158 comments

[ 2.7 ms ] story [ 181 ms ] thread
Simple, buy AMD
Not so simple when you want to buy server quality motherboards.
Possibly, assuming they make something that fits your needs. The market is really shallow at the moment.
The other issue is buying the chips themselves, they're only available to OEM partners right now AFAICT (not listed on Newegg or anywhere else I'd expect to be able to buy retail boxed server CPUs).
Exactly. If you just sit and play games and don't care if your computer kernel panics and don't care if bits flip in your memory (with no ECC) then AMD is for you!

But if you're doing real work, don't bother.

Not in channel yet. Intel are making hay from those customers who need to buy in the next quarter.
In an interesting (to me at least) move intel instead of competing on price when AMD put out a competitive offering they decided to keep prices the same more or less or increase them by adding value. Faster AVX units, more cores, more SKUs (not sure about that last one). And while intel has margins to maintain AMD has put features into their chips for years. For example AES has been shipping in most every desktop and server CPU while only some intel ones get it. Most AMD chips can do VT-D etc while the same feature is relegated to higher end chips generally on intel.
I could be wrong but I assume that there's enough inertia in the high-end business-to-business side of things (the part that matters for high-end server processors) that they probably can keep milking that cow for a few years more.

People who have only been buying Intel CPU in the last decade won't jump ship immediately. Nobody gets fired for buying Intel CPUs, at least until very recently. You know you won't have compatibility issues etc...

Of course within the next few years Intel will have to react somehow, either by outperforming AMD significantly to justify the premium or by slashing prices.

I know I hope AMD can keep up the pressure and keep iterating.
> Nobody ever got fired for buying IBM.

Replace IBM with Intel for the 2010s version.

And yet there are still people out there buying IBM and somehow not getting fired for it.
Do you frequently see people getting fired for purchasing decisions?
Systems Engineer here;

Absolutely.

Good example is that we bought 1,000 servers, 256G memory, dual socket Xeon E5-2687w v4 etc;

The things didn't come with a TPM module and it was a business requirement to run:

A) Windows

B) Full Disk Encryption

TPM modules were not available to us in such quantities.

So, um, ... what happened to those now "useless" servers. Wondering for a friend ;-)
Available software constraints choices in many of those cases. Put two choices in front of an application team exec: rewrite and port an application from POWER/AIX, POWER/System i, POWER/System z for $X million, with all the attending risks that entails. Or spend same $X million on new IBM hardware, and get more capacity, performance, etc., while the system administrators/system programmers update the underlying OS, with lower magnitude of risks and impacts, and if the sysadmins screw up, that's another department, not yours, anyways. Even if porting is a one-time charge, and saves Y * $X million in N years (where N usually <5).

The choice to avoid risk is easy.

This is why, by the way, selling to C-suite, or above middle management, as much as it is hated in technical circles, is used. The higher you go in management, the more risk is able to be shouldered.

If Intel hypes up any AMD compatibility issues Intel can find, then that urge to avoid risk will play to Intel's advantage.

Port away from? We moved everything TO POWER/AIX.
If its not confidential, could I ask why? IBM has been pivoting POWER to Little-Endian and has been focusing more on Linux with Red-Hat, SUSE and now Ubuntu available.
Power is big iron reliable. if you can pay for it it likely pays off. Probably easier to manage a handful of million dollar Power/AIX setups than 1000s of servers.
I don't follow, this is the same POWER 'big iron' just running Linux instead of AIX.
The hardware is where the magic is not the software.
To complement siblings answer, there are several types of failures where PC hardware will just crash and burn, while POWER will just work as usual.
Yep, I've seen that, too. There are some use cases where IBM still excels. There is a popular perception that IBM is a has-been dinosaur, especially with the System z (mainframe) stuff, but I still haven't heard that Visa International has moved even one of their three world-wide super data centers off of that platform. The Linux world could avoid reinventing a lot of wheels in the future looking to the lessons IBM has publicly published about in the mainframe world. Even POWER/AIX is scoffed at in many Linux circles.

I've certainly seen my share of underutilized System p servers, but in the right hands, they're a powerful option, with features that Linux still hasn't caught up to yet. For a cool feature that I would love to see on Linux someday, peek at checkpoint-restart, which enables some really sweet capabilities [1]. Generally, anytime I hear a business tell me uptime reliability is nearly a blank-check requirement (when downtime losses are described in 5+ figures per minute and up, and it is a 24x7x365 operation), but they must still stay on a Unixen box because that is what their IT shop knows, and it's not a parallelizable application, I start looking in AIX's direction. It can be a right PITA to work with if you grew up with Linux or BSD, but once you figure out some add-ons to use, it gets much more pleasant to use.

[1] https://www.ibm.com/developerworks/aix/library/au-aix7.2.1-l...

I find sad that many aren't aware how many cool features are available in IBM and Unisys mainframes, still not available in mainstream computing.

The AS/400 (IBM i) and ClearPath are my favourites.

One with its TIMI architecture and kernel level JIT, the other with an history going back to the 60's using an almost memory safe systems programming language, while people are still debating how to replace C.

Is there an article that compares the features present in Power vs x86 and this almost memory safe language?
I don't touch Aix since 2003, but really enjoyed using it.

Just curious, do they still support the Windows/Symbian style of dynamic libraries, with import libraries?

Sometimes IBM are better than the alternative. Hard to believe in the HN bubble, I know, but there you are.

(The trick is not to end up buying a bunch of rubbish IBM products while you're getting the one you want.)

I suspect anyone ordering top-bin CPU skus by the pallet-load has an ongoing program for evaluating performance-normalized TCO of at least a few candidate platforms based around a few candidate processors.
Heck, why has Google been investing heavily in OpenPOWER? I bet it's as much as keeping the option there and making it as attractive as possible: gives them a far stronger negotiating position with Intel.
I saw mentioned in another thread that all AMD processors support ECC RAM, provided you pick a motherboard that does it.
This alone could save you thousands.
Not really. Low-end Xeons with ECC cost >$300.
Sure, but with a $300 budget on Intel you're choosing between an i7-7700K or a Xeon E3-1245v6, which are all but identical.

The Xeon has a slightly higher base clock, slightly lower turbo (all clocks are changed by 0.1 GHz), ECC, and a locked clock multiplier. It appears to benchmark consistently a couple of percent below the i7.

What's even the point of that product segmentation existing? To annoy their customers? I'd rather just buy a $300 processor that checks all the feature boxes instead of all but one.

http://www.cpu-monkey.com/en/compare_cpu-intel_xeon_e3_1245_...

Best I can come up with is binning. Maybe some CPU's fail on the ECC front, so they sell those as 7700 and up the factory boost clock to compensate and 'market to consumers'.

Or maybe the MSRPs vary wildly, but in practice prices converged?

> Best I can come up with is binning

Bingo.

The memory ECC is a tiny bit of logic right next to all other memory logic in the MC, it seems very far fetched to me that there would be a binning process for specifically the instance of that logic being broke but the data logic next door working just fine.

Note that the ECC on the processor's internal data buses and caches is completely independent of whatever ECC the memory controller may or may not use.

The product segmentation is basically for different markets, but also for process maturity.

Mobile chips are very small, under powered, and under clocked. They get fabbed first.

Then your core i3/i5/i7/E3's are all fabbed. As the process matures and can be expected to regularly handle 4GHz.

Next your E5/E7's get ran, normally in a few waves. Your _enthusiast_/workstation cores are generally fabbed with E5/E7 engineer samples which get sent out for testing. These have [1] 20+ cores so you need to know your process is stable so you don't have to sell >50% of them as 8-16core Xeons.

---

But yeah the product segmentation is mostly to gate access. Only Xeons offer ECC [1].

And only [1] i5/i7 offer unlocked multiplers.

Ultimately this started to be a organization issue. Consumer sales (i3/i5/i7) doesn't even talk to enterprise sales (E5/E7). Consumer support can't answer questions about enterprise chips.

My personal opinion is the only reason Intel is still in business is they have a monopoly. Their anti-consumer practices are down right twisted.

[1] can/only are strong words there are a lot of exceptions/gotchas in Intel's product lines. Celerons/Pentiums sometimes have ECC, i3's and Pentiums have had unlocked Multiplers. There is a line of Xeons with unlocked multiplers. IDK

> Only Xeons offer ECC

> And only [1] i5/i7 offer unlocked multiplers.

These statements are false, which you yourself acknowledge later in your post. You don't get credit for noting a caveat when the caveat is that your point is entirely false, at least in my book.

Intel no longer manufactures 2-core Xeons so Pentiums and i3s support ECC nowadays. This isn't uncommon, this has been the norm literally since they stopped making 2C Xeons (Haswell?).

And there are currently unlocked i3s in the lineup, just like there have been unlocked Pentiums in the past. It's the norm that Intel offers a barnburner overclockable SKU somewhere in the low end. They don't always, but they do it probably every other generation.

There has been one Pentium and i3 that are unlocked since the start of the Nehalem era.

What's up with the i3-6320 supporting ECC but the i3-7320 no longer supporting it? Intel is at the least, very inconsistent with ECC on the consumer side.

ECC was probably not for consumers. I think they needed to have an offering for processors smaller than E3 for pro needs, and somehow merged it with some i3. Also check for embedded options: those are basically LTS, so low volume and/or high latency designed specialty electronics can use them without having to update all the time. I've not checked if there still are some offering on some newer gen i3, but Intel can very well have internal restructuring between their consumer and their pro-branch, which I remember having read are not very integrated...
I once heard there is certain mentality that when given more options, more people end up making purchases where they otherwise would not. For instance, if you have one fast food chain along a short stretch of road, less people will go to that one location than if there were multiple, competing chains in the same area.

I believe this is called agglomeration[0][1], and may have parallels beyond physical, competing stores, into realms like similar, but differently named CPU's.

[0]http://journals.sagepub.com/doi/abs/10.1177/0042098017694131 [1]https://en.wikipedia.org/wiki/Economies_of_agglomeration

edit: rplst8 mentions binning, and that seems like another likely factor as well.

(comment deleted)
Sure, if you're tinkering at home where the stakes are low and your time is free. You can also save a bunch of money running Xeon engineering samples.

But if your production environment goes tits-up then your savings go out the window and a whole lot more besides. And that's what you're risking by running a hardware configuration that is non-validated/unsupported.

People constantly make the mistake of assuming that businesses need to pinch pennies on their money-generating assets, just like they do on the rigs they use for entertainment at home. In the business world, reliability and official support trump an extra $1k in your pocket every time.

Forget Xeons entirely: there are still a lot of customers who buy Sun/Oracle/IBM hardware even though it costs tens or hundreds of thousands of dollars extra. $1k does not even register for a lot of businesses.

(ECC is validated on Epyc, of course - but the prices are a lot closer to Xeons there, it's no longer a slam-dunk. You have to look at it on a case-by-case basis to determine which is actually a better deal for your workload. And frankly AMD has been doing an undue amount of massaging on their benchmarks here, like in their launch data where they just straight-up removed 43% from all the Intel benchmarks. )

http://techreport.com/review/32125/amd-epyc-7000-series-cpus...

A fair point, and well articulated!
ECC is not an officially supported/validated feature on Ryzen. Support comes and goes across various motherboards and BIOS revisions and if it doesn't work you have no leg to stand on. Also, UDIMMs only, no RDIMMs.

It is officially supported on TR and Epyc, as well as Pentiums, i3s, and Xeons (entry-level Xeons are nowhere near as expensive as people think they are).

If you think you need ECC my advice is to build out around a platform where you won't be the first person to try something and run into the bugs. There is not a large community of ECC-users on Ryzen.

Ryzen doesn't support RDIMMs, only EPYC do. AFAIK ECC support is broken on Ryzen even with UDIMMs.
I suspect Intel is not too worried about AMD because they've seen this script before. When AMD's Opteron server CPUs came out they offered vastly superior performance and scaleability compared to contemporary Xeons. Even with a clear technical advantage, AMD never came close to matching Intel's volume in the server market.

Today, AMD's Eypc CPUs are a much closer match to their Xeon competitors than Opteron was. So if history is anything to go by, Intel has little to fear from AMD's latest efforts.

AMD wasn't able to match Intel's volume because Intel engaged in outright bribery of PC and server vendors on a impressively massive scale (they literally paid out hundreds of millions per year — and that's just to one vendor — as an incentive not to buy AMD). Other factors obviously didn't help, but this was clearly a huge one.

Given they've been slapped down for this, I wonder if they'll try it again. If they don't, AMD could have an easier go this time round.

I don't completely understand the "need" for bribery. Could they just not lower their prices? (which could have resulted in the same balance in the end)
I think you're confused about "couldn't match their volume". Parent means (I think) that AMD couldn't _sell_ enough units. Not that they couldn't _make_ enough units (which is I think your interpretation).

They couldn't sell enough because someone was paying their customers not to buy their (superior) product.

I think temac was suggesting Intel could sell cheaper chips. Why would Intel need to bribe the competition?
To maintain product pricing and value perception.
And to block competitors from selling out their own products (AMD in that case).
There are few things at play. And this time is a little different.

1. Opteron in the early 2003 to 2004 did have a good run. It forced Intel to lower prices as well as using dirty tactics to compete. Let's not forget Opteron was AMD first entry into Server market. AMD then failed to execute after that.

2. Software compatibility was a thing, and may still be a thing judging by the comments here. But remember Datacentre is now largely dominated by Cloud and Web Host Providers, and most of them depends on Open Sources Software. All of a sudden, Google Amazon and Microsoft have incentives to improve software compatibility, if there is such problem exist.

3. IO compatibility / performance used be a problem for AMD. I am surprised it is not mentioned anywhere. Intel has the best IO ever with their Chipset. From PCI-E, SATA, AGP etc. But now none of these IO implementation are secrets anymore. Thanks to the SoC IP industry you can get the best and leverage the battle tested IP from other vendors. And not to mention the really only IO left is PCI-E. Which is very well defined.

I expect there will be lots of interest to test out EPYC, but the real volume will be Zen 2, when the industry has settled on EPYC being good enough. And I cant wait to see how Intel would react, after all if you look at their recent management teams changes it seems they dont have a clue what they are doing.

I have been thinking about whether GitLab should buy the M versions that can do 1.5TB of RAM per socket with 128GB 3DS LR-DIMMs. They will probably not use them immediately though. Of course, I think the price gap between 64GB LR-DIMMs and 32GB RDIMMs are falling.
I thought Gitlab went back on the idea of using their own hardware anyway?
They can't even manage the managed servers, it has issues too many times
I am obviously not suggesting that they buy the servers now.
Server CPUs are way too cheap compared to the remaining parts of the server. Paying 10% more in exchange for software compatibility is a steal.
What software compatibility issue?
Different CPUs perform subtly different. Not to mention that server-class workloads tend to expose bugs and architectural shortfalls. Ryzen hasn't had great press in this regard, so decision makers are going to regard AMD with some skepticism to begin with. It's about risk: you know Intel chips roughly perform as they always have, just a little better each generation.

Additionally, if you rely on Intel's compiler to get more performance out of your chips, well, switching to AMD doesn't make a whole lot of sense.

AMD needs to be back in the server space for a while before they'll be trusted again.

What's a server class workload in this context?

Remember that Intel's embarrassing FDIV bug was spotted in Excel...

I agree with most of what you said, but Kaby Lake had issues with Hyperthreading, and Haswell had to fully disable the TXT extensions, both modified (fixed/disabled) through microcode updates. You could say Intel hasn't had great press either.

That being said, I still agree AMD still needs to prove their new platform in the server space.

For what it's worth, it's TSX (hardware transactional memory), not TXT (secure firmware). It's a very niche feature, compared to hyperthreading.
Don't you know AMD and Intel aren't compatible.
Not the OP, but maybe he's talking about the kernel panic that some users are getting during heavy-duty compilation workloads.

https://www.phoronix.com/forums/forum/hardware/processors-me...

Newer steppings of Ryzen have a fix applied, and AMD is allowing people to RMA those processors affected. Threadripper is unaffected by the bug.

Source: https://www.phoronix.com/scan.php?page=article&item=new-ryze...

AMD has never stated that there is a fix for this. If you look at the link you posted, this is Larabel making this claim on the basis that he hasn't seen any forum posts noting this problem on samples manufactured after week 25. Quote:

> AMD has not provided an official public explanation of the fundamental problem, but from those in our forums and elsewhere, it appears to affect Ryzen CPUs manufactured prior to week 25.

This problem only affects a minority of chips in the first place (likely a litho flaw in the cache section), it's probable that Larabel's second CPU just didn't happen to be one of the ones with problems. And Week 25 ends June 25, i.e. stock for these date ranges has hardly even made it through packaging/distribution to end users let alone been extensively tested yet.

Larabel is jumping the gun on insufficient evidence and people are racing to make the claim that AMD has fixed this, when that's far from the case. AMD certainly hasn't made any statements to this effect.

AMD is RMA'ing afflicted units, which is really all that can be done until they get a new stepping out. But I've seen no evidence that this is actually fixed.

> This problem only affects a minority of chips in the first place (likely a litho flaw in the cache section)

What are you basing this on? I've seen no good reporting on this* but the AMD Community segfault thread (https://community.amd.com/message/2796982) seems to suggest that almost all Ryzen CPUs manufactured before the end of June (which I assume is most of the current stock) are affected. Very few people load their Ryzens sufficiently to hit this bug, but a large number of those who attempt it seem to get the segfault.

* I would like to know the percentage of processors with the bug, percentage of review units with the bug (I would be very interested in this), and maybe some long-term testing of "bug free" processors etc.

It may be ridiculous, but I'm not spending a million dollars on CPUs with information gleamed from a forum post.
It's not ridiculous, but there's pratically no more information about it. No major tech sites have covered it, beyond Phoronix. AMD haven't acknowledged it, beyond stating that "a performance marginality problem exclusive to certain workloads on Linux" exists, which is factually incorrect as it also seems to occur on other operating systems.

I'm not spending hundreds of euro on a processor without knowing something more concrete about this issue. If you budget is really ~million dollars, do you have additional information about this that you can share?

My ~$400 CPU budget went toward a Skylake i7 a couple of years ago, I'm afraid.
there is no kernel panic whatsoever. the mentioned issue causes some of your GCC parallel compilation jobs to fail, everything happens in user level. This issue is caused by manufacturing, not by design, it has been fully addressed in recent batches.
(comment deleted)
At least there are some linux kernal support issues on AMD Ryzen and Threadreaper CPUs for now.
There is a fix for Ryzen in newly shipping chips which fixes the gcc compilation issue on Linux. From what I understand if you are facing the issue you can contact AMD and they will ship you a new chip without the issue. ThreadRipper / Epyc never suffered from the issue.
A several people have replied to you by mentioning the Ryzen kernel panics. That's a thing and it's getting fixed by hook or by crook, but it's not everything. Here[1] is an example; Ryzen only received no-workarounds-needed support for ESXi 30 days ago. Another[1]; you can't monitor CPU temps in Linux yet for lack of driver support. That sort of thing is a non-starter in a cloud data center when you're managing megawatts. The truth is that the 'enterprise' has been wrapped around the Intel axle for years and AMD has a lot of catch up to do before it has parity with Intel.

There are enterprise people that will adopt AMD for the shear $/compute value, but until you can find rackable Dell SKUs and Supermicro server motherboards with AMD sockets and the whole 'enterprise' software stack is fully ironed out Intel can and will command a premium. It will happen, eventually (assuming AMD continues to compete,) but it's not here yet.

[1] https://www.reddit.com/r/Amd/comments/6que0q/esxi_65_now_wor... [2] http://www.phoronix.com/scan.php?page=article&item=amd-ryzen...

Intel agreed to license the x86 to AMD decades ago, it is the precondition to allow Intel to avoid being sued to bankruptcy for monopoly. For current software, the current arch is called AMD64 for a reason - the mentioned compatibility is built entirely on AMD inventions.
Not only to avoid enforcement, but also because some customers will only select products where a https://en.wikipedia.org/wiki/Second_source is also selling them.
Well that's stretching the meaning of second sourcing quite a lot. Normally it means that you can get an exact replacement part from someone else. For example, industry standard things like nuts, bolts, ball bearings, SRAM and DRAM (most stuff specced by JEDEC) can be sourced from many sources. There is literally no difference at all between a M8×20 8.8 bolt between manufacturer A and B. Similarly, if you have a design with a LM358 (a non-standardised part), then there are multiple manufacturers that will sell you a part with identical specs under a slightly different name or even the same name.

This is also the way it started with x86 and other second-sourceable processors. An Intel 8086 and an AMD 8086 are drop-in compatible. This is not the case with modern processors, they don't even fit the same sockets.

But you can change a larger module. If you want to cover that kind of risk with that kind of supplier companies (hint: a marginal risk, to begin with), you are not going to go anal about the drop it characteristics of single chips. Or at least you should not, IMO.
Until your single-source vendor decides to turn up the heat. Realistically if Intel doubled all prices tomorrow most enterprise customers would just pay it.

The huge cloud vendors should be buying and deploying AMD systems right now to ensure that never happens; it doesn't matter if there are some kinks to iron out. As an insurance policy it is very cheap.

Ah, the classic "computers are cheap, developers and software is expensive" argument. That may apply for small companies, but when you've got 100's of millions or billions of dollars worth of hardware, 10% starts to easily be worth the developer investment.
With the concentration of server buying by AWS I am surprised Intel believes it can get away with this. AWS has the expertise and the incentive to push for supporting a competitor that offers a superior value.
I'd bet that the cloud-scale customers like AWS negotiate pricing that has very little to do with the published list price of these CPUs.
More than that, Intel builds custom chips to AWS specs.
Kind of. From what I understand, any customer large enough can get small features added, and they just get turned off for the other customers.
Depends on the feature, as far as I can tell. Intel has been known to do entirely custom chips, if the price is right (and realistically, you need lots of volume to be able to justify the setup cost of it).
If we are talking about modern processors, I doubt they are going to do custom masks.
There's no way the price is right for custom Skylake Xeons, especially considering how easy it is to disable features on chips after you make them.
And I think this is possibly one reason for the pricing. When a >50% (guessing, any better numbers?) of new chip sales volume comes from a few hyperscale buyers with individually negotiated contracts, you're far more flexible with what you do for the remainder.

Heck, Intel might even not really have the production volume on all SKUs to sell to non-hyperscale! And if that's true, then why would they do anything other than milk the few customers it deigns to sell to?

That could be one reason why it's priced so high: because they wouldn't be able to make enough to meet demand otherwise. They can price it extremely high to keep demand under control, and negotiate special pricing with a few selected customers that they really care about pleasing.
What % of Intel's sales by volume do you think goes to AWS, GCP, Azure?
While AWS is one of the largest buyers, I doubt that they have a significant share of Intel's profits. There are probably a handful of companies buying (nearly) as much and sales will be dominated by small datacentres and enterprise customers. A lot of Xeons don't end up in servers but in workstation computers.
From the reports I've seen, the big vendors (HP, Dell, Lenovo, etc.) ship about 3x more servers than workstations. And that probably understates the degree to which Xeons end up in servers rather than workstations since HP, Dell, and Lenovo account for a significant percentage of the workstation market whereas there are a lot more vendors and custom-built hardware in the server/datacenter markets.
Given AwS's leadership in being the world's datacenter, they must purchase enough Intel CPUs to have some say, so I'm surprised we haven't seen Amazon offer of CPUs clocked at 10 GHz.

Home users don't have liquid nitrogen piped into their homes or even really the expertise for hot vs. code aisles, but AWS is building their own datacenters, at which point they could pipe liquid nitrogen in if they chose to. At the scale and concentration to AWS is building at, I certainly hope it leads to faster computers, if not bigger AWS bills.

The way I recall, AWS had to be embarrassed into offering GPU instances rather than taking the lead in, so maybe there's a competitor out there building for speed.

Check out Softlayer / Bluemix we have customizable bare metal offerings.
And that if case and point of the problem we face with the sheer size of AWS.
Most markets feel comfortable and safe with Intel.

AMD will have to go through the tough adoption curve of early adopters to mature markets and while this happens Intel has relatively free reign with their pricing and value.

Markets with long established leaders have the additional burden of too many vested interests, misinformation and dirty tricks. However if AMD continues to deliver the naysayers will eventually lose credibility.

AMD fans like to separate RTG from AMD, but anecdotally, I just got a Vega 64 GPU and it's horrible. Loud, constant crashes, my desktop freezes when a video starts playing, etc.

Stuff like that makes me unwilling to go with AMD when I have a choice. I view them as offering competitive performance on paper but completely falling short when it comes to reliability.

On another hand, in the cpu space, I went to top of the line Intel CPU overheating and throttling, to 8 core Ryzen, doing 45 celsius at 100 percent load and almost 4ghz across 8 cores.

Stuff like Intel unable to perform below 80 degrees on 4 cores, makes people switch with ease.

> I went to top of the line Intel CPU overheating and throttling

Frankly I've never heard of a 6950X overheating and throttling, the lid is soldered on those processors. Sounds like a personal problem.

I wasn't taking about the lid. The raw thermal output is different under a full non-transient load. Why the cynism?
Real-world power consumption between the 6950X and the 1800X aren't supposed to be terribly different either. What did you measure before and after?

https://www.bit-tech.net/reviews/tech/amd-ryzen-1800x-and-am...

https://www.pcper.com/reviews/Processors/AMD-Ryzen-7-1800X-R...

> Why the cynism?

Because in performance and efficiency Ryzen is basically equivalent with Haswell-E/Broadwell-E. The sole advantage it has is that the 8C Ryzens are cheaper than the 8C and 10C Intel SKUs (the 6C have been competitive for a long time), and the tradeoff is you are moving to a much more immature platform.

Claims like "oh my Ryzen was so much cooler than my 6950X" set off my bullshit detector real hard because that's not what the people who are actually measuring things are saying. They usually end up being made on the basis of the TDP figures, which AMD's marketing department has massaged significantly downwards from their actual measurements.

Ryzen is basically on par with Haswell, it's not the second coming of Jesus unless you've been in a coma since Bulldozer was released.

I think you might have forgotten to type a post to go with those links.

Later iterations of Ryzen are undoubtedly going to fix some of the teething issues (segfaults, memory clocks, etc), and probably get the clocks up too. But they will also be going up against Coffee Lake and others - Intel can iterate on their products too.

Consumer Ryzen is basically just Haswell with an AMD sticker on it, and it's just so transparently fake to see all the fanboys cooing over a product they wouldn't even give the time of day to when it was released under an Intel badge. It's an incremental improvement on pricing, but not really that drastic except in a few special cases (the 8C and 10C Intel processors had very steep pricing).

As it's on a newer stepping with a better platform and fully-validated ECC support, Epyc is much more interesting product to me. But it's also a market that doesn't care about throwing money at the problem if that's what it takes to get performance/stability - there are plenty of companies who are not even using COTS hardware let alone off-brand.

Also, bear in mind that "nm" measurements are pure marketing. It's now the spacing between features that determines total density, not feature size. Intel is at least a full node ahead of everyone else in density, even if others claim to have "14nm" nodes.

(comment deleted)
If you change the thermal resistance, you can very well overheat or not with the same constant TDP.
The top of the line Intel CPU comparable to an 8 core Ryzen CPU is probably the 7700k, which has a higher absolute power draw "ceiling", but also has better performance and higher clocks to match.

I think Intel's value based on performance isn't good compared to Ryzen, but the product isn't inherently flawed, and that combined with AMD's shaky history is enough to keep myself, and lots of people I know from switching. (Right now stories of RMAs for the Risen GCC crash bug taking ages are popping up)

7700k is an awesome CPU. I just prefer 4 more cores at the $400 mark price, and was willing to risk it :)
7700k seems kinda overpriced. Why not pay $40 more for the i7-7800X and get 50% more cores and double the memory bandwidth/channels?
Plenty of reasons:

Higher base clock on the 7700k, biggest reason for entertainment alongside:

Higher OC potential (back in the 4790k days 5Ghz on air was something else, for the 7700k not getting 5Ghz is considered "losing the silicon lottery")

Cheapest 7800X motherboard is 100$ more than cheapest 7700k motherboards (which are still quite good) because X299 is a HEDT platform

Most entertainment machines won't benefit from quad channel over dual channel in a meaningful way.

The 7700k is generally regarded as the fastest gaming CPU available, that's why so many benchmarks use it. I'm pretty sure AMD used it to showcase Vega because even if it becomes a bottleneck, no other CPU on the market can be "less of a bottneck"

And I'll add, this is all with my entertainment/gaming hat on.

For a dev machine where I'd expect to work with compiled languages or compiling OSes I'd definitely go with more cores.

> constant crashes, my desktop freezes when a video starts playing

my r9 270 started doing that and died completely (as in, stopped POSTing) a couple weeks later. i'd try to RMA.

The freezing is well documented with Vega 64, it happens stops when Hardware Acceleration is disabled but I can't (easily) change that in anything other than my browser

The crashes seem to be manifesting the heat issues it has. Even at stock voltage and power it hits 80 degrees and starts running into stability/throttling issues

Vega 56 and 64 would be decent, but AMD basically overclocked them way too far past the point of diminishing returns out of the box, in order to make them, on paper, beat the GTX 1070 and 1080 respectively. If put on a power-saver profile, they're suddenly much more respectable. But AMD keep gambling that a loud, power-hungry and hot by default card will do well if it narrowly beats the closest NVidia card in benchmarks. They should perhaps stop making this gamble.

Or so this guy claims: https://www.youtube.com/watch?v=AavuWT17X48. Seems to be borne out by reviewers' general impressions though.

What I've seen points more towards Vega 56 actually not being a bad value all things considered.

But Vega 64 is overvolted severely to get to the 1080's level of performance so it runs hot and loud. Some people have been able to undervolt their cards until they reached reasonable power draw for the performance level, but AMD wouldn't set a high voltage for no reason, implying they did it to improve what percentage of their yield could be shipped as Vega 64

Q: Why is Intel charging so much?

A: Because enough people pay.

Q: Why do people pay so much? Why do they pay a high premium for top performance?

A: Because, considering all factors, it's worth that much for them. For many it's too much but for many others it's ok.

Q: But it's the same silicon!

A: Manufacturing cost plays little role when the price is not dictated by a race to the bottom between different manufactures. Instead, marketing, market segmentation and customer value are the cost determining factors.

That's really not what the article is about at all. It's talking about the history of the tech industry and associating this kind of price motion with "peak" revenue for given hardware products.

Your same answer would apply to IBM's mainframes in 1990 and Sun's servers in 2001 too, and the point of the article is to associate Intel's current position with those products, which clearly reached peaks at those moments.

Personally I don't really buy it completely either, but quipping "because the market!" as an answer is missing the point. The article is about what that market desire means for future revenue.

Another example missing in the article was the graphics workstations with SGI.
I remember considering spending lots of student-budget money in 1993 for an SGI Indy (https://en.wikipedia.org/wiki/SGI_Indy), a low-end graphics workstation from SGI. I think this was shortly before peak-SGI.

I'm glad I didn't, because (a) PCs started to see impressive graphics performance improvements not so longer after; (b) I went to industry and ended up working in enterprise software instead.

SGI market cap peaked in 1995, here's a HN discussion of a 1997 article of SGI's "Sad Saga": https://news.ycombinator.com/item?id=8462841 . They shifted focus to scientific/enterprise compute and storage after and dwindled away.

Exactly, and for pretty much the same reason. Once CMOS prices got to the point where 3D graphics acceleration (not SGI's only core competency, but pretty close revenue-wise) was desired at consumer volume, they were toast. NVIDIA had they beat on almost everything that mattered by 1998 or so.
I remember when my scientician friends started switching from SGI to multiple super-beefed-up Dells in the early 2000's. They could get five monster custom Dell machines for the price of one Octane or Indigo or whatever. And they delivered better performance. (Or so they told me.)
There was about two or 3 generations of either PC hardware or software (Windows) not being able to handle what SGI hardware could, no matter what the price.
Don't forget the nail in the coffin: the windows nt-based SGI workstation...
I want to introduce the factor of Intel being protected against the chinese chip manufacturers possibley having an unintended consiquense of supressing state side competition.

In general, I cannot imaging running an AMD processor in a flagship. Intel is an example of superior architecture and amazing developments with, most recently, the Core i productline, more impotantly, the most amazing, brand identity positioning i have ever experienced.

AMD very possibly could work well; I will never know. ... No really, i have to go buy an AMD just to keep an open mind.

One thing missing from the comparison to IBM and Sun is that precisely because those companies made things so expensive people started looking for alternatives.

The ES/9000's were really good machines for the time. Had IBM not raped people on those, companies might have stuck with them for a little longer. The AS/400 was priced more reasonably and people loved those. You couldn't get someone with an AS/400 to move to a Unix workstation if you used a nuclear bomb.

As for Sun in 2001, the problem was that they were so absurdly overpriced and underpowered that commodity x86 simply blew it away. If I didn't have direct knowledge of the dysfunction that was Sun at the time, you would have thought they were taking kickbacks from Oracle, Cadence, Synopsys, etc. to limit performance. (Explanation: Oracle, Cadence, Synopsys, etc. all had very expensive software packages that were licensed per processor core per year and effectively only ran on Sparc. So, the lower the performance of the processor, the more cores you needed and the more expensive your license to Oracle, et al.)

Even the IBM machines were flagging in performance in 1990. That was the beginning of the RISC revolution, and all the benchmark records were being taken by MIPS (and later Alpha) boxes. It was a slower process than the Intel takeover[1] only because ECL logic was still available to expensive machines, allowing clock rates in the mainframe world to stay higher than CMOS chips even as their architectures got eclipsed.

[1] Ironically on the back of a "CISC" architecture. What happened there was that CMOS density had gotten so high that the relative advantage of a RISC ISA (not architecture: every thing form the P6 onward looks like a pipelined RISC box on the inside) was just a pipeline stage or two and a 1% hit in the transistor budget for decode.

I own two SPARC servers (a 2003ish V245 and a 2007 T5120) and both are rather ridiculous ... idle power draw for a V245 is like 240 W and just 150-180 W of that are the fans. [1] A T5120 draws some 180 W idle as well. Especially the latter was also heavily marketed as green computing / efficient data centres / throughput computing. Completely ridiculous offering at the time.

Needless to say, I didn't even bother to set these up lest build a rack for them (my normal rack is both full and too short). If packaging weren't so annoying with computer servers I would've sold them long ago.

[1] The good stuff. High static pressure, high flow fans from Delta with ball-bearings, and of course only speed monitoring, no regulation. These Sun servers make for excellent magnet boards for holding papers just without using any magnets.

> You couldn't get someone with an AS/400 to move to a Unix workstation if you used a nuclear bomb.

Still can’t. The new bottom end model (iSeries) is cheaper (sub 10k) and will last for another 10 years until a new one is needed. The place I work is buying a new one because it is hitting the end of support (damn thing was bought before 2008). No problems that lasted more than a couple of hours in all that time.

Interesting, seems like IBM still has a couple of competent departments surviving the rampant idiocy.

If you don't mind my asking, what tasks do your folks have the AS/400 doing nowadays?

"raped" is not an appropriate word to use in the context of this discussion. Please try to be more considerate.
"I bought an IBM/Sun server and the salesman ripped my face off". Should adequately describe server pricing strategies.
Dipshits like you are the reason why a word from 1563 which has no relation to a similar sounding pejorative can't be used in modern parlance.

https://en.wikipedia.org/wiki/Controversies_about_the_word_%...

As such, I feel no need to censor my language to meet your lack of understanding--to paraphrase Julian Bond, the chairman of the NAACP.

I also am familiar with the concept of markets.
btw it there a name for studying the value of "price" ? or is it just economics ? the history of the difference between cost (material, employees) and sale price.
I've got at least one book titled price economics. Basically that's a huge area in economics. There are both micro and macro contributions to price. It's also the one communicative factor between suppliers and demanders.
Price theory. Microeconomics… which is largely how a single actor makes a decision, whether to buy something or where to set a price, and how that leads to market equilibria when many people do that in some utility-maximizing manner.
Also, does this include the psychological effect of price ? selling at 0$ doesn't mean people will grab it, they might assume it's crap.

Thanks nonetheless to both you.

I think a question like that would fall under the umbrella of "consumer behavior".
That's not always the case. Sometimes monopolies are born and the companies charge more money out of arrogance that there isn't anyone out there that can "truly" compete with them. But that same arrogance is often what ends up bringing them down.

See Nokia and BlackBerry's arrogance in relation to the rise of the iPhone. They were still charging similar or higher prices than the iPhone for their flagships a few years after the iPhone first appeared, and with no touch features to speak of or anything close to an iPhone experience.

The inertia of success definitely plays a role. For instance, BlackBerry was logically concluding that they "must be doing something right" a few years after the iPhone came out because their financials still looked pretty great, due mainly to the global expansion (the iPhone wasn't in that many countries at first, and Android still had single digit market share). So that's how they rationalized keeping their high non-competitive prices.

I assume something similar is going on at Intel. They continue seeing a large influx of server buyers even with AMD's EPYC being out because of inertia - customers already deciding to buy Xeons in 2017 a few years ago - like say Google which bragged so much about being the first to get a Xeon Skylake. So Intel rationalizes this as "having no reason to reduce its prices, and in fact it could even increase them since it sees so much demand!"

But that logic is going to hurt them a few years from now, when AMD or perhaps some ARM competitors like Qualcomm start becoming more established in the market and they've had a few years of "people hearing good things about their products." Right now the new AMD and ARM chips are untested and the majority of companies like to play it safe.

You're not really being fair on Nokia or BB; Nokia had an iPhone competitor when they added phone functionality to their Maemo 'Internet Tablet' OS in 2009, creating the N900. Sadly it had to compete with another in-house project to add multi-touch and HW acceleration to Symbian. After Nokia announced the merger of Maemo with MeeGo, they were unable to forge a clear way forward for their smartphones ever again.

Blackberry took far too long to make touch devices their numero uno, from the Storm in 2008 to the Z10 in 2013. In the meantime they had to cope with a lacklustre platform, as just about everyone did until Android came around. They should also have cross-platformed BBM far, far earlier than they did.

The n900 had a resistive touch screen. That by itself is sufficient to disqualify it as a modern smartphone.
A really good resistive screen, probably the best put on any phone. For 2009 it was perfectly usable.
Now is the only time you can charge premium for new chips. Nobody will pay that in a year or two. In five years it's on Craigslist, and in ten, it's ancient crap you find on motherboards pulled from machines thrown into the dumpster.
Hey, my 2003 iMac still works as a great dns resolver.

Of course so does a $5 raspberry pi.

There's also the huge difference in power consumption.
(comment deleted)
You lost me between Q's 2 and 3, and then gained me back in A3. Market segmentation happens when customers think there is a difference; it doesn't matter if the difference is real or not.
Basically off-topic, but this might be one of the few places someone might know: does anyone know which of the quad-core 8th generation Intel CPUs that are about to come out in laptops support TSX-NI? I can't find this information on their site.