Slashdot Mirror
AMD Ryzen Threadripper Launched: Performance Benchmarks Vs Intel Skylake-X (hothardware.com)
Reader MojoKid writes: AMD continues its attack on the desktop CPU market versus Intel today, with the official launch of the company's Ryzen Threadripper processors. Threadripper is AMD's high-end, many-core desktop processor, that leverages the same Zen microarchitecture that debuted with Ryzen 7. The top-end Ryzen Threadripper 1950X is a multi-chip module featuring 16 processor cores (two discrete die), with support for 32 threads. The base frequency for the 1950X is 3.4GHz, with all-core boost clocks of up to 3.7GHz. Four of the cores will regularly boost up to 4GHz, however, and power and temperature permitting, those four cores will reach 4.2GHz when XFR kicks in. The 12-core Threadripper 1920X has very similar clocks and its boost and XFR frequencies are exactly the same. The Threadripper 1920X's base-clock, however, is 100MHz higher than its big brother, at 3.5GHz. In a litany of benchmarks with multi-threaded workloads, Threadripper 1950X and 1920X high core-counts, in addition to strong SMT scaling, result in the best multi-threaded scores seen from any single CPU to date. Threadripper also offers massive amounts of memory bandwidth and more IO than other Intel processors. Though absolute power consumption is somewhat high, Threadrippers are significantly more efficient than AMD's previous-generation processors. In lightly-threaded workloads, Threadripper trails Intel's latest Skylake-X CPUs, however, which translates to lower performance in applications and games that can't leverage all of Threadripper's additional compute resources. Threadripper 1950X and 1920X processors are available starting today at $999 and $799, respectively. On a per-core basis, they're less expensive than Intel Skylake-X and very competitively priced.
Let's just say this thing is going to put out some serious heat at 180 watts TDP. You will need a big and loud fan. And any money you save on the cost of the CPU, you will pay to the electric company. And you will have to hope that you do not use the CPU for long, because the longer you use it, the more this space heater will cost you over an Intel CPU.
This is interesting now but in 4-8 months is when the market will begin to really adjust and the competitors will be more squared off. Expect a very interesting holiday season. Now if we can just get that memory price down.
It is pretty safe under heavy workload. Ryzen only crashed when creating many many many processes were created each second and the only workload which does that is a c/c++ compiler.
And that bug have been fixed.
Actually there is still more work and research being done. And the problem can be triggered under many scenarios involving various loads. There are "attempts" and some "workarounds" that mitigate the problem, but it's still there.
I have a Ryzen 1700x (so basically half of a theadripper) and overall it's good and stable but there are some issues still around -Ram it seems the memory controller is vary sensitive -VMWare ESXi (i think a lot of it is fixed but there is always room for improvement.) things i would like AMD to really work on for the next socket for desktop systems: -4 channel ram - more PCI-E Lanes off the chipset - better public documentation (it was unknown until a AMD Blog post that the X line of cpu's had a 20 degree offset so they read high in the UEFI but also would have a much more aggressive fan response based off that reading.
without crashing, that is...
server / workstation ver in the works? MB's that will run them with ecc.
Or an server / workstation desktop socket system?
amd needs to have a server system for the users that don't need the full EPYC but still want to have server class hardware. They systems in the $1000-$1200 range (the server case + PSU can be $100-$300 of that price) Or just towers at $800-$1000 with an server board (IPMI) and maybe have the X16 (CPU) slot cut into X8 X8 + X4 or even X4 X4 X4 X4 X4. So storage (HBA / Raid card / pci-e) and networking (10G). And be better then the intel ones at the same class that only have X16 + DMI on the cpu that have less cores.
It would cost me $13.03 (CDN) to run it 24/7 for 30 days at full load that's at $.10/kwh. I think at that point whatever I'm doing is either worth the money or its making me some serious money dripping all those threads.
by TheSpoom (715771) Uncaring Linux user here. I have nothing to add to this but please continue. *munches popcorn*
ASSEMBLE!
more PCI-E Lanes off the chipset needs an better link and not just pci-e 3 x4 to pci-e 4 x4.
The next socket may be 2019-2020 when pci-e 4 hit's. But can they boost the link speed in the same socket with zen2?
It's worth mentioning the CPU bug doesn't seem to be present in Epyc or TR, just the original Ryzens.
I've been burned too many times by AMD's claims of performance with CPUs and GPUs only to find that my games actually run better on Intel.
The money saved is never worth it, to me. I always end up wishing I had Intel.
You are welcome on my lawn.
More facebook, uber, tesla and gender diversity articles please.
Warning to those who would use Treadripper where life or saftey is involved: DON'T.
You're not going to be using an Intel in those situations either. You're going to be using an embedded CPU that meets ISO26262/IEC61508
It's a tech news site. Alternative offerings to Intel CPUs are big news. I think most of the announcements and releases are done now though.
central compute core spends 99+% of it is time idle. It is really silly...
Syntax error while processing sentence due to inappropriate apostrophe usage.
For great social justice!
It looks like AMD will have some sort of RAID support in the X299 chipset, but at launch, they don't have bootable RAID-0 support for NVMe drives.
Intel promises this on the X299 motherboards, but hobbles it with the DMI interface for motherboard-mounted M.2 slots, and the need for an expensive "VROC Upgrade Key" (i.e. DRM nonsense) just to run non-Intel parts in a bootable RAID-0 array... oh, and the "Key" isn't actually available yet, at any price.
VROC was the last nail driving me away from their platforms. Sad really, considering their RAID technology promised almost direct multiplying of bandwidth in RAID-0 up to 25+ GB/s. Intel has crippled RAID support moving forward, and there is little point to using their stuff when AMD has managed to catch up and costs much less. I just wish they'd move faster to provide decent RAID support in their X399 chipset... though apparently there is a promise to deliver support in a future update.
Just be happy the constant 3D printing articles finally tapered off.
X399 is hobbled by cpu's with only 16 or 24 pci-e on the cpu so they stack a lot of stuff on the DMI bus.
AMD has USB on die and at all levels more cpu pci-e and all cups on each socket have full pci-e lanes.
Even worse, Threadripper is an anagram of the rad ripper - echoing the famous 'Yorkshire Ripper' serial killer from last century. Would you trust your children around such a CPU?
I have never seen an article on slashdot pushing some social justice agenda but I have sure seen a ton of them bitching about it. Like most conservative things they are against a problem that doesn't exist.
More CPU diversity articles, please!
Damn. You forgot amazing battery breakthroughs! You forgot batteries!
Little bored of AMD threadripper destroys i9...
No.
AMD's best is better than intels worst..doesn't have the same ring to it.
Let's see how everything looks when the i9's are out that people want ..7940x, 7960x, 7980xe...
Remember price isn't an issue we care about the end performance not the efficiency spread of my dollar per core.
Still on the fence we see the actual intel competitors.
It's not a Ryzen article, so?
Since the TR does support ECC, there's no reason for any of the TR boards to not support it and in fact, according to the Manual - the AsRock TaiChi board does support it (unbuffered only). Don't know the difference between the two but for 8 sticks, the extra cost doesn't seem to be worth it when buffered ECC is less then $20 more on Newegg for the same speed/size
Every now and again Slashdot will do interviews with people like RMS, Linux, ESR, Krebs, McAfee, Stroustrup
How the fuck did Brianna Wu manage to score an interview then?
https://interviews.slashdot.org/story/15/07/22/1317222/interviews-brianna-wu-answers-your-questions
I'm a LIBERAL and even I recognize there is a flood of annoying young students who have been embedding themselves into companies and projects so they can write codes of conduct and promote the careers of other rich white women and transsexuals.
Uh...
X399 = AMD
X299 = Intel
BenJeremy got it wrong on the first mention (he called AMD's chipset X299), but he got it right at the end.
Most AMD MBs already support ECC, even the Ryzen ones, and the generation before that.
Most of the comments I have seen online boast about how Intel "crushes" AMD's offerings on gaming. Here's your wake up call...looking at today's benchmarks, Threadripper is within 2-3 fps on just about every major game tested when using a higher end graphics card. Sure, at 1080p, where games show more of the limitations of the CPU, you will see some more dramatic differences...but for REAL WORLD applications, a TR makes a LOT more sense than the X299 based Intel offerings. At the $999 price point, you get a LOT more performance for the dollar. If ALL you are going to use the computer for is gaming, I think you should find a different processor...but for those who are using their CPU for more, Intel is the one getting crushed.
Will supermicro make TR and ryzen boards?
Be nice to have a good non gamer board with IPMI for listed sockets.
oops the numbers are very easy to flip around.
And by the time todays horde of useless programmers finish writing software for the new CPUs the resulting performace will be about the same as a 286.
Joe Dork president of Dork industries is already delighted to announce that thanks to the performance of these CPUs his 2Tb RAM 2 x Threadripper R&D machine is able to utilise exactly one function call from 2,000,000 different frameworks.
"Our new executable is a joy to behold. Thanks to being able to leverage function calls from nearly every available library ever written, not to mention our now total use of design patterns, we are able to calculate the value of pi to 7 decimal places in under a week with a finished executable size of only 3 Tb."
Hey, did you know there's a crash bug in every freakin' Skylake / Kaby lake CPU with hyperthreading enabled? And that includes Skylake X?
Requires a microcode update.
These things happen. Read your errata.
But most hardware that affects "life or safety" is ancient, crusty, and full of security holes anyway.
Companies have to make a balance between waiting what seems ages to release information (until they are wholly certain of the cause, lest they get called out as withholders of information by your ilk) vs. releasing information in a rapid fashion (often with varying degrees of correctitude, opening them up to claims of ZOMG FAKE NEWS and LIES, also by your ilk). Funny how that works out.
"The reason for the relatively high idle power is two fold -- first, there are more cores to keep powered up versus the other processors and second, Threadripper's clocks never dropped below their base frequency."
Why tho? Are these hardware settings or could the OS control it? Maybe Linux already does? I would want most cores off and frequency quite low, with both ramping up on demand if I was using it as an all-day general-use desktop.
Forgot AppleGoogle
There is no XUL, only WebExtensions...
As far as I can see, the Asrock X399 Taichi supports ECC already and that's a consumer board.
My main problem is that I want high clocked ECC memory and that seems to be a market that has been pretty much neglected so far.
“Threadripper” is obviously a compound word. I know what a “dripper” is, but what is a “threa”?
You would prefer more google internal politics articles?
And how is it being used? "Used" is a generic word. "Leverage", while arguably fluff speech, also has a real meaning, and if the object is being used to put pressure on something (say, with a fulcrum, literal or metaphorical), then it's clear it's being "leveraged".
Cunt's a fun word but no reason to get pissy over a legit use of a word.
5 trillion cores at 3.4 ghz .....or 4 at 3.4 ghz ......really nothing really new unless you have 50 thngs going like a idiot
He called them "Lab Queens".
In other words, there were fantastic in the engineering lab, where conditions could be tightly controlled and optimized; but in the real-world, they just didn't work out so well.
So, what we have with the AMD Ryzen CPUs is something which, when benchmarks are constructed like virtually NO software actually is, they appear to kick ass. But, with software that is written like 99% of developers and their development toolchains do it, they are actually LOWER-performing than their "slower" counterparts.
TL;DR: LOL!!!
amd needs to have a server system for the users that don't need the full EPYC but still want to have server class hardware. They systems in the $1000-$1200 range (the server case + PSU can be $100-$300 of that price) Or just towers at $800-$1000 with an server board (IPMI) and maybe have the X16 (CPU) slot cut into X8 X8 + X4 or even X4 X4 X4 X4 X4. So storage (HBA / Raid card / pci-e) and networking (10G). And be better then the intel ones at the same class that only have X16 + DMI on the cpu that have less cores.
Except EPYC CPUs have fairly similar pricing, so why not just go with EPYC?
It's for production work. Workstation tasks. Things like a desktop at Dreamworks for the rendering and visualisation department, not the sales department.
In which case you want it fast. Time is money when your movie's effects depend on how much computation you can do before the deadline for release.
$800-$1200 (Full hardware cost) range server systems are desktop class. For the 1P next level it's like $1300 - $2000.
and right now there are really no 1P EPYC boards out.
For lower end needs AMD need to have stuff in the $800-$1200 range.
You are absolutely right about this. Right now there is a huge move to replace H.264 with HEVC and VP9. The encoding times are up to 5 times longer for the same content. Throw in 4K and it gets even crazier. A couple years from now content producers are going to be a huge market all the multicore magic that AMD and Intel can produce.
He called them "Lab Queens".
In other words, there were fantastic in the engineering lab, where conditions could be tightly controlled and optimized; but in the real-world, they just didn't work out so well.
So, what we have with the AMD Ryzen CPUs is something which, when benchmarks are constructed like virtually NO software actually is, they appear to kick ass. But, with software that is written like 99% of developers and their development toolchains do it, they are actually LOWER-performing than their "slower" counterparts.
TL;DR: LOL!!!
Actually not true. Where Ryzen helps in the real world is having the system remain responsive when having a million things open that uses threads like Chrome for example. Ryzen is a skylake i5 with several cores basically. An i7 has more performance per core which you are correct.
Keep in mind phones today have more cores than i7 and Intel has now woken up and redesigned coffeelake to include 6 and 8 cores.
A ryzen r3/5 is cheaper than an i5 and has SMT (hyperthreading) where you need an i7 with an intel. I think in the real world Ryzen is a great value as even the budget r3 1200 is a true quad core CPU at an i3 price! The speed difference is not that great for single tasking but things are changing and having +30 tabs in Chrome, running a game, virus scan in the background with Office will show some benefits having SMT and +4 cores.
http://saveie6.com/
Actually there is still more work and research being done. And the problem can be triggered under many scenarios involving various loads.
Usually, once a reliable way to reproduce the problem is known, identification and correction follow shortly after. Since it is easy to reproduce the problem via gcc, I presume that the cause is already known and fixed in the next stepping. The current silence would cover the period where work is being doing on a microcode-level patch for already-shipped parts, SOP for any processor vendor.
When all you have is a hammer, every problem starts to look like a thumb.
He called them "Lab Queens".
In other words, there were fantastic in the engineering lab, where conditions could be tightly controlled and optimized; but in the real-world, they just didn't work out so well.
So, what we have with the AMD Ryzen CPUs is something which, when benchmarks are constructed like virtually NO software actually is, they appear to kick ass. But, with software that is written like 99% of developers and their development toolchains do it, they are actually LOWER-performing than their "slower" counterparts.
TL;DR: LOL!!!
Actually not true. Where Ryzen helps in the real world is having the system remain responsive when having a million things open that uses threads like Chrome for example. Ryzen is a skylake i5 with several cores basically. An i7 has more performance per core which you are correct.
Keep in mind phones today have more cores than i7 and Intel has now woken up and redesigned coffeelake to include 6 and 8 cores.
A ryzen r3/5 is cheaper than an i5 and has SMT (hyperthreading) where you need an i7 with an intel. I think in the real world Ryzen is a great value as even the budget r3 1200 is a true quad core CPU at an i3 price! The speed difference is not that great for single tasking but things are changing and having +30 tabs in Chrome, running a game, virus scan in the background with Office will show some benefits having SMT and +4 cores.
You could be right, actually; especially with OSes like macOS and iOS, which uses Grand Central Dispatch (GCD) to dole out threads to various cores basically automatically and transparently to the Application Developer.
https://developer.apple.com/li...
https://en.wikipedia.org/wiki/...
BTW, Apple Open Sourced GCD under the Apache License model; so there's really no excuse for any OS not to use it! But apparently, gcc doesn't support "blocks"; so neither Linux nor Android use it (or at least not generally). But then, there's always LLVM/clang, which again, Apple contributes to (and uses) as Open Source.
Unfortunately (perhaps), Apple has not yet put anything but Intel in their desktops/laptops. But things can change, and it has been quite awhile since AMD was competitive in the CPU department; so we shall see...
$800-$1200 (Full hardware cost) range server systems are desktop class. For the 1P next level it's like $1300 - $2000.
and right now there are really no 1P EPYC boards out.
For lower end needs AMD need to have stuff in the $800-$1200 range.
Epyc 1S Motherboards:
Gigabyte MZ31-AR0
Tyan S8026GM2NRE
SuperMicro H11SSL-i / H11SSL-C / H11SSL-NC --- Coming Soon
There are a few single and dual Socket MBs. With a 2 Socket board you will need two CPUs to enable ALL the features of the MB, otherwise you'll lose half your Ports and Memory (and some Slots, but NO Lanes).
So don't say there's no single socket. You can scrap together a board and chip for under a thou, next year the 7nm CPU is guaranteed to fit. I'd pay the extra and wait for the Supermicro but the others are not 'bad'. Best to wait until they've been out a month and tested/reviewed by someone else.