Now Zen 2 seems to overall beat Intel on every metric: single-threaded perf, multi-threaded perf, perf/dollar, perf/watt. No matter how you look at it, Zen 2 comes out on top.¹ Very impressive.
Man the folks at Intel must feel the heat.
¹ Except perf/socket when competing with the Xeon 9200, but that's just a PR stunt no one cares about: https://mobile.twitter.com/zorinaq/status/113576693566724096...
When Zen first came is was a huge deal, but it merely put them as a real competitors, with a decent advantage on many cases, reinforced with Zen+. But Zen 2 put them ahead in almost every category, and in all markets; threadripper and EPYC are just as strong in their areas.
Either Intel has something strong about to appear, or they're going to face a truly difficult few years with customers going AMD now that it's not merely "one generation of chip" that was good. It feels like getting their 10nm working will not be enough by itself.
I hope that in a few years time that ECC will become standard, as for many it is driving a car without a seatbelt.
This and with security exploits (rowhammer), ECC would be the solution that is a price gap that for many, is still one that can and should be closed.
That all said, it would only take the one or two big mobile manufacturers to go ECC and marketing virtu the whole security and integrity aspect for the rest of the market to follow suit. Which would be a bigger driver in reducing the price gap premium over non-ecc memory.
Which is how I see things panning out as mobile phone makers are running out of sell features to add and this would be an easy one for the premium market of phones out there today. Also one in which would go down very well for that feature alone.
They do (or did?). My home NAS is running a Sandy Bridge Intel Celeron with ECC memory. Support seems to be randomly distributed thought through the product line though and obviously depends on the motherboard manufacturer to implement it as well.
In general, Intel has a problem with branding. Their product lines are confusing mess requiring looking up each specific part to get a list of features it does or does not support. There's little rhyme or reason to it.
I got bitten by that back in the Core 2 era. When I built my very last Intel system, around an Intel DG43NB board, I picked up a Q8200, thinking it would be a great bang-for-the-buck chip. Little did I realize from the store display that the Q8200 was the only Core 2 Quad that lacked virtualization support.
The DG43NB died prematurely, due to capacitor plague. I didn't shed any tears for it.
Of course you need to know what you're looking at, the suffix of the number matters a lot (eg 7100 vs 7100U), but that is nothing Intel-specific. A Ryzen 2700 and a Ryzen 2700U are very different processors as well.
Its not because they are "very different processors".
If you look at the products that are actually compatible with your system, it's not confusing at all. Pentium/Celeron/i3/Xeon = has ECC, i5/i7 = no ECC.
For me, the reason to go ECC was just to prevent silent file corruption. With 64 GB, the math is in favor of me seeing bit flips. Moreover, I tend to put my machine to sleep, rather than shut it off, which also increases the likelihood of memory errors. I wouldn't say I have a highly specialized workload outside of the occasional VM, consisting of large files, for development. A lot of it is I just didn't want to deal with silent corruption of family photos & videos, even if the underlying file formats are pretty resilient.
Personally, I think the situation should be flipped. Everyone should run with ECC these days and only run without for specialized environments like gaming where you want to squeeze out every FPS you can. Faster memory isn't going to matter for most situations.
But for everything else, it's a scale of needs, sure the low-end would be a gaming console or graphics memory, but if that gap gets smaller, uptake and usage will scale as that advantage becomes cost effective at various consumer levels.
But having the option of choice, is and always been a huge plus for any consumer - AMD makes that choice far more accessible than Intel. Though I do see a mobile phone maker going ECC as the turning point in uptake and making that price gap more palatable in the end.
If you use it just for media consumption and playing games, anything goes.
A good power supply and ECC RAM make a very stable PC.
Whether one cares about that is one's own tradeoff.
A few i3 CPUs support ECC, though.
All desktop segment chips with ECC have i3 designation.
 double check your board manufacturer for bios updates to support new CPU, my cheap B350 board does.
My understanding is that they pushed multi-patterning a bit too far and they have too much defects. 7nm, which use a different tech and is developed in parallel seems to get better.
So if Intel has a big thing in the making, I expect it to be 7nm, not 10nm.
People are hypothesizing this to be true given that Apple is putting Intel rather than AMD chips in the Mac Pro, and Apple doesn’t usually make dumb purchasing decisions, but does sometimes have private access to product roadmaps.
Apple has a lot of optimizations for Intel at the moment from the instruction set down to the motherboards and chipsets. A great example is all the work they do in undervolting mobile chips so they perform better (when the latest MBPs shipped with this disabled, everyone really complained). Re-writing all that software definitely has non-trivial R&D costs.
When making a new motherboard design, a ton of stuff simply gets reused and moved around. Switch to a different chipset and you start all over for a lot of stuff. Even if AMD were 10-20% faster overall, their current "fast enough" Intel chips would still win out.
AMD's zen+ 3000 mobile chips don't compete with Intel in single-clock performance, clockspeeds, or total power draw. With the exception of the mac pro, Apple's entire lineup uses mobile processors. In addition, Intel probably gives amazing discounts to Apple. Zen serves them best as a way to squeeze out an even better deal.
A final consideration is ARM. Given the performance characteristics of A12, Apple most certainly has their sights set on using some variant in their laptops in the not-too-distant future. They already run their phone chips in their macbooks as the T2 chip. They are probably working on the timing to allow those chips to run more than the touchbar and IO.
> With the exception of the mac pro, Apple's entire lineup uses mobile processors.
In fact their entire desktop lineup now uses desktop-grade CPUs. (except possibly some entry-level iMacs that weren't subject to the recent refresh)
* iMac Pro: Workstation processor (Xeon)
* iMac 27": Socketed desktop processor (e.g. Core i9-9900KF in top config)
* iMac 21": Socketed desktop processor (e.g. Core i7-8700)
* Mac Mini: Soldered embedded desktop processor (e.g. Core i7-8700B)
An i9-9900K has a 95w TDP, but Anandtech puts the real load number at around 170w TDP. I've seen people undervolt these down to around 110-120w in the 4.7GHz range. I imagine Apple's dynamic undervolting and custom motherboard can shave 10% or so off that total while their dynamic undervolting can go much lower with fewer cores and lower frequencies. While even that isn't going to make their tiny cooler keep sustained loads from throttling, it could get much closer.
In c't magazine's review of the current iMac, they found that the whole machine appears to have a power limit which is shared by CPU and GPU, so yeah, Apple are definitely doing something fancy in that regard.
Likewise, I do agree that Apple is likely switching to ARM for mobile—but are there any ARM cores on anyone’s product roadmaps that could power a Mac Pro, or even an iMac? Nah.
I do agree with the greater point: Intel probably do have an exclusivity agreement with Apple right now, and so Apple sticking with them right now isn’t evidence of anything in particular.
But to me, it looks like a natural shift for Apple, in the near future, to adopt a hybrid strategy: if they can switch to ARM entirely for their highest-volume segment (laptops et al), then they’ll no longer need the benefits of having Intel as a locked-in high-volume chip supplier, and will thus free to choose anyone they like for the much-lower-volume segment (desktops/workstations) on a per-machine basis. That might be Intel, or AMD, at any given time. They won’t get such great deals from Intel, but in exchange they can play Intel and AMD off one-another, now that they’re in healthy competition again.
Intel is probably just as aware of what Apple has on its roadmap as Apple is aware of what’s on Intel’s roadmap, so I would expect, if anything, that they’re scrounging desperately around for a mobile architecture that’ll be competitive-enough with the nascent A13 to stave off that collapse of a partnership.
It's true that quite a few have had AMD GPUs, and they made the more difficult PowerPC to Intel switch back in 2006 with OS X 10.4. But it would be a significant effort to change a processor partnership more than a decade old. No Apple developers have anything but Intel in their machines; it's not just an item on a BOM.
Given that Apple almost certainly has a research lab maintaining machines running macOS on top of Apple’s own ARM chips, to watch for the inflection point where it becomes tenable to ship laptops running those chips; and thus, given that Apple already has a staff for that lab whose job is to quickly redo macOS’s uarch optimization for each new A[N] core as Apple pumps them out; it doesn’t seem like much of a stretch that those same people would do a uarch optimization pass every once in a while to answer the “can we use AMD chips on desktop yet?” question, does it?
Apple wouldn't care that AMD narrowly beat Intel for 1-2 chip generations either. They'd care about AMD's ability to produce chips at large enough volumes and keep the pace going forward.
They've been burnt by Motorola before and Intel now, to just jump on a short-term bandwagon.
If AMD manages to keep this up (and ramp up their production) for 5+ years, then they might have a chance with Apple. But again, Apple is more likely to go for their own ARM based chips in 5+ years...
I think there's no reason to bother switching to AMD if/when they plan on moving to their own ARM based CPU within the next few years.
Not talking down about prior motherboards, but the next run will have some very high end designs and features compared to prior gen Ryzen as well. I'm really looking forward to upgrading in September/October. Looking at a 3950X unless a more compelling TR option gets announced before then.
K8, too. AMD's IPC was so far ahead of Intel's at that time it was crazy. 2.2GHZ Athlon 64's were keeping up with or beating the 3-3.2ghz P4's.
It was so strong & competitive Intel resorted to straight up bribery to compete, resulting in many anti-trust judgements against it as a result. But they succeeded in preventing K8 from hurting their market share, and kept AMD down despite a vastly superior product. Here's hoping that doesn't happen again this time around, but maybe Intel will decide the wrist slap is worth it.
How good are AMD's laptop chips? The improvement in IPC and efficiency in Zen 2 can go a long way in improving this, and then, of course, they must improve perception.
Anecdotally: I've owned almost exclusively AMD chips in my desktop builds for the past 15 years. I've never once owned an AMD laptop. When Intel built the Core line of chips, they seemed to nail laptop first, and then apply the efficiency to their desktop line with higher clocks. It worked wonders.
In my opinion, AMD really needs to nail laptop CPUs/APUs now more than ever. I hope they do!
Of course, this is the longest I've held out on upgrading, and getting itchy about it...
One has to wonder how things would have gone if they didn't have the Core-M architecture on the side back then.
RDRAM, though, that was rapid. But that was only a fraction of Netburst's problems.
Well, their processors are very effective space heaters, so yeah.
Here's a top of the line Intel Pentium 4 3.46 Extreme Edition being unable to compete with AMD processors running at 66% of the clock speed of the Intel.
I really don't count Core Duo, even though I had a Macbook equipped with one. The Core 2 Duo was Intel's first real competition to Athlon 64.
That was a slight jump.
Yep the 2600 is Zen+, 3600 is Zen 2, decent uplift in IPC (about 10%) on its own not worth upgrading.
That 3900X looks tempting though to replace my 2700X.
50% more faster cores than an already very fast processor, AMD are on fire at the moment.
Announced and priced at $749.
If it performs well and lasts, I don't care if it's nothing but goats blood and binder twine holding it together.
What part of that illustrates they're defective, or that it legitimately costs more than if it was sold separately?
As for frequencies, modern day finfets can clock way way higher than planar devices, and are more limited by thermals than anything else.
The other thing is, these 12Cs are actually clocked significantly higher (and have much more cache enabled) than the lower-end parts. So these are not simply "bad silicon". Really this is a misnomer in general since there is a variety of ways silicon can be "bad". Bad clocks, damaged cache, damaged cores, etc etc.
Binning gives you the opportunity to pick the best cores on a chip too. So if a chip had two cores that could only do 4 GHz but the rest of them could do 4.4 GHz or whatever, then all of them might be functional yet it might fail binning as a fast 8C chip, but you could disable the two derpy cores and ship it as a very fast 6C chip.
The binning process is a deeply trade secret kinda thing, but is undoubtedly much more complex than people generally believe. It's not "top X% silicon becomes Epyc, next X% silicon becomes threadripper", etc. All that is known for sure is that AMD is very efficient at using every part of the buffalo.
“We enabled it but didn’t test it and we don’t guarantee it will work.”
* Not all motherboards support ECC, make sure to check before buying.
* Only unbuffered DIMMs are supported by Ryzen/Threadripper.
* According to motherboard vendors, Ryzen APUs do not support ECC. Ryzen Pro APUs do support ECC but those SKUs are typically not available in retail.
I have an asrock + unbuffered ECC ryzen 5 1500 setup. I've checked windows does indeed detect errors when induced.
I am using an ASROCK - B450M Pro4, Crucial - CT16G4WFD8266 16GB ECC with a Ryzen 5 2600.
Worse could happen if a RAM error corrupted the in-memory copy of filesystem metadata (like an extent map) - subsequent writes that rely on that metadata could then cause gnarly corruption.
We will only use Intel Xeon for our work because of this. You'll get about 1 bit flip/GB/year. With 128 GB or more in our standard builds, this would be more than 2/week. We just can't have that uncertainty in the data we provide.
And while Cinebench is a useful benchmark, all our heavy number crunching is done on NVidia 2080 architecture so the fact that AMD may have an advantage on some cases isn't that interesting for us. Perhaps if you're a gamer, who doesn't care about an occasional bitflip, looking to squeeze the last drop of value out for his dollar....
AMD doesn't disable ECC support entirely on consumer CPUs like Intel does, but as far as I know it's also not officially supported and guaranteed to work, it's up to the mainboard vendor how to handle this. In the Intel case you simply can't get ECC with non-Xeon CPUs.
Not quite. There are some Core branded CPUs that support ECC, including funnily enough the i3's.
Hell, there are Celeron and Pentium chips that they have it enabled on. Not because they expect desktop users to buy them, but because it allows them to keep their Xeon brand premium while letting OEM's like Dell advertise the T140 "starting at $549" (in a configuration nobody would ever want to buy).
For example in the 7000 series, the i3 7100 has a 3.9 GHz base clock and you have to go almost to the top Xeon (the equivalent of an i7) to get anything equivalent. And even then it's a turbo, not a base clock, so in principle the motherboard should not let you turbo forever (PL2 time limit may actually be enforced on a server chipset).
Also depending on workload you may not even be able to exploit an increased threading capability anyway, without 10 GbE on the box, or link aggregation capability.
The Celeron and Pentium chips that have infiltrated entry-level servers are absolute trash though.
It seems like it's mostly any chip that would compete with the Xeon-W gets ECC removed.
That said, the new Xeon-W series has more memory channels (6) and supports more RAM (up to 2 TB) than any existing Threadripper product. I.e., AMD doesn't have an equivalent product for all use cases yet.
However, we don't know the Zen 2 Threadripper lineup and the frequencies for the different Zen 2 Epyc SKUs are also not public yet. AMD could release Threadripper with support for RDIMM/LRDIMM or Epyc chips with higher clock speeds to better compete against Xeon-W.
Both cases are hardware errors of some form because usually swapping ram/motherboard/powersupply/etc will clear it up.
See google's study: https://static.googleusercontent.com/media/research.google.c...
I've seen googles study, and out of the few thousand or so machines I've had statistics collections from, the few machines with soft errors were fixable and stopped reporting soft errors after having something swapped.
The google study itself goes on and on about the variability of errors with such wonderful sections as "These numbers vary greatly by platform. Around 20% of DIMMs in Platform A and B are affected by correctable errors per year, compared to less than 4% of DIMMs in Platform C and D."
The paper really leaves a lot of holes, I don't remember (nor do I see after skimming it) any note of how aggressively they are running the ram. Did they say try to reduce the ram timings/bump voltage on the platforms they were having issues with? Did they compare how mature the technology was when the commissioned it? Did they try to diagnose the machines reporting high error rates by seeing if they could convert a machine with a high error rate to something lower? They do spend a lot of time talking about temp though. The only valid conclusion I think can be drawn from the paper is "ECC is important use it because you will have RAM failures, better to know about it than not".
To me the paper speaks to googles diagnostic/repair system more than anything. I took a proactive approach and replaced DIMMs/Motherboards/Powersupplies/etc that reported correctable errors. When we were self supporting we would swap the questionable parts into other machines to see if the failures would follow them in an attempt to see if we could prove a failing part was marginal. Then return/exchange it if it failed in more than one machine.
I've seen a lot of different failures over time, and when I was partially in charge of designing/picking platforms I even managed to find actual design bugs a couple times that caused low rate error rates (not in the RAM subsystem thankfully). I tended to use the "any kind of failure when run normally is instant disqualification" metric when I was initially picking new platforms before buying them to put in production. I would never have qualified a platform that had a 20% DIMM failure rate. (well at least not purposefully, we got some stinkers but we tried to correct our mistakes).
Given what i've heard of google, i'm not sure I would really extend these reliability metrics unless your buying the latest bleeding edge parts and running them well into their design margins. These days its pretty common to design systems that have error correction and push the physical topology to the point where there is an expectation of a pretty solid error rate (think SSD flash chips). So for a company like google pushing the RAM timings/etc right out to the margin where they are experiencing a low but statistically unlikely error rate would seem to be the right thing to do. Its different if your a bank/etc running financial data. In that case you buy for reliability first.
That's incorrect. Uncorrectable errors are properly reported to the OS. Wendell from Level One Techs has tested this: https://www.reddit.com/r/Amd/comments/b1qmgy/ars_technica_th...
If you're worried about a single event causing two flips in a single row... I suppose that's possible, but it could also cause three bit flips. So a Xeon has a non-zero error rate. Is Ryzen meaningfully worse?
I had the chance a long time ago to work on a product that as a side effect was corrupting system memory... Think of it as a kernel module that picks a random number between 0 and MAX_RAM and flips a byte. Its truly amazing how many of those can happen before there is any visible evidence something is wrong.
Though note that the chip in this article is not going in a NUC form factor because it does not have integrated graphics. The Ryzen chips with integrated graphics top out at 4 cores and the new ones are still based on the older Zen+ microarchitecture.
First is that you need a gpu, so the chips just announced aren't going to work.
The second is heat dissipation. It's easy to think any heat sink and fan will work, but the reality is that an integrated gpu and dynamic clocking can generate a lot of heat. The computer will still function, but can end up laggy and much less capable because it is being throttled.
Then there are bios options to be aware of. Dynamic overclocking generates a lot of heat for little gain. The defaults also throttle too soon. Relaxed temperature throttling runs hotter but gives a much better experience. CPUs run just fine at 50C anyway.
This is all to say a small PC is possible now and a NUC might be possible later, but they are still tricky to really get right.
But you can also find Mini-STX boards like this one: https://www.sapphiretech.com/en/commercial/amd-fs-fp5v which should then fit in something like this: https://www.silverstonetek.com/product.php?pid=708&area=en
You can also find some very small Mini-ITX cases on things like aliexpress: https://www.aliexpress.com/item/32994653418.html?spm=2114.se...
The date is rumored to be the 7th of July.
Similar to how product details are no longer "announced" or "released", but rather "leaked".
I'm searching for the fastest C++ compile machine, that includes linking (which is single core).
For instance, if you're working on a large C++ codebase with incremental compilation, rarely do full recompiles and heavily rely on link time and profile guided optimizations, I can see a case where a heavily tuned Cascade Lake X makes a lot of sense (overclock the mesh, tune the memory latencies and get enough cooling to sustain the single core boost as high as you can). It's actually one of the few usecases (AVX-512 being another) where I still see a niche for Intel CPUs purely based on technical merit... saying that still feels weird, but that's where we are now.
It'd be stupid for AMD to not release another generation of Threadripper with up to 64 cores, but we don't know anything for sure just yet except what has been detailed about the consumer Ryzen and the server EPYC.
But I'd say it's far from being too slow. The boot times and load times for most software is fine for my use cases.
It was just too expensive of a build to justify an upgrade so soon.
Is Ryzen a better choice for a daily driver?
For more general benchmarking something like Geekbench is probably more useful, as it tests performance for memory intensive, cryptographic, integer and other workloads.
Here are the benchmarks: https://browser.geekbench.com/processor-benchmarks The top appears to be dominated by Intel, but AMD is often better scoring at the same price point.
For web browsing and playing video (i.e. 99% of all users' "daily driving"), I don't think it even matters to consider these sorts of benchmarks.
I do edit video and work with 3D graphics almost daily though, so this is sort of comparison is very relevant.
In other words, of you were to actually measure page load speed, are you saying the measirements would correlate with the benchmarks?
In reality I think for general web browsing the difference, if it exists at all, in either direction, would not be perceptible.
I know that's not EXACTLY what you're asking, but I think it's worth mentioning the practical aspect as well.
Source: I work at Intel currently.
His advances literally dictate the efficiency of 2% of the worlds power output, and that's just one metric of influence.
And that's very much in the middle of the zen 2 offerings... and at a significantly lower price.
I thought this benchmark was about single thread perf? From the article -
"However, according to a Cinebench score shared by Videocardz on Twitter, the 3600 levels with the Core i7-9700K in the single threaded test."
Whilst this is only one benchmark, it's looking like they've caught up or even overtaken on single thread performance too.
Anecdata here, but...
Depending on your type of workload, you don't need anything more than your typical heat sink and fan. Using Prime95 as a torture test, my 9900K will clock to 5 Ghz just fine on just air cooling without thermal throttling unless I'm doing the small FFT test, which creates a workload that fits entirely within CPU cache. I'm not sure if I'd consider that much of a real-world benchmark though.
Should be interesting and I suspect on even games AMD will be 10%+ ahead.
Come September they're going to release the 3950x,and if folks can wait a bit, you get 16 cores/32 threads on one socket.
But even on a budget, the 3600G looks like a great buy with Navi 20+3600 on chip.
The deal is that they have reasonable spectre mitigations working since last few gens, and Intel has to fix this if they need to be competitive. Not to mention standardizing a chipset.
Intel runs things the other way lately, mobile gets released first, then desktop, then server.
9600k has a 3.7ghz stock base clock, 9700k and 9900k have identical base clocks of 3.6; in order, their stock single core turbos are 4.6, 4.9, and 5.0.
Now, you could be asking yourself "but Diablo, in single threaded, the 9900k is 8.6% faster, and the 9700k is 6.5% faster than the 9600k, respectively". They're all k parts, just overclock it the tiny bit of the way for a fraction of the price ($230 vs $410 vs $490), and the 9600k has far more thermal and power budget per-core to play with than the other two do.
I almost built a new desktop with a 9600k, the $230 6/6 part that shames (in both single and multithreaded) the highest end $340 or $350 4/8 parts from the previous four generations (4770k, 4790k, 6700k, 7700k); vs the $360 8700k (6/12), it performs identically in single threaded, and illustrates that hyperthreading only gains you about 25% extra performance if you can saturate all 12 threads.
Remember, these are desktop chips, you are very unlikely to ever need more than 8 threads (and use them effectively), even if you game. If you happen to have a use case where you can easily saturate >8 threads, anything LGA115x is probably inappropriate for you anyways.
However, with the 9000 series release, Intel admitted AMD scared the fuck out of them, and re-released Coffee Lake at a lower price with some tweaks in speed and core/thread count because they were afraid of Zen 2 being a success; their fears were warranted.
The $250 3600x (a 6/12 part) beats the 9600k in single threaded, widely beats it in multithreaded, and has PCI-E 4.0, with an extra 4 CPU-bound PCI-E lanes (20 vs 16 on AM4 vs LGA115x), and uses a higher DDR4 clock (stock 3200 vs 2666, with an effective maximum of possibly over 5000 vs around 4133; AMD's IMC seems to continue to be effective at lower CAS latency than Intel's is).
Side note: 4790k, a Devil's Canyon part, is Intel's stand-in for the non-existent 5700k.
Devil's Canyon is a Haswell Refresh part that got a second set of 4000 series model numbers instead of 5000 series. Haswell Refresh was a Broadwell core paired with a DDR3 controller, fabbed at Haswell's node size; there are no architectural changes between Broadwell and Haswell in the core, only major changes was the decrease in node size and the swap to DDR4, the core design remained nearly identical. Broadwell largely ignored the desktop, focusing on LGA2011 and mobile parts instead, leaving Haswell Refresh to fill in the gap with the desktop and Xeon E3s, and the only notable exceptions being a small set of Iris Pro GPU parts.
There were certainly some changes. The gather instructions were dramatically improved, taking ~5 uops instead of ~30 and with much better throughput.
Conditional moves only take 1 uop in Broadwell, down from 2.
Some other changes listed here:
Intel's tick-tock model was never black and white: even the "ticks" (node shrinks) received some changes and even new instructions.
Is there any public access to a shell on one of these things to run your own benchmark?