* Posts by msroadkill

45 publicly visible posts • joined 13 Jun 2011

Intel abandons XPU plan to cram CPU, GPU, memory into one package



Way back in Q1 2017, starting with a simple 4 core ccx/lego block, AMD could assemble 4/8/16 CCX into socket modules for desktop/ws/server respectively.

Even further back in 22014 or sooner, they were doing similar with 2x 2core ccx + a very impressive GPU in APUs like the 7850k.

It is now nearly q3 2023, & we have lame, same ol 8 core monolith intel, plus ecores since feb 2022 & a now delayed & now canceled APU

Nvidia is similarly unimpressive - they cobbled together their old risk cpu & gpu & gave them separate caches & a nvlink inteconnect.

In over 6 years of Infinity Fabric/chiplet omnipresence in the market, this is the best NV & Intel can manage?

amd is slaughtering intel in DC, & MI300 slaughters grace hopper.

~NV can manage a cpu/gpu combo, & intel, a perf cpu w/ 1 or 2 (sometimes useful w/ MS help) e core chiplets.

Even amd chiplets have chiplets. MI300 has ~15 (high yield & cool running) chiplets &15billion transistors.

Intel says AI is overwhelming CPUs, GPUs, even clouds – so all Meteor Lakes get a VPU


Re: Chipzilla is name challenged.

... only to have them cancelled or flash to insignificant obscurity.

Server shipments fall ... just as AI drives demand for costlier kit


The usual argument is that power, space & cooling costs are almost as important.

I think improvements from Epyc in particular make new gear very cost justifiable. but there is so much uncertainty than minimal capital costs is trumps.

How do you boost server efficiency? Buy new kit, keep it busy


AFAICT, FGPAs, an AMD strong suite, should help at keeping data center's utilization rates optimal, buy switching various resources off and on to suit needs.

For its big comeback, Intel needs to spend money – and it's making less and less of it



I am amazed its legal.

Intel: Please buy these new 13th-Gen CPUs, now with 24 cores


If Intel were a person they would be rightly shunned as ill-mannered. This SKU proliferation is just one example of their time wasting marketing ploys to gain shelf space etc.

China may have to reassess chip strategy in face of US sanctions


5000 years of misery & sudden prosperityvia a blend of nationalistic controlled capitalism - yay - now lets throw our weight around.

Uts this that is the real agenda here. China's BS in the south china sea is way over the top, & so is missile stunts by their vassal nth korea.

confronting a big nuclear power on their doorstep on the other side of the pacific is a lot harder & riskier than screwing them economically - the silly bastards got rich from the even sillier laissaz faire capitalists throwing their own workforce & industry to the wolves to shave a few dollars from costs ("let them shoot smack")

I despise trump, but it took one mad bastard to confront another in a plausible way, and Biden stayed the course.

war is always a bitter pill, but this economic war beats the real thing & is very necessary to check the Chinese leadership.

Years late and 36 cores short of AMD, who are Intel’s 4th-gen Xeons even for?


Chiplets compromise latency to gain well proven advantages in ~everything else.

SR is half assed chiplets, w/ a slight latency advantage, & disadvantages in ~all else.

AMD 8 core chiplets are in fact 2x 4 core chiplets - making the simple elegance of AMD's architecture even more contrasting.

An interesting crossover point is the core limits of 2 socket genoa - 2 socket x96 = 192 core Genoa's vs ~4 socket x56 core SA. I very much doubt the intel option has better latency.

They say that in this market, folks dont buy chips, they buy roadmaps, & Intel could hardly have done worse in this regard for many years.

As the author says - AMD has now become the default choice in servers. As AMD inevitably resolves its supply shortages, it will be an interesting market.

Google datacenters use 'a quarter of all water' in one US city



Samsung slaps processing-in-memory chips onto GPUs for first-of-its-kind supercomputer


Which is why this is about its inclusion in an AMD product. Not their competition.

The decade+ AMD invested in Infinity Fabric is the reason only mad have mastered the core to core coherency/latency problems posed in cpu & other processor chiplets.

The others are dreaming if they think they can come up with a timely response.


re article conclusion: "We'll know if Samsung ends up making any inroads if PIM starts appearing in new supercomputers set up by research labs, academic institutions, and other organizations over the next few years."

Isnt that exactly what it reports? PIM is included in the MI100 server module of which 96 are included in the announced system, & the MI100 is soon to be supplanted by the spectacularly performant MI300.

Nvidia's CUDA moat looks more bridgeable every passing quarter.

Google Cloud is super keen to keep certain customers on pricey Intel VMs


IE., Google's Xeon instance infrastructure are now a tough sell.

To extract more instance value, this marketing exercise selectively identifies clients who can use helpful Intel/google expertise.

Europe's most powerful supercomputer is an all-AMD beast


It seems flawed "green" logic to "entirely" run it on hydro power.

Hydro is the precious equivalent of the battery, as an antidote to the intermittent nature of most other green power sources

Intel says costly 10nm ramp will counter PC slowdown


"Intel's expensive comeback plan to introduce five new nodes in four years, with the goal of surpassing foundry rivals TSMC and Samsung by 2025. "

Use rival's fabs to dominate them?

Yeah right.

AMD takes another crack at Intel's server stronghold with more Epyc silicon


In the cited comparison, intel 1TB of ram, amd 4 tb. Sounds a deal breaker to me.

Cloudflare says Intel is not inside its next-gen servers – Ice Lake melted its energy budget


Re: Is Intel the new IBM?

I like ur analysys - u sound like u have some inside knowledge.

I am inexpert, but I too am suspicious of the alleged threat of ARM to X86.

Yes it exists, but not to the extent, or the way in which it is portrayed.

I liked ur analogy of x86 vs mobile vs arm vs x86.

Given ARM's history, some points?

Yes cloud is a huge new force, but its not a complete compute solution (wan interconnects will ~always be relatively puny), & cloud companies with the massive apps to justify custom chips wont be vendors of them.

AFAICT, AMD's Fabric could integrate ARM processors for the best of both worlds

The sheer range of AMDs simplified core unit lego blocks - the same unit serves from massive servers down to laptops and embedded devices - spreading the massive development costs very thinly and at a power penalty which is almost imperceptible to the end user where it counts.

AMD'ss model allows it to cross subsidise and see off competitive threats with low pricing.

I doubt even Toyota have the volumes to justify serious custom ARM chips.

Microsoft cops to 775% Azure surge, quotas on resources and 'significant new capacity' coming ASAP


afaict, the take home is that MS cloud appeals to home workers more than Amazon?

Intel server chip shortages continue to bite: HPE warns of Xeon processor supply drought for the whole of 2020


"Intel told us in a statement:

We continue to focus on improving supply for our customers. We have invested record levels of CapEx the last two years to expand our capacity and increase our supply, but customer demand has outpaced these improvements. We continue to prioritise the production of our highest performance processors as we also continue ramping volume on 10nm and working closely with customers to align our available supply to their demand.

Rising customer demand isn't coming from the traditional server makers: sales haven't exactly been flying of late. IDC stats show that market's global revenues declined 11.6 per cent, year on year, in the second calendar quarter of 2020 to $20bn, and dropped 6.7 per cent in Q3 to $22bn.

So who is snapping up all these Intel server chips? The hyperscalers. Intel cannot make enough processors fast enough for Amazon, Microsoft, Google, Facebook, and others of similar size that are constantly expanding their data center operations. With Intel effectively auctioning off its Xeons to the highest and largest bidders, smaller IT suppliers, such as HPE, have to wait in line."

Where in the above is it shown demand is rising? Perhaps supply is falling.

where is big demand from cloud cited?

how do we know cloud isnt going short too?

We're so, so, sorry you're not able to get PC chips, says Intel to everyone who hasn't gone with AMD yet


A little discussed aspect of this, is the potential for ructions in the AIB/OEM industry.

Those rats too slow to abandon the sinking Intel ship, will suffer the financial pain of no 10nm as planned & repeatedly promised, and lost market share to more fleet footed OEMs able to deliver what customers are demanding.


"10nm is capable of making low clock speed (sub-3.5GHz - maybe less), low power CPUs in low volume. Its not going to fundamentally address any shortages before Intel's 7nm process is available. So only very limited innovations in the desktop space until 7nm."

This ~universally recited dogma is very wrong.

Intel have many problems, an important one is being at a process disadvantage to AMD, but they could all evaporate tomorrow, & their outlook would remain grim.

Their hoped for 7nm "response" in 2022, does not compete with TODAYS AMD CPUs on the inarguable metrics that decide sales: IO lanes, bandwidth, ram capacity, cores, efficiency, heat dissipation, costs, ...

What intel lack is far more fundamental than a mere fab process, and there is little hope of them having it in a realistic time frame to recover dominance, no matter what their R&D budget. Processor validations e.g. cannot be rushed and take years.

Their problem is architecture. They have no response to amd's Fabric bus and its very cost effective family of compatible modular co-processors.

AMD have effectively changed the duopoly playing field to one where they are just getting into stride as Intel begins to need expensive compromises to boast similar metrics - ~at & beyond 8 core.

7nm IF they execute ok this time in 2022, will not suddenly mean they have a 64 core answer to TODAYS 7nm Rome Epyc, let alone amd's 2022 lineup.

The only true solution is to completely demolish and rebuild, yet they barely even acknowledge the problem. They cant. Wall street & others wont allow it. It is an admission that vast sums for IP & goodwill on the books are worthless.

The good news is it needs doing anyway - to fix their unfixable security holes.

Nvidia keeping mum on outlook for year as data centre slows, channel chokes on crypto crap


Re: You sold your soul

A PA for Alexa.

Time to Ryzen shine, Intel: AMD has started shipping 7nm desktop CPUs like it's no big deal


A curious twist to Intels problems, is a recent mini boom for them, for the inglorious reason that their server chips are now up to 20% slower due to security patches, which of course means servers need 20% more chips installed to do the same workload.

This has led to some deceptive recent revenue results for intel.


Re: AMD is hot right now

Having followed amd's resurgence closely before and since their seismic Zen CPU processors, I can assure you that Intel's cupboard has already been emptied of potential countermeasures.

They have been in trouble since Zen in early 2017 & at best have mounted some decent rear guards (as you would expect after so many years of almost total domination of a conservative market) during what has clearly to insiders, been a rout.

CPUs have always been a very monopolistic duopoly (both started at the same time in the ~same neighborhood ~50 years ago), so this role reversal is one of the most amazing biz stories of all time.

As can be imagined, Intel has plenty of moats embedded in the market, chiefly control over the predominant pre-built PC suppliers. Savvy who have freedom to switch - the millions who assemble the 6-10 sub components themselves, have long been flocking to Zen - the available (mindfactory.de e.g.) recent numbers indicate an amd intel split of ~60/40 by volume and 50/50 by revenue in this "canary in the coal mine" market.

They have been fortunate to have a diminishing but plausible case among big spending competitive gamers, whose sole metric is what drives their expensive graphics cards fastest at lower resolutions. With new Zen2, neither dominates the other.

The climactic perfect storm is now upon Intel. A spectacularly superior Zen2 has arrived after they have; badly betrayed their pre-built & other partners, gouged and treated customers with contempt, had a huge fail in going from years ahead to years behind in manufacturing process and they have very serious problems arising from security problems due to ill advised hot rodding shortcuts in their past designs.

Incredible as it sounds, Goliath is in deep, deep trouble. The bigger they are, the harder they fall. Its all about confidence, and their halo is rapidly tarnishing. Even the ~PC illiterate are getting the message - Japan e.g. has always been a ~all intel, and now seems split equally in the DIY market.

To top it all, they cannot resort to the key advantage overwhelming scale advantage usually grants - predatory pricing.

Perhaps the killer aspect of Zen architecture is scalability. They can take a ~single, relatively simple & high yielding processor chip, and recombine it in a myriad of multiples to address the widely differing power levels of all segments of the market - from ~lap tops to mega servers.

Contrary to other posters here - it is in data center where amd is the biggest threat of all. There are now extremely few user cases where intel have a competitive pitch to make. Intels moats dont work nearly as well with big boys like the big 7 cloud providers.

It is this market where the real money is. Consumer PCs is a relative side show, but a very nice complement for AMD. Servers demand perfect chips. AMDs modular architecture mean lower binned or partly defective chips can be combined into competitive products. An entry level desktop 6 core cpu e.g. is 2x 4 core chips combined, each with one core de-activated. ~Nothing is wasted.

This ingenious aspect of Zen architecture gives AMD an overwhelming cost advantage. Whatever intel offer, amd simply match it with 30% more power and 30% cheaper, and still make 50%+ margin at intel set market prices.

Their failure in manufacturing has deep consequences too. They now have eye wateringly expensive plants unable to produce competitive product. This, at best, underutilised capital, is a huge burden for them.

While they had a technical lead, owning factories gave them huge market power. Now they are years behind specialists like tsmc and samsung, who do subcontract manufacturing for the likes of amd, the factories are a huge liability.

All intel can do is concede market share in an orderly retreat, & keep as much of their high margins as they can for as long as they can - keep that share price propped up til insiders can sell their shares w/o a panic.

HP boss: Intel shortages are steering our suited customers to buy AMD


Re: AMD is cool again

My response to your post and the above series of similar ones on basic office PCs is ~"well yes, but they are a false economy to implement on scale".

as a preamble - AMD have an excellent minimalist solution in their zen/vega APU. Extremely competent, but not over priced over kill.

Many of these basic rigs have also had an obtrusive discrete GPU added at some point.

The list of improvements a sub $100 APU offers is as long as your arm.


Intuitive GUIs - graphics are useful.

smooth operation causing fewer distractions from the thought process.

The perception is that support will be a challenge, but the reality is that the pervasiveness of zen/vega in so many tiers of the market (from mobile and embedded, all the way to exa servers), and the unified sibling processors from a single source, makes support much simpler than any intel/nvidia assemblage.

This last point relates strongly to the article - even if clients are only being driven to tentative steps toward a plan B, it will wise them up about a lot of intel myths when they test the water.

Having swallowed its pride and started again with 10nm chips, Intel teases features in these 2019-ish processors


Just doing what intel and others have long done with perceived threats - wax lyrical about fanciful things on the drawing board, as if they are just around the corner, in the hope of spoiling sales.

He i s really talking about where they may be heading now if they had 10nm - BUT THEY DONT.

It is compounded by the need to reconstruct much of what they have done in the past.

Start trek, the next generation: PCie 4 flash controller demo flaunts speedy peripheral vision


Who needs it? Oh wait...

There are apps with an ~insatiable appetite for ram, and many more would code for if it were mainstream.

Dram is expensive & a bit thirsty.

We can page memory overflows to storage to give an illusion of large memory, but historically, there is such a chasm between memory and storage speeds, it is of very limited use.

Yet when the possibility of storage arrays almost as sequentially fast as memory of not so long ago arises (i.e. ~unlimited, cheap ~memory), its a giant yawn apparently.

And here's Intel's Epyc response: Up-to 56-core, 4GHz 14nm second-gen Xeon SP chips, Agilex FPGAs, persistent mem


Re: True story*

Maybe it's just me, but this seems profound - a kinda Rosetta stone of Intel management logic. If reality doesn't suit their model, assume it isn't there and proceed.

Epyc move: Supermicro plunges into Cascade Lake’s Optanical waters


Re: Revolution/Evolution

I agree optane isn't the magic bullet intel make it out to be. It expensively has a read iops/latency advantage over regular nand nvme of 2-3 times it seems, but not much else. It remains a fraction the speed of dram, an advantage non-volatility power savings wont overcome imo.

Regular nand still seems to have plenty of room to grow, with smaller nodes and pcie4, or using memory channels similarly to intels proposal.

Amazon's homegrown 2.3GHz 64-bit Graviton processor was very nearly an AMD Arm CPU


Re: Expensive? We'll never know.

"no one actually says that Annapurna managed to meet the same performance requirements." - yes I posted similarly a bit after you. I don't know when it was said, but they bought Annapurna in ~jan 2015 for ~$350m. They may have realistically realised "so that's what arm yields, & we still wanna play with it, but we will buy our own test kitchen where we have full control". Annapurna may have other revenues/prospects, which would make the price chump change vs their intel cpu spend. If it adds intel leverage, it pays for itself on that alone.


I declare I am an inexpert amd stock holder, but hopefully objectively, I agree with others here that its just a plan B for AWS to use as an "arm twister".

They said amd's arm didn't live up to expectations, but nor does this it seems.

It bears noting annapurna cost ~$350-370m back in 2015


I think they just got a better deal this way than getting much the same arm twister from amd. 350M$ isn't huge considering their presumed cpu spend - they just bought their own designer team & arm licence.

Its just another logical step in the very slow continuum of the arm solution finding a problem to solve in exascale.

Instead of using amd's workshop & team for their experiments, they got a bargain on their own setup, team & arm licence. Their own ~test kitchen, just as arm instances seem to be for their clients to experiment on.

Nobody seems to be saying arm just took a big leap vs x86.

Hmm, there's something fishy about this graph charting AMD's push into Intel's server turf


Re: Lisa Su Comment

To this layman, it seems common sense it is more like a snowball than linear, more so given the nature of the whole server ecosystem.

It was announced in march 2017, but only publicly available recently. Its a slow process while everybody tests the water, but when a few take the plunge, many follow.

Of course it is exponential.

A bit of intel on AMD's embedded Epyc and Ryzen processors


not noted afaik, is that the embedded version of the zen/vega apu, is officially rated for 3200 ram. afaik the desktop apu is only rated for 2933 ram.

This could indicate the situation has improved generally for apuS, and that more ambitious mainstream ram clocks are doable.

Dell goes on Epyc server journey with AMD


As we know, buyers have different priorities, but I think the largest group are focused on IO, and in this respect, intel is lame on single socket rigs. amd's 128 lanes wins hands down.

The most exciting improvement in IT has been nvme, but they are very lane hungry.

AMD scores EPYC gig powering new Azure instances


Cloud providers are in the biz of selling VMs.

The biggest reservation about cloud by far, is security. Even priority over perf/$.

AMD security is inarguabley better.

So providers find epyc vmS are disproportionately popular with existing clients, and new clients now feel secure using cloud.


Yes its a foot in the door - of course, given its freshness.

Its one among many vm offerings by azure.

But what a foot and what a door?

It seems very liberating for system admins, - do a proof of concept on alternative amd processors so painlessly.

Web uni says it will get you a tech job or your money back. So our man Kieren signed up...


Re: "...a good percentage of the population is not suited to programming..."

Then I presume you could devise a test for schoolchildren, which could identify those with the innate gift of which you speak?

Surely; we, the kid and his family are better for knowing early that this path is an option to be nurtured?

Azure fell over for 7 hours in Europe because someone accidentally set off the fire extinguishers


Some failsafe clouds are more failsafe than others - G Orwell

In the Epyc center: More Zen server CPU specs, prices sneak out of AMD


Very helpful, Ta.


"AMD has split its Epyc SKUs into dual and single socket classes – they can be used in either configuration, though, unless they are a P-coded SKU, and a couple appear twice because they straddle both classes. So, for example, AMD recommends using the 7301 in a dual-socket system as an alternative to a pair of $800-plus Intel Xeon E5-2640 v4s, and the 7551P in a single-socket server versus a pair of Xeon E5-2650 v4s."

I cannot find clarity on this. Surprisingly, as its a very common scenario I hear - future proofing with an unused second socket on the chosen mobo.

Part of the reason for 2P popularity is simply adding more cpu grunt later if needed.

The epyc options in this regard are far from clear to me.

NVMe SSDs tormented for months in some kind of sick review game


For a very different nvme milleieu, howzabout local raid0 nvme storage on a gpu?


AMD Llano vs Intel Sandy Bridge


Not having read HP specs, it seems the intel is much heavier from the performance per pound bars. If so, can only be explained by bigger batteries methinks.

A choice between ~30% better cpu vs kick ass battery life (no gpu discrete), lighter & 200% better graphics. I know which I & I suspect most would prefer.

Integrated usb 3 is a big plus w/ amd chipsets too.

Assuming 500 vs 630 pounds is right - some say its a pittance over life of the unit, 2 years is not uncommon for these - when you consider the future worth of that money, its a big chunk of your next , way cheaper & better mobile device.

Due to a shortage & huge demand, amd prices are at a premium. expect lower once sorted.

Not sure what ram used but NB - going from ~1333 ram to ~1600 ddr3 pays huge graphics dividends. If you add a similar graphics radeon discrete GPU, expect a 75% graphics boost on amd APUs. There is no dual gpu mode for intel.

intels mantra about grunt cpus is getting as tired as muscle cars. Vast majority could do w/ a 10 yr old cpu for what they do - not so w/ graphics & power consumption tho. We are at a watershed where AMD/ATI can make intel dance to their tune cos no way intel can compete w/ ati graphics for years.

If, as you say, you are smart enough to need cpu grunt on a mobile, u should be smart enough to offload it to a server on the cloud or at the office.

AMD may be a bit slower on cpu, but intel grinds to a halt on some apps on its igp. I know which would annoy me most.

Wasnt there an obscure scandal about intels igp cheating at graphics conversions - they are lossy & crappy?

AMD targets tablets and cloudbooks in Intel showdown


hope its ok to link - v on topic



good thinking

"supplementing or replacing x86 cores in Fusion with ARM. That would still leave AMD owning most of the silicon on the die. Almost all Intel's value is the x86 core (no one willingly chooses an Intel GPU) so its not as attractive an option. AMD have more choices here -> can adapt faster."

good thinking - x86 drinks power i hear. Dont abandon it, but use it as little as possible.

My posts elsewhere here, fancy similar logic for servers. AMD cloud servers w/ 2 x cpu & 8 x gpu on 1 die, running open cl for the grunt jobs. The x86 cores act as traffic policemen & run legacy code.

Power is a big problem for these farms. A new approach is needed.


brazos is 40nm - ok but not great - same as amd GPUs, about to go 28nm - brazos next?

brazos is 40nm - ok but not great - same as amd GPUs, about to go 28nm - brazos next?


if thats brazos vs current similar intel offerings, thats not what I heard.

GMA? not sure? but if thats brazos vs current similar intel offerings, thats not what I heard. Yes, the x86 benchys are worse, but the overall user experience is way better.


general amd stuff

I just bought AMD shares, minutes before reading this.

intelligent story & comments.

"vital that Intel and AMD make the shift, tapping into their entrenched positions in PCs, while refusing to be dragged down by the legacy."

well put. Odd for me to a bit negative, but amd especially so cos of their discrete gpuS. Fusion will slaughter the low end gpu business.

We all seem agreed on the merits of fusion, now if only AMD could do decent cloud servers. Thats where the money is.

How about a dual cpuS w/ 8 gpuS fused in, running open gl server apps.

As another said, brazos is a good product and selling now. Its 40nm, same as the discrete gpuS, which are due to shift to 28nm (brazos next?).

Hope it wasnt here I read it, but they got ~124 design wins for brazos & 145 for llano. Sounds impressive anyway.

AMD dont care much about the market, the real action is stealing share from intel, or new value conscious markets (india/china etc)

I read 400 smart phones or 120 tablets = 1 new server on the cloud.

Also good money in emerging markets tho, & AMD can kick ass.

Just a guess, but a hybrid lap/desk top llano which can handle a few hours of power outtage may sell well in 3rd world etc.

SB seems foolish to me. Who would want a grunt CPU & crap integrated graphics? Those guys go the discrete route.Hope I dont sound a shill. Just sayin