Oh it's not selling at all just yet.
Late Q1 was just the inflection point.
For client in general: it doesn't matter if it is much faster Intel will spam cheapo RPL laptops at slightly lower prices and it will sell better still.
5% YoY revenue gain, while switching to a new socket is not bad.We've been saying that for 5 years and they've only made modest progress.
At this point, AMD is basically dominating hyperscaler sales. If you are purchasing mainly the highest-end chips, and are doing so by the truckload, you are probably already getting more than half your x86 product from AMD.
Papermaster said as much in one of the investor convergences late last year. That among US based hyperscalers, AMD reached 50% market share.
Wich mean that there s 50% who are buying outdated Intel hardware wich is one gen late, at this rate they need to spend at least 50% more for an equivalently performing configuration, not counting the much bigger power comsumption bill, so there s 50%
of the executives who are incompetent.
Even with the discounts that s a certainity that they ll lose marketshare because they will have less throughput/VMs for sale than those who made the right choice, and at a higher cost given the augmented TCO, if they want to subside Intel at their own expenses so be it, their competitors and AMD buyers approve such a foly.It's not so simple because of the discounts. Intel can get away with it because they have so much slack 10 nm capacity... well not enough to fund the future nodes, but hey.
Wich mean that there s 50% who are buying outdated Intel hardware wich is one gen late, at this rate they need to spend at least 50% more for an equivalently performing configuration, not counting the much bigger power comsumption bill, so there s 50% of the clouds executives who are incompetent.
Wow - that IS a long time in manufacturing for a semi-product. Must be giving the CFO some sweats.Lisa said it takes 7-8 months to produce Mi300
IMHO, that'll only work for workloads that go a lighter on the CPU and much heavier on the GPU. Often they are tightly coupled, which is why ST still matters. I haven't looked at this for LLMs - so in that one arena, I'm not sure what makes the best system.In not-too-distant future, NVidia will try to phase out SPR, and try to force-freed customers with their own Grace CPUs.
I agree. Intel benefited from NVidia reference systems having SPR CPU. Clearly, because NVidia does not perceive Intel as a threat, rather than pairing it with far more capable Genoa CPUs. The customers are forced to use SPR. I wonder what percentage of Intel server CPUs have shipped this way.
In not-too-distant future, NVidia will try to phase out SPR, and try to force-freed customers with their own Grace CPUs.
FIFY.In much less than a distant future, that is this year, AMD will provide a full platform with Zen 5 based Epyc andMI400MI300, for the time this is Zen 4 based Epyc + MI300 and they have best perf/watt and absolute perf even with the latter.
In much less than a distant future, that is this year, AMD will provide a full plateform with Zen 5 based Epyc and MI400, for the time this is Zen 4 based Epyc + MI300 and they have best perf/watt and absolute perf even with the latter.