The past few flurries of conferences have been incredibly informative as the information war aspect of the AMD-Intel feud grows ever hotter. AMD didn’t let go of the opportunity to throw a few potshots at Intel – and we must say this is something the company has more than earned (although TSMC gets a fair share of the credit here too considering it is their node AMD fabs on). AMD processors are (for the first time in history) at a process advantage over Intel – something that hasn’t happened in the x86 industry for the past 30 years.
AMD: We thought we were going to be at parity, we didn’t dream we would be ahead of Intel
AMD’s stock price recently $42, very close to its all-time high of $47.5 back before the dot com bubble burst in June 2000, and it’s very clear to even the naysayers that the company is back in all its glory – and still with tons of room to grow! I am talking of course about their GPU side, which has taken a back seat to all of the action ever since Zen become front and center. But that is a discussion for another article, and without any further ado, here is a transcription of AMD’s presentation at Barclays:
…we had some deficits vs Intel in some of the low thread count or single thread count , single threaded applications. With Rome, we not only doubled the throughput or more than doubled the throughput in terms of doubling the cores per socket over our previous generation and more than doubled versus Intel’s contemporary generation.
But we also did a lot of work on IPC so that the architecture of the machines was just faster on a per thread basis, and of course, with the process advantage that we have, where we had originally planned to have process parity with Intel, and we were super excited about that.
First time in the industry, we’re going to break the laws of nature that had governed the semiconductor industry for the last 30 years, which is — it’s like the fourth wall of Physics in semiconductors, Intel has a process node advantage. We were excited 4 years ago, because we thought we’re going to be at parity . We didn’t dream that we would be ahead.
And so, that is also beyond the architectural advantage that we design for, that we plan for, that’s also been a huge uplift.
And so with Rome, we have leadership not just in throughput oriented workloads, not just in HPC and cloud and large virtualization forms, all of which were positions we’ve staked out pretty well in the first generation, but pretty much across the board, you look at our 16 core or our 24 core parts versus Intel thread for thread.
We’re delivering substantially better performance and better power. And so, there’s sort of no place to hide. Pretty much across the board in terms of the vast majority of workloads, except for maybe a few that have been very tightly tuned for some peculiarity of Intel’s ISA . We think we’re in great leadership position even better than we had originally planned. -Forrest Norrod, SVP, AMD Datacenter Group
In what is a pretty candid conversation, Forrest Norrod frankly admits that the last generation of AMD Zen was lacking in some aspects as far as single-threaded performance goes – something their latest generation has completely solved. He also talks about how AMD was hoping they would achieve parity with Intel processors but never even dream that they would be ahead – something which is a testament to their commitment to the cause.
With TSMC looking to start risk production of 5nm, it doesn’t look like Intel will be able to take back the lead anytime soon – or even achieve parity. The closest they can hope to achieve parity is after 2021 when and if their first EUV processors ship on time. Intel’s CEO had also had a very frank discussion of what went wrong with 10nm and its really refreshing to see both of these companies making these usually banal conferences worth listening to with all this commentary.
Share
Tweet
Submit
[ad_2]