The calm before the storm: AMD's Zen bears down on Intel CPUs AMD is continuing to drip feed information about Zen, its long-awaited designed-from-scratch x86 processor microarchitecture. Zen is a big deal, or rather needs to be a big deal, for AMD because its previous Bulldozer microarchitecture was, well, bulldozed by Intel's Core series. Thus, AMD needs Zen to succeed in order to win …
The Intel HSFs aren't *that* bad. The Pentium G2120 in my VMWare server has been going more-or-less non-stop for over 3 years on the stock Intel fan, it's already killed the fan in one PSU.
I'm still using the stock HSF on my i7-6700 too. I've got a Corsair H-90 ready to go in, I only put the Intel one on to check the rest of the kit was working. One day I'll get around to taking it to bits and putting the Corsair in, but the Intel is coping well so far.
"The Intel HSFs aren't *that* bad."
And surprisingly, Intel stock coolers are pretty efficient.
Foil/air bearing fans aren't terribly expensive compared to any given CPU. These tend not to fail ever (unless dropped). I'm guessing the mentality is that a system is only expected to last 3-5 years at most so why spec the fan beyond that? (Because some proportion of them die early!)
Of course most home systems end up with clogged up heatsinks long before the fans die. There has to be a better way of getting heat out of the box which doesn't cost the earth.
>Used to like AMD as they saved on heating costs in the winter.
My 3930k fills that function quite nicely.
I'd love AMD to do well. I just hope people are buying new PCs in sufficient quantities to give them good sales. However much I might like them, I doubt I'll be changing to anything anytime soon. Let's hope the server market is kind to them.
The Lenovo laptop on which this is typed runs an AMD processor, and, with openSUSE and tlp, I have had it down to under 5w, screen on, admittedly at idle. The days of AMD=excess_power are long gone. I wish the chipset was available as a mini-ITX or similar, as it would make a great home/small office server with that power consumption.
The days of excess power are 'sort of' gone if you're using an AMD APU but that's because they're low end/low(ish) performance parts more than any real efficiency improvement. Intels equivalents still suck less power. Just harder to care about a few watts.
The performance parts are a very different story. My current FX8370 under load supplies 50%+ of the winter heating in my office. Fire up the GPU and I have a heating problem ;)
AMD needs to survive, and, if possible, grow. Intel hasn't any other credible competition.
Recently I built my new desktop workstation, all AMD. For my uses it thrashes the relevant i7. But then I don't game, I need cores and lots of them.
I am looking forward to hearing more about the Zen.
(AMD fanboi.)
Intel need competition, or they'll try to milk their customers dry. Whether that competition comes in the x86 space or from the up and coming ARM platform (chromebooks and desktop-focused android e.g. remix OS, as well as ARM-based servers) is another matter. AMD are pretty much unique, in that they are able to field products in both the x86 and ARM space.
I wish them all the best... but I can't wait for them to catch up. My next machine is going to be based on intel - maybe next time I can go back to AMD.
The Athlon 64 days were even better. Cheaper and faster than Intel's offerings, and 64 bit.
Intel came out with Prescott, which turned out to be rubbish (unless you were chilly), and had to go back to the drawing board and come out with the Core architecture.
So, even if you prefer buying Intel, a strong AMD is good because they force Intel to up their game.
(Unfortunately I just bought a new i5, ho hum)
And by Core Architecture you mean based on Pentium M which means based on Pentium 3 which means based on Pentium Pro.
Netburst really was a fuck up of legendary proportions when Intel thought as long as they could market with the highest clock speeds of CPU and RAM then the performance didn't matter.
"Intel thought as long as they could market with the highest clock speeds of CPU and RAM then the performance didn't matter."
They're right, up to a point. People buy based on clock speed, not on actual benchmarks.
When I point out how small the differences are between E5 2.4GHz and 3.5GHz cores for most _real world_ operations people look at me as if I've sprouted a second head.
"Netburst really was a fuck up of legendary proportions when Intel thought as long as they could market with the highest clock speeds of CPU and RAM then the performance didn't matter."
I can remember Intel's boast that it would be selling 10ghz chips by 2011:
http://www.geek.com/chips/intel-predicts-10ghz-chips-by-2011-564808/
They didn't explicitly say that the Pentium 4 would reach those speeds, but it's an insight into their thinking - clock speed was the only thing that mattered. Since the late 1990s my main desktop PC has almost invariably been an Intel machine, not because I love Intel but because their chips have been mostly good. But I avoided the Pentium 4 era entirely. I skipped from an overclocked Celeron 300A to an AMD Duron, and then from the Duron to the Core Duo era.
Based on this report Zen doesn't seem very impressive, which is a shame because I have a certain amount of residual affection for AMD. For a few years they took on Intel and won! On the other hand AMD's share price has gone up by 10% today so obviously the market thinks that Zen is a good thing.
I have stuck with AMD since my early days of self-build PC, and I liked that AMD changed their socket design less often than Intel. My current 4-year old processor is running on a 6/7-year old motherboard, and handles itself rather well - only sticking point is the memory bus speed.
The other great thing is that they are cheap as ... well, you know.
AMD has to rely on external fabs for their semi-con production expertise, while Intel does everything in-house. Because Global Foundries just doesn't have the same sort of cash to burn as Intel they seem to be a step behind in terms of production expertise and capabilities. This hampers AMD as well. It's all well and good they can deliver a chip that can compete with Intels CURRENT offering, but what about the next generation?
GlobalFoundries has the Sheikh of Abu Dhabi as a Sugar-Daddy, They just got paid $$Billions to haul away the trash that was IBM Microelectronics and they just got given Samsung's 14nm process for free !!!! Also ...they don't have to publish accounts.
How can they NOT have "the same sort of cash to burn as Intel" ???
@P0l0nium
Intel spends those billions on R&D alone. There really is a massive scale difference there. Getting a fab process "for free" still doesn't give them the sort of deep know how about nano-lithography Intel gets from doing all the research itself. Once things are worked out the slightly less clever can get/keep it working. But when things inevitably go south because of some unforseen variable you want the really brainy chaps around who have the "fingerspitzengefühl" to get it fixed. From what I'm seeing Global Foundries is still playing catchup (Though they are making good progression and are certainly gaining on Intel)
Things aren't exactly rosy around Intel's fabs right now. A lot of their better people took the early retirement and left including most of their more experienced engineers. I would stay away from their newer products until Intel can build up the same level of institutional experience that's walking out the door and over to Samsung, Qualcomm, ARM and GlobalFoundries.
Competitive as in similar to but ultimately worse than Intel.
My current i7 is one of the 65W ones, it stays cool most of the time without needing the fan spinning. Current AMD chips that get anywhere near it on speed are over 100W.
The old competitive AMD forced Intel to develop decent processors and release them quickly. Now my processor is 3 generations old and I don't see anything from benchmarks that would encourage me to upgrade. Hopefully something new from AMD will shake up the CPU market.
"Current AMD chips" have a 40% lower IPC than these new ones, or so they say. So don't base your assumptions about TDP on those old ones.
And there's no way your i7 runs without the fan spinning. You mean it is spinning quietly, not that it isn't spinning at all. Getting any desktop CPU to run without a fan at all requires either a solid copper heatsink the size of a Rubik's Cube, or using one of those expensive ULV chips with a 15 watt TDP (and accepting reduced performance)
Intel's reduced progress on performance has more to do with reaching physical limits than lack of competition. Increasing frequency increases power draw exponentially, so we've kind of settled at a ceiling of around 4 GHz, with little progress being made. You can have more cores, you can have more new instructions to speed up certain types of encryption, but the only improvements beyond single digits between generations of Intel chips you will see in the future is on the GPU - because graphics is easy to parallelize.
"but the only improvements beyond single digits between generations of Intel chips you will see in the future is on the GPU - because graphics is easy to parallelize."
Yup and for the kind of stuff I'm supporting, that's a major problem.
The mantra for the last 50 years in research computing has been "computers always get faster" and they've relied on it when predicting delivery of results.
We've been getting a steady stream of complaints that "the new server is no faster (or slightly slower) than the old one" - and invariably the culprit is badly written, singlethreaded code that simply doesn't know how to run in a multicore system.
Physicists refuse to take computing or programming courses (they think that IDL(*) is a good language to do heavy lifting in FFS), let alone accept that they have to know how to multithread. The current kludge is to run lots of individual processes from the command line but this comes with its own gotchas.
(*) IDL is to scientific computing as MS Excel is to business operations.
"We've been getting a steady stream of complaints that "the new server is no faster (or slightly slower) than the old one" - and invariably the culprit is badly written, singlethreaded code that simply doesn't know how to run in a multicore system."
We have a similar problem, but the root cause is PHBs thinking that more cores on the same memory + cache config = more speed. They are finding out that more cores is f.all use when memory is the bottleneck. With respect to threads, they tend to make the cache/memory contention problem *worse*, the ideal is a bunch of loosely couple processes that share as little memory as possible. :)
Large performance gains will be easy for Intel to show when they bite the bullet and
introduce high bandwidth 3d memory cache on Xeons, just as Knights Landing, Nvidia Pascal
and AMD GPU's already have.
Interestingly, AMD may be even ahead of Intel here, since they already have experience
of HBM with their GPU's.
Its the memory that's the problem.
I'm not a gamer, nor in the market for Zen... yet. When I need a couple more cores to keep most of the the 32GB ECC (unbuffered) RAM, 10 SATA3 + 6 USB3 ports, and 4 PCIe3 slots busy (SABERTOOTH 990FX/GEN3); I'll swap the FX6300 for a liquid cooled and somewhat overclocked (~4.5GHz?) FX8370e. Primarily for storage with an occasional compile, and maybe a VM or two, it does yeoman's duty - thank you very much AMD! It's EVGA Superclocked 750Ti is more than fit for purpose for the user side - while a Geforce4 64MB chugs away in the PCI(no e) slot as the Console. (Presuming it lasts a few years longer a Zen+ is in that box's future.)
There's no rush, but a Raven Ridge APU will be a fine foundation for graphical workstation late next year. Ye Olde 1.6GHz 4GB dual 64bit Athlon (RS780) AIW is currently sufficient for web surfing so long as NoScript blocks most of the useless CPU Cycle Thieves.
The one lung "Walmart Special" (Compaq) 2GHz 32bit Athlon upgraded to 1GB RAM with the Radeon Xpress 200M (1150 chipset w/ATI Radeon X300 GPU) is getting a little tired. A "lightweight" OS like AntiX works fine, but the graphics are weak. In a laptop I'd like an APU with HBM2 to assist the graphics and plenty of ports. Even just 8GB of HBM2 with no DDR4 would be sufficient, though mobile APUs with HBM2 probably won't arrive till '18 or later. An additional battery draining GPU need not apply at it's endurance for hobbyist Astronomy and Amateur Radio applications rather than gaming prowess that I'm wanting. ;-)
"As for Zen...let's just wait for the reviews shall we? No point getting all excited. I think we've been here before."
This. Yes, we have been here before expecting a new processor/gpu/architecture to eclipse the competitor's offering. We see smoke and mirrors tests (like this article) and get our hopes up for true competition to the reigning champion.... only to be disappointed when the final product comes out and third parties start testing.
I still have my Athlon chip (on the shelf), and I'm proud of that little one. It's good to see that AMD is getting back to where it should be : toe-to-toe with Intel. Good for everyone, as we all agree.
But I am not ashamed of my current i7. I just want the best performance my money can buy for my gaming needs. And it has been a (long) while since AMD was in that picture. I do sincerely hope that my next CPU choice will be AMD.
Here's to looking forward to that.
But I am not ashamed of my current i7. I just want the best performance my money can buy for my gaming needs. And it has been a (long) while since AMD was in that picture. I do sincerely hope that my next CPU choice will be AMD.
My current machines are i5 and plenty of life left in them for my needs. No shame here either, but also looking forward to building an AMD machine again. It has been a long time...
Long time AMD CPU fan here, I've built a few AMD gaming rigs over the years (late 90s onward)..
But my current rig, built in 2012, is an i7 (3770k), as unfortunately for gaming, single core speed, not number of cores, rules, and AMD just were not there!
I really hope this 3GHz speed is just due to it being a test, and that this isn't indicative of the real world speeds we can expect from this new chip set in the final production version.
Hopefully if they can match proper desktop speeds, as in 4.5GHz (and upwards), then this could mean AMD become a contender against the top end i7 chips, and become a viable option in time for my next build.
I have to admit, I was a bit disappointed to read that it is on par with current Intel architecture. I expected a ground up chip design to blow away the competition, not something to get the to the current level of performance.
I was an AMD fanboi back in the day, but when I looked for my new laptop recently, I didn't consider AMD as a viable alternative. As long as PC cycles are these days, I'll probably consider it in several years. By then it might be too late t hough.
I recently upgraded to an FX-8370 (currently mildly overclocked at 4.5GHz) to replace the FX-8120 that I built this system with* in the hope of extending the life of this system until I feel I can afford to build something new.
I was telling a colleague this and explaining that I did so because it's the most powerful chip that will fit in the socket and that next time I'll be looking to (possibly spend more that I would for AMD and) build an Intel system as AMD just aren't performing as well.
As somebody who admires AMD for the architecture they invented (AMD64) and some Intel-beating chips of the past I hope that this means I may be able to choose AMD and be buying something close to top-of-the-table (assuming I can afford it, of course).
*I knew at the time it wasn't impressive with benchmarks but knowing that all virtualisation features were definitely enabled rather than Intel's "You must read all the specifications, regardless of price, as sometimes we just disable things for fun" approach and knowing that I'd at least 4 proper cores and potentially 8 (usage-dependent) for my money meant it was generally fine and I still think it was the right choice.
I'm still using a very old x2 AMD as my main work PC, it's been upgraded to 4GB ram, SSD, and an AMD 250 graphics card to run 2x 30 inch monitors. Works fine for me as I actually work and don't FB, LinkedIn etc all day.
Server side its several 8350 cpu's with 32 GB each ram runing VM's on a cost effective SAN. At half the price of intel kit they get upgraded more often and I have more redundancy.
Unfortunately when we need new desktops in a hurry it always ends up being intel based. AMD are either out of stock or the lead time is too long. Sort that out and they woud probably sell more machines :/
I think the issue here is that OEMs just don't like pushing AMD kit (why would they when 95%+ of their lines are Intel) and most IT corporate buyers just know they will be fine if they buy 5000 i3 machines like everyone else.
They have them listed because I reckon Intel asks them, to give the impression there is some choice and competition.
But I still went Intel / Nvidia.
Last time I had an AMD processor (Phenom II X4 965 140W) my system would crash randomly. After RMA'ing several parts (Video card, memory) the issue didn't go away. After about a year, it turns out that this particular chip (but not the lower power Phenom II X4 965) randomly crashes when a full set of 4 memory sticks are installed in the system.
According to their warranty, design erratas are not covered.
Thanks for a year's worth of troubleshooting pain AMD, I wish you luck but fuck off.
After about a year, it turns out that this particular chip (but not the lower power Phenom II X4 965) randomly crashes when a full set of 4 memory sticks are installed in the system.
Similar happened to me, but the cpu/chipset were from Intel. Possibly it was a firmware problem but fixed it by using only two bigger sticks.
Last time I had an AMD processor (Phenom II X4 965 140W) my system would crash randomly. After RMA'ing several parts (Video card, memory) the issue didn't go away. After about a year, it turns out that this particular chip (but not the lower power Phenom II X4 965) randomly crashes when a full set of 4 memory sticks are installed in the system.
I had some dual core variants. The weak memory controller (AthlonII/PhenomII) was (IIRC) quite well documented (even if not by AMD themselves) and they had that TLB bug until stepping B2/BA. Perhaps you were one of the unfortunate early adopters.
Yes, I agree errata should be covered by warranty if acceptable work around doesn't exist (say, the TLB bug, where you could disable TLB but would take 10% performance hit which for many would not be acceptable).
Admittedly in was in a lucky position where using just two DIMMs was sufficient. In fact still running one of those as a server (with ECC memory on Gigabyte 990XA mobo) and it hasn't missed a beat.
Come October a dual proc 32 core Zen dedicated server hosting will cost 4 times the price it was in September. £25 PCM for 2 processors now, £100 PCM for 2*32 cores in October.
That's based on MS statement that 1 proc license cost now will be equal to 8 cores price come October. You may hazard a guess at how much when they put up fees in January next year.
This will go for on premise too.
aka it's going to be slow. This is a workstation processor without workstation features (except ECC). To impress me it'd need improved virtualisation features (Intel is undeniably the clear winner here), ideally dual socket support by default (which would of course eat into AMD's lacklustre server offerings), and an integrated APU (OK, maybe that should go in the server processor, but it destroys the lie that AMD is committed to HSA. Basic VDI built in would be useful)
Another site claims AMD underclocked the 6900K to 3GHz to match the clock speed of the Zen sample, which still didn't convincingly win. In AMD's current generation specific heavily threaded benchmarks offer the highest performance, so this test is probably a high point.
Zen will be repeatedly and embarrassingly trounced on any single threaded benchmark on release. The performance desktop market is not Broadwell-E - that's a set of niche enthusiast CPUs for virtualisation and rendering fans that can't stomach buying a Xeon. The target market for Zen should be the 6700K/4790K i.e. four high speed cores, eight threads. Very fast for most desktop apps, and the best choice for gamers.
In terms of the Intel vs AMD argument - I don't really have a preference either way - as a pragmatic Scotsman I'll buy go with whatever works best for the task at hand... In truth that has meant Intel 99 times out of 100 for the last 10 years - but as soon as AMD have a competing solution worth deploying, I'll deploy it.
I have used AMD chips for a couple of decades now. I've had no complaint with Intel's performance or reliability either, but am I the only one that thinks that despite whatever benchmark, that in real-world use, clock speeds, etc. being similar, AMD processors just 'feel' faster in random day-to-day use? I'm not talking about hardcore gaming or a single drag race, but the type of stuff you normally do with a PC, such as opening a couple of dozen things, switching between them, copying huge files while playing video while working on something else, etc.
Perhaps I just like to root for the underdog, but also Intel has used a lot of shady business practices over the years to 'lean on' their partners and ensure that their chips are chosen by laptop manufacturers. Yeah, no one ever got fired for buying Intel (unless they work at AMD, perhaps), but when it comes to the machines I use outside of work, I will keep buying AMD.
If nothing else, the Zen series might force Intel's top-end CPU prices down from their current crazy levels. According to Amazon, the 5960X 8-core *desktop* CPU has an unbelievable 1,336.73 pounds RRP (though in reality, Amazon are selling it for 849 quid). If a roughly equivalent 8-core Zen comes in at 500 quid or less, I can see it selling like hot cakes (but hopefully that doesn't mean the CPU burns cakes :-) ).
...customers asked me to build about 4-5 years+ ago mainly due to the motherboards being far better specced than the budget Intel boards.
Back then a cheap AMD mATX board would have at least - DVI or HDMI, USB3.0, eSATA and decent surround options.
On the Intel side a cheap Intel mATX board would be rocking USB2.0, VGA, and a serial and parallel port like it was 2003.
Thankfully things have improved on the Intel front.
"We estimate that performance boost slots Zen somewhere between Intel's Broadwell and the Skylake microarchitecture in terms of single-threaded IPC."
Whoopy doo!
They are going to bring out a chip that for most desktop users is slightly less good than Intels current offering, just give up now it'll be less painful to watch.
Considering that there appears to be a lot of people who, like me, are still happily running an old i5/i7 (2500/2600 vintage) Intel CPU because they cannot justify buying a new PC for marginal gains of the new Skylake kit, do they really think these people are going to be willing to switch to an AMD system that is 'less good' than the one they cannot be bothered with upgrading to right now?
AMD/ATI, whatever you want to call yourselves, you need to produce something that is actually better than Intel/Nvidia not nearly as good but £50 cheaper, because if that extra £50 means people will stay with good enough kit for longer they will probably just buy th better kit.
Im an intel owner and am buying a new 6700k next week, the last AMD chip I had was the Barton core yet I still don't this hate on them "just give up now it'll be less painful to watch."
AMD did drop the ball massively they are well behind Intel in term's of performance on their chip so to catch up to intels current tech (if thats true I still want to see full benchmarks etc) is certainly very commendable. If they price it right they could actually bring back some competition to Intel.
The Zen+ core is supposed to be even better though again id want to see full benchmarks, but no matter who I choose for my CPU I want there to be options. Look at intel their latest offering have been a bit lack luster, 6th gen is only slighly faster than 5th gen and its got a slower IGP. If AMD can get this right intel will up their game. Cometition in the market place means we will get better hardware for cheaper prices something that benifits all users from both the AMD/Intel camp.
Who knows maybe Zen+ could beat intels 7th/8th gen? If so even better because you bet the 9th/10th will be amazing. If AMD just gave up now intel wouldnt even need to try anymore.
I would love AMD to pull a great chip out from somewhere that would really give Intel a run for it's money but they just seem to have given up, at least for games/desktop. This new chip with the multiple cores might make a great server chip or some other job.
Perhaps if game makers actually started using multiple cores for their games then we might have some real competition again, and with chips appearing to be reaching their limits of speed they might start doing it.
Personally I think AMD lost their way when they merged with ATI.
"We were shown a PC powered by a 3GHz eight-core Summit Ridge running alongside a system sporting a comparable 3GHz eight-core Intel Broadwell-E 6900-series Core i7. At the push of a large red button, both machines raced to render on screen a very large 3D model of a Zen processor package in Blender. Whichever CPU finished drawing the scene first would be the winner."
Let's hope AMD can make a comeback and put some price pressure on the CPU market.
However the demo would be almost useless, especially with no comparison of power consumption.
And has been pleased with the cost/performance ratio. I own an FX-6300 at the moment which I want to replace with a Summit Ridge processor if the price and performance is right. Intel i7-4790K is an alternative which I might also consider if the price drop after the Zen release.
Anecdotally, I bought a Phenom II hexacore just as they were going out of production and paired it with a more recent AM3+ board for SATA III and USB3. Although it required some overclocking, the system managed to reach parity with a bog-standard i7 920 (first gen) running SuperPi.
Switching over to multi-threaded tasks (video encoding) the Phenom II, which cost me $120 (tax included, there was a sale) retail at time of purchase, was comparable in performance to Intel CPUs in the $300+ range. At the same price point, the best I would have gotten from Intel would have been some pitiful Celeron or i3 system.
So yes, it requires a lot of power (150W) and overclocking is more or less mandatory to make up for some of the deficit in memory bandwidth versus a Core system, but it doesn't fall short as a workhorse for anything from gaming to heavy VM usage and video encoding, not bad overall for a 6-year-old $120 CPU (which, IIRC, retailed at ~$150 from the start).
AMD sold to users and cost less but that market has died, game consoles took people away, endless up grading ended and users embraced IOS and Android.
AMD has let Intel get into the graphics market for casual users too.
So now they must make a better than Intel chip for substantial less cost to the buyers!
The more elite the market the smaller the customer base, so AMD need to think hard.
They need a AMD CPU and GPU combo that substantially out perform Intel at a lower cost. Which is a stuck between a rock and a hard place situation.
IMHO.
"Summit Ridge is able to keep up with high-end Intel desktop offerings from earlier this year"
So AMD's new chip will be about as good as Intel's year-old ones, on an architecture that will be three years old by that point? By which time Kaby Lake will already be out (unlikely to be a big step, but surely with at least some improvement), and then Cannonlake will come soon after to completely blow it out the water. It's all very well saying that Intel need some competition, but this really isn't what competition looks like.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
