They ’re mortal enemies . Sworn rivals . The two top names when it fare to thebest art card , duking it out even over petty performance difference . And yet , AMD and Nvidia are partnering for the launch of Team Red ’s 5th - gen Epyc server central processing unit , code - named Turin .
Through what AMD called a “ proficient partnership , ” Nvidia is providing counselling on how to partner off its HGX and MGX datum nub GPU bunch with AMD ’s new Epyc CPUs . Despite the fact that AMD makes its own Instinct AI accelerators , the partnership seems like a read on realness . client will require to use Nvidia GPUs with AMD CPUs , and the two companies came together to offer that counselling .
And that ’s a good thing because AMD ’s fresh data center potato chip take care mighty impressive . The bedrock they are establish on is the Zen 5 architecture that we saw in CPUs like theRyzen 9 9950X and Ryzen 9900X , but massively expanded . you may see the dizzying product stack below , which place from CPUs from eight burden all the way up to a massive chip tamp 192 of AMD ’s Zen 5c cores . And as a cherry red on top , the sweet spot 64 - core 9575F can hit 5GHz — a first for AMD ’s data point center chips .
AMD is prove some pretty massive operation gains , too — 2.7x the performance on the SPEC CPU bench mark compared to an Intel Emerald Rapid Xeon CPU , and up to four multiplication the go-ahead performance , which looks at transcoding video . Intel just of late launch its Granite Rapid Xeon CPUs , but AMD did n’t share any performance numbers for those fries .
Turin CPUs are monumental , and it ’s no admiration that Nvidia is working with AMD to bring home the bacon counsel to customer . Still , AMD is n’t down for the count when it comes to data nerve centre GPUs . The ship’s company formally set in motion its new Instinct MI325X AI throttle valve alongside Turin CPUs , which it previously preview at Computex earlier this yr . A single GPU pack 256 GB of HBM3E memory , and when machinate into a cluster , AMD says the platform can suffer up to 2 TB of HBM3E computer memory and 48TB / s of memory bandwidth .
AMD is still gun after Nvidia ’s AI crown with the GPU , too . The company demonstrate the MI325X outmatch Nvidia ’s H200 in AI inference and supply performance parity in AI training . Nvidia has moved onto itsBlackwell information center GPUs , but the H200 comparison is still bonnie given how quickly AI computer hardware is roll up out these day .
Although most of the readers here at Digital Trends will never interact directly with this hardware , it ’s still nice to look and gape at just how much power company like AMD can squeeze onto a CPU . Behind the scene , AMD ’s Epyc CPUs help power dozens of divine service we all interact with every sidereal day , from Netflix to Snapchat and beyond .
