head start on an advanced process node. Intel is expected to launch its Gaudi 3 AI accelerator next year based on the 5nm process node and a follow-up in the form of Falcon Shores in 2025.
84 to 128 if we're looking at actual products is not a 70% increase though.Going from 7nm GA102 to 5nm AD102 was a 70% increase in SMs
Blackwell should bring some serious architectural changes though so comparing SM numbers may be pointless.
GA102 was Samsung 8nm (really just an enhanced 10nm class process), not 7nm, no? Only GA100 was TSMC 7nm.Interesting. Going from 7nm GA102 to 5nm AD102 was a 70% increase in SMs and 50% increase in clocks. AD102 is really a beast.
192 SMs on GB102 would be a relatively small increase of just 33%. Maybe clocks are going up again.
GA102 was Samsung 8nm (really just an enhanced 10nm class process), not 7nm, no? Only GA100 was TSMC 7nm.
(either way, AD102 really was a heck of a step up, reminded me of GP102)
Ampere felt like a big step up because Turing was ultimately quite disappointing in terms of general performance architecture, and because it was still stuck on TSMC 16nm family. This meant that Ampere had more room for improvement than normal. Not to say Ampere wasn't a great architecture, it was, and it was also not as good as it could have been thanks to Samsung 8nm, but still, the disappointment of Turing created the sort of expectations that helped Ampere perhaps look better than it otherwise would have.GA102 was Samsung 8nm (really just an enhanced 10nm class process), not 7nm, no? Only GA100 was TSMC 7nm.
(either way, AD102 really was a heck of a step up, reminded me of GP102)
Ampere felt like a big step up because Turing was ultimately quite disappointing in terms of general performance architecture, and because it was still stuck on TSMC 16nm family. This meant that Ampere had more room for improvement than normal. Not to say Ampere wasn't a great architecture, it was, and it was also not as good as it could have been thanks to Samsung 8nm, but still, the disappointment of Turing created the sort of expectations that helped Ampere perhaps look better than it otherwise would have.
the french are probably angry because nvidia made wine and cheese with aiNvidia offices raided by French competition authority
Nvidia’s reign in the graphics card market is getting attention.www.theverge.com
Yea crap, my bad.AD102, not Ampere.