Leaked Internal Intel Memo Acknowledges 'Resurgent', 'Formidable' AMD (hothardware.com) 162
Slashdot reader MojoKid writes:
AMD announced its 3rd Gen Ryzen 3000 series processors at Computex earlier this month and the company's Zen 2 architecture is promised to bring single threaded performance parity with Intel but exceedingly better multithreaded throughput in content creation and other high-end workloads.
Intel has obviously taken notice of AMD's Zen 2 advancements and nowhere is its renewed keen focus more evident than in an internal memo that just leaked out to public venues. The memo was originally posted on Intel's internal "Circuit News" employee portal and it's quite revealing. The memo, which is entitled, "AMD competitive profile: Where we go toe-to-toe, why they are resurgent, which chips of ours beat theirs", is a surprisingly frank look at how AMD has managed to get the best of Intel, at least currently, and how the company should manage this renewed or "resurgent" competitive threat.
What's most surprising about the memo, which was penned by Circuit News Managing Editor Walden Kirsch, is how flattering it is in general to AMD, pointing out that it was the best-performing stock on the S&P 500 for 2018. In terms of Zen 2 and AMD's Ryzen 3000 series, the author notes, "Intel 9th Gen Core processors are likely to lead AMD's Ryzen-based products on lightly threaded productivity benchmarks as well as many gaming benchmarks," Kirsch writes in the memo. "For multi-threaded workloads, such as heavy content creation workloads, AMD's Matisse is expected to lead." All in, the internal memo is a rather insightful and well-reasoned look at the threat that AMD poses to Intel and how the company might respond.
Intel has obviously taken notice of AMD's Zen 2 advancements and nowhere is its renewed keen focus more evident than in an internal memo that just leaked out to public venues. The memo was originally posted on Intel's internal "Circuit News" employee portal and it's quite revealing. The memo, which is entitled, "AMD competitive profile: Where we go toe-to-toe, why they are resurgent, which chips of ours beat theirs", is a surprisingly frank look at how AMD has managed to get the best of Intel, at least currently, and how the company should manage this renewed or "resurgent" competitive threat.
What's most surprising about the memo, which was penned by Circuit News Managing Editor Walden Kirsch, is how flattering it is in general to AMD, pointing out that it was the best-performing stock on the S&P 500 for 2018. In terms of Zen 2 and AMD's Ryzen 3000 series, the author notes, "Intel 9th Gen Core processors are likely to lead AMD's Ryzen-based products on lightly threaded productivity benchmarks as well as many gaming benchmarks," Kirsch writes in the memo. "For multi-threaded workloads, such as heavy content creation workloads, AMD's Matisse is expected to lead." All in, the internal memo is a rather insightful and well-reasoned look at the threat that AMD poses to Intel and how the company might respond.
Intel needs to fix there CPU bugs! and more pci-e! (Score:2)
Intel needs to fix there CPU bugs! and more pci-e!
Re:Intel needs to fix there CPU bugs! and more pci (Score:4, Interesting)
Deer Intel,
Last time your dirty tricks forced AMD to spin off its FABs, adding to the pile of rent-a-fabs that are now crushing you. Did you consider the consequences?
Re: (Score:1)
Deer Intel, or Elk Intel?
Re: (Score:1)
Ewe, Intel.
Re: (Score:2)
Last time your dirty tricks forced AMD to spin off its FABs, adding to the pile of rent-a-fabs that are now crushing you. Did you consider the consequences?
They did and were crushing it all the way to 2014-15 with 14nm. The GloFo spinoff flopped, Bulldozer was a dud, AMD was barely hanging on the ropes and it seemed like Intel was ready for the KO punch. The only thing that got in the way was their own failure to execute on the 10nm process, nobody at Intel wants to say it but they've massively flopped the last 3-4 years and the only reason it's not worse is that AMD was so out of shape. If they don't get back to innovating there'll be a crisis in Intel soon a
Re: (Score:3)
Re: (Score:2)
but don't be surprised if Intel tries pulling some nasty shit in the future as we have seen in the past its SOP for them when they have more than token competition.
Here's an AMD response [engadget.com] to some FUD that probably was probably paid for by Intel.
MOD PARENT UP (Score:2)
This is entirely correct. I think there are legitimate complaints about phoronix, but overall they're by far the most trustworthy benchmarking out there because all the work is open and the workloads are representative of real ones. It was a very underrated CPU family, mostly due to dirty tricks.
Re: (Score:2)
Re: (Score:2)
Re: (Score:2)
You can't though.
Come on, man. Piledriver was a Pile.
I can make a 2003 Sempron kick the snot out of the latest i7
OK, I get it now. You're just making random shit up.
A 2003 Sempron would need to run at 9Ghz to have the clock-per-clock performance of a 9700K.
Get the fuck out of here, man.
Re: (Score:2)
Re: (Score:2)
You're wrong, you're just going to have to accept that.
ICC never produced 100% speedups on Intel parts (which is what would be required for an FX87xx to compete with an i7-3770k).
It was simply a shitty core. Deal with it.
Re: (Score:2)
Re: (Score:1)
8-core Piledrivers barely exceeded or matched a 4-core i7 of the time period.
That limits its usefulness to very multi-threaded applications, which I am sorry to inform you, simply does not include average Joe. You'll notice on the *one* single-threaded benchmark they ran, the i7 had roughly double the performance of the 8350, meaning on any workload from 1-4 concurrent cores (see: Games, average stuff running) the i7 was *twice* as fast, for $140 more.
Re: (Score:2)
Intel is thousands of products and while CPUs are really important to them, they are much much more than that.
I think a world where AMD and Intel are both doing great in the CPU world is a nice place to live. But it really doesn't matter what AMD does, they sell CPUs and GPUs and that's about it. If they have any other products, no one knows about them.
Intel has had problems and the fact is that most everyone on earth is more than happy with what CPU they currently have. Most peo
Re: (Score:2)
Last time your dirty tricks forced AMD to spin off its FABs, adding to the pile of rent-a-fabs that are now crushing you. Did you consider the consequences?
The consequences were it bought them almost 20 years of monopoly pricing.
Re: (Score:3)
SIMD instructions that operate on 512 bits at once
512 bits mean a whole cacheline in one go. This is a win in quite a number of scenarios: unless the processor can guess you won't overwrite the rest, any smaller write will first need to read in the whole cacheline, pointlessly. There are techniques to avoid that but the processor and compiler are only so smart, and it's not an easy thing for the programmer.
AVX512 simplifies all of that away, requiring no effort from anyone of the three.
Re: (Score:2)
Good grief, no SSE2 support in 2019? The newest AMD CPU to not support SSE2 was the Athlon XP, released in 2003! You could probably dumpster dive something 10 years newer than that for free. The power efficiency gains alone would make it a highly worthwhile upgrade. Surely you jest.
Re: (Score:2)
AVX 512 would be a nice upgrade for anything doing matrix math and you can expect most BLAS implementations to be updated quite quickly. It would certainly help the software we use here.
Re: (Score:2)
Re: (Score:1)
Why is it so hard to get their vs there right?
Re: (Score:3)
Re: (Score:2)
Slashdot keeps asking for more pci-e lanes but can never give a reason why.
Re: (Score:3)
Running multiple gpus, running lots of pcie storage. Two perfectly good reasons. Maybe you want to add some 10 GigE network cards, who knows. Lots of pcie is good. You get it for the same reason you went for a tower form factor instead of a laptop. Of course, you just posted from a laptop, right? Hard to explain PCIe lanes to somebody who never gets further into tech than social networking on a mobile device.
Re: (Score:2)
You and your imaginary friend don't.
Re: (Score:2)
Re: (Score:2)
You let your imaginary friend post?
Re: (Score:2)
Re: (Score:3)
A pacifier I assume.
Re: (Score:3)
Slashdot keeps asking for more pci-e lanes but can never give a reason why.
To run bigger Beowulf clusters, duh.
Re: (Score:2)
Throughput. The pci-e lanes isn't a reference to "how many slots" are on the boards (although you definately gain that out of it) , but how much CPU io throughput you have (defined by pairs of Input and output lanes to the CPU. If you have a low lane count your stuff will still work, but its going to be slower, and once you throw some SATA driives and a big firebreathing GPU in there, your already struggling.
Re: (Score:2)
Bandwidth for 16+ 8K cameras plus the GPU bandwidth needed to see and process them all simultaneously, while streaming+playing a game.
Re: (Score:2)
It makes no sense to work and game on the same PC at the same time. Back when they were expensive, that made sense, but PCs are cheap now.
Re: (Score:1)
This- for the pimp playing games while running a voyeur house, all on one rig.
Re: (Score:2)
Because I want enough lanes to run my GPU and more than one NVMe drive? Intel has been giving us a pittance of PCIe lanes for some time now. With thunderbolt now officially USB 4 that means we can completely strip out any legacy PCH/Southbridge driven interfaces. NVMe replaced SATA, USB 4 (thunderbolt) replaced legacy USB 3 etc. We can run everything off PCIe and provide tons of bandwidth. And since we are just talking about allocating lanes at this point, now motherboards can be specially tailored to the p
Re: (Score:2)
Because I want enough lanes to run my GPU and more than one NVMe drive?
What do you mean more than one. Intel doesn't even provide dedicated lanes for NVMe. On Intel boards you need to slave of the PCH for that.
Re: (Score:2)
On Intel boards you need to slave of the PCH for that.
Gasp!
Are you telling me that my NVMe, NIC, and USB ports will be limited to 33Gbps of *aggregate* throughput?
How will I ever manage?!
Re: (Score:2)
Re: (Score:1)
Re: (Score:2)
Even if you don't need them today, tomorrow you might want to add a USB 4.0 card, or an extra NVMe SSD. Having plenty of PCIe lanes future proofs your system.
I still use 6-7 year old CPUs regularly. The main issue I have with older machines, apart from being crippled by Spectre, is that they increasingly lack PCIe lanes. If you want USB 3, >gigabit network and an NVMe SSD you are going to have to compromise due to lack of PCIe lanes.
Re: (Score:2)
Extra? Intel chipset's NVMe slaves off the PCH as it is. The CPU only provides 16 lanes for peripherals.
Re: (Score:2)
That's true. Optane is crap too. Intel's storage stuff really sucks at the moment.
Re: (Score:2)
Am I confused, or isn't that essentially the same thing as a Ryzen 2700X?
Re: (Score:2)
I'm curious how often you're running into that bottleneck within the PCH PCIe lanes provided that are multiplexed over that. Are you often pushing >4GB/s across that bus?
Re: (Score:2)
The problem is the distribution of lanes to slots. If you want a GPU and an 8x storage device you have to drop the GPU down to 8x as well, at least on the older boards.
Re: (Score:1)
You've got your x16 on the GPU, and you've got our x8 (wtf storage device is x8, btw?) on the PCH.
Are you concerned about being able to only transfer 4GB/s across the DMI bus?
Re: (Score:2)
Well, yes actually. Single NVMe drives hit 3.5GB/sec back in 2017, and USB 3.1 Gen 2 is 1.25GB/sec per port, and 10Gb LAN is another GB/sec.
BTW RAID cards have been using PCIe 8x interfaces for a decade now. It's inadequate for your basic JBOD NVMe card.
Re: (Score:2)
BTW RAID cards have been using PCIe 8x interfaces for a decade now.
Sure, PERCs and such. Those aren't for you. Your AMD 2700X can't run one of those without compromise, either.
Re: (Score:2)
But that's assuming that the system can actually saturate that much. I find that in practice Intel systems, at least ones from a few years back (I'm all AMD now) tend to have problems saturating available PCIe bandwidth.
I've on Threadripper now.
Re: (Score:2)
Threadripper parts are analogous to Intel X-series HEDT parts with lots of lanes.
Intel parts (16 PCIe lanes, 4 DMI) are able to benchmark the fastest NVMes on the market, and are the fastest gaming machines on the market, so it would seem to me that 'unable to saturate that much' scenario probably doesn't exist, since that covers the DMI link, and the 16 CPU lanes.
I'm not arguing that t
Re: (Score:2)
They constantly give reasons you just willfully ignore them. NVMe is a big reason. Right now you can't sustain disk throughput and graphic throughput on an Intel motherboard at the same time. We're not talking about magic RAID or SLI configurations either. It's possible to saturate what the 9900k offers with a single GPU and a single NVMe drive. This has led to all sorts of comical solutions including the use of spare PCIe lanes to the memory modules being used as NVMe riser cards on Asus motherboards.
Mind
Re: (Score:1)
You're talking out of your ass, dude. This stinks like an AMD marketing piece. Your statements are more loaded than a toddler's diaper.
Re: (Score:2)
And you need to fix your basic grammar errors. :P
Re: (Score:2)
No they don't. AMD needs to introduce some bugs. I don't want to sacrifice performance for a completely irrelevant security gain thankyou very much. I don't run a datacentre and don't manage the worlds finances from my PC.
Intel need to stop with the rest of the bullshit though. They are being well and truly out priced. For PCIe through put they are being outclassed. And for god sake their CPUs have features which are DRM locked which need a dongle to activate, something AMD provides for free. https://www.am [amazon.com]
Re: (Score:1)
intel 10nm was supposed to be out 2 or 3 years ago
My next PC will have a Ryzen 3950X in it. (Score:2)
Not on price/perfomance though. (Score:1)
There, Intel doesn't have a grain of a chance.
I'm all for Intel improving due to AMD. But what's creepy is how they do *not* focus on actually improving, but merely on *spinning* it in their favor.
That is textbook psychopath behavior.
It makes it look like they think they can still save their old ways. The last times that happened, they used some really dirty, and even illegal tricks, like blackmailing mainboard vendors, deliberately crippling compiler code paths, and of course speculative execution shortcut
Re: (Score:2)
Intel is misstating AMD's TSMC-related advantages. Yes, moving to TSMC (and their 7nm process) allowed AMD to get a jump on Intel's delayed 10nm process (which is roughly equivalent to TSMC's 7nm), but it's not what let AMD drive core counts higher. The original Ryzen was on GlobalFoundaries' 14nm process, and it too managed significant core count advantages over Intel. Those advantages had more to do with their design than the process node.
That's not to say that being on TSMC 7nm isn't a big positive for A
They are already working on that. (Score:2)
Intel recently hired the non-shill remnants of the Computer benchmarking and performance press.
Kyle from Harocp was an ass, but they were the Go-to site for honest performance ratings. :(
He was hired by intel this spring, and has effectively shuttered his site.
People will put us with a lot for insurance; I guess that's the last handle the big companies have on people these days.
Others less capable were hired away as well; if you're into this stuff, you know who I mean.
We will see lawsuits, threats, and some
Re: (Score:2)
As an AMD investor, the part that scares me is how Intel will decide to compete. I am not worried if they try to 'out-engineer' AMD - but I am very much worried they will offer short-term cash somehow to large OEMs to keep AMD from being able to compete on a fair playground. For example, there are zero business laptop options that have any sort of premium feel to them. Every AMD laptop has one option missing - be it a docking port, 4k screen, decent battery, thin and light, etc.. Also, the business desktop
Re: (Score:3)
Content creators. Doesn't matter if you're rendering stuff in blender, or need to stream to periscope and youtube at the same time. You want and often need any and all CPU power you can get.
If the leaked benchmarks are to be believed (Score:2)
Intel might have rested on their laurels too long. Either that or their engineers couldn't figure out how to massively improve performance.
OTOH I hate to say this but after a series of unsolvable crashes in Far Cry 3 I'm probably going back to nVidia... And speaking of which leaked benchmarks show the Navi underperforming the "super" 2070 in FF15 (though to be fair that's a game that heavily favor's nVidia and my rx 580 wa
Re: (Score:2)
Either that or their engineers couldn't figure out how to massively improve performance.
Oh I think the engineers have figured it out . . . but they cannot convince their managers to believe them.
Re: (Score:3)
Intel might have rested on their laurels too long. Either that or their engineers couldn't figure out how to massively improve performance.
They didnt rest at all. Let me give a bit of a time-line.
Intel leaped ahead of everyone with a sensational new transistor design, 3D tri-gates. This allowed them to pack transistors more densely, more or less equivalent to a 22nm FinFET but not actually on a process node that would be called 22nm without the 3D tri-gates. Over time they improved their process, culminating in transistors densities approximating 14nm FinFET, but still required tri-gates to get that density.
They then blew off about 5 year
Re: (Score:3)
Re: (Score:1, Troll)
Everyone loves to talk about the death of Intel and how Intels goose is cooked everytime AMD brings something new to the table. Just to point a couple things out about Intel. 98% market share and name recognition. Of course AMD has name recognition in the market and its bad. Managers think AMD as a second rate competitor to Intel. That will never change until AMD brings something revolutionary to the table again and not just slightly slower chips at a cheaper price.
Re: (Score:1)
That will never change until AMD brings something revolutionary to the table again and not just slightly slower chips at a cheaper price.
Have you been asleep for the last 12 months? 16 core Ryzen is due in a week. 64 core Threadripper is due at the end of the year.
Re: (Score:3)
Tossing more cores into the pot is nothing new. Intel can easily add more cores to any of its produce and people will buy them just because of the Intel name. When I say something new I mean being the first to break 1Gzh or the x64 instruction sets. Those where all AMD firsts.
Re: (Score:2)
Tossing more cores into the pot is nothing new. Intel can easily add more cores to any of its produce and people will buy them just because of the Intel name.
Oddly enough, no, they can't. Not right now. Because a TDP of 600W is completely untenable. Intel's many-core chip, the Xeon Phi, was clocked at no more than 1.7GHz. Any higher and it got much too hot.
Funny thing is, later this year AMD will be releasing a full speed (~3.7 GHz) AMD64 architecture processor with 64 cores, right in the middle of the core count range of Intel's Xeon Phi, which was 57 to 72 cores, depending on binning. With vastly more capable cores. Xeon Phis are still in current product
Re: (Score:2)
Obviously name recognition can be gained or lost over time, as a lot of people are talking about AMD these days.
Re: (Score:2)
A quick Google suggests that AMD is at around 19% share for desktop and 14% share for laptops.
For games consoles, both the PS4 and XBOX One use AMD CPUs and GPUs.
Re: (Score:3)
Re: (Score:1)
Interesting. What's your username on Seeking Alpha?
Re: (Score:3)
It may beat the 9900k in some particular single core benchmarks, very few though. Don't believe the lies. It's a good chip but 9900k is going to win single thread mostly, regardless.
Great they are closing the gap though.
Dream on. (Score:1)
Are you using the Intel compiler again?
Conspiracy (Score:3)
Re: (Score:1)
Proof Intel thinks AMD has no problem competing.
But from all 3rd party reports and leaks it looks like AMD really has no problem competing. I think the memo is more of a pep talk for Intel's employees.
Re: (Score:2)
It's definitely a pep talk, but it's also true. 10 years ago Intel was a process node or two ahead of its peers; since then Intel has stalled and the peers have pulled even.
Re: (Score:2)
I see only one peer, AMD. Are there others?
Re: (Score:2)
INB4 RiskV!!!11
Re: (Score:2)
I see only one peer, AMD. Are there others?
Depends on how you look at the market. You have Chromebooks and Raspberry Pis running ARM cores but there won't be any volume until Apple makes the switch, the technology is state of the art though. I guess far off on the sideline you have nVidia doing computing tasks but CUDA/OpenCL is fairly niche. There's a helluva lot of software that's prepared to be cross platform now though due to mobile, I don't think Intel should feel too safe.
Re: (Score:1)
ARM and a few others in low power applications (tablet, laptop). IBM in (expensive) servers.
But on the desktop? Currently none. There it is Intel and there is AMD.
AMD is back to what they were 20 years ago (Score:1)
Back in 2001 I wouldn't use Intel to make a serious PC rig back in the Athlon days. And they invented X86-64.
Remember the crappy Intel Itanic?
Re: (Score:2)
Actually, no, AMD invented x86_64, which is the vendor neutral name for AMD64. Intel originally came up with ia64 (the aforementioned Itanic) and had to adopt x86_64 years after AMD came out with it, dumping ia64 into the wastebin in the process.
Perhaps you're thinking about 32-bit x86?
That didn't take long, lol. (Score:2)
Let me guess; you work for intel's marketing department?
LOL.
Re: (Score:2)
Well the closest thing mentioned were the six pillars as their ingredients of the secret sauce. One of those was "security". I've seen many vloggers fail to not chuckle when reading out loud that part.
The other five were process (they're behind the competition here), architecture (AMD possibly taking the IPC crown, this is already an arguable thing), memory (they're admitting they're challenged on the RAM front; they still do have the optane technology though), interconnect (nothing yet really to show here;