Slashdot Mirror
Preview of AMD Ryzen Threadripper Shows Chip Handily Out-Pacing Intel Core i9 (hothardware.com)
MojoKid writes: AMD is still days away from the formal launch of their Ryzen Threadripper family of 12 and 16-core processors but OEM system builder Dell and its Alienware gaming PC division had an inside track on first silicon in the channel. The Alienware Area-51 Threadripper Edition sports a 16-core Ryzen Threadripper 1950X processor that boosts to 4GHz with a base clock of 3.4GHz and an all-core boost at 3.6GHz. From a price standpoint, the 16-core Threadripper chip goes head-to-head with Intel's 10-core Core i9-7900X at a $999 MSRP. In early benchmark runs of the Alienware system, AMD's Ryzen Threadripper is showing as much as a 37% percent performance advantage over the Intel Core i9 Skylake-X chip, in highly threaded general compute workload benchmarks like Cinebench and Blender. In gaming, Threadripper is showing roughly performance parity with the Core i9 chip in some tests, but trailing by as much as 20% in lower resolution 1080p gaming, as is characteristic for many Ryzen CPUs currently, in certain games. Regardless, when you consider the general performance upside with Ryzen Threadripper versus Intel's current fastest desktop chip, along with its more aggressive per-core pricing (12-core Threadripper at $799), AMD's new flagship enthusiast/performance workstation desktop chips are lining up pretty well versus Intel's.
When setting a mug of coffee on the AMD CPU it will heat it faster than the puny Intel CPU for the same amount of processing!
Gaming is great and all but my real interest is on the computing power per Watt. This is a tech site and I would think people would want to know if datacenters are about to switch their boxen to AMD in the near future. This actually is something that matters.
Anons need not reply. Questions end with a question mark.
Um if you only look at synthetic benchmarks yes it does win but sadly rest of the results are don't put it so amd chip's way.
Could you revise your post so that we know what "it" refers to with regards to winning? And could you revise your sentence "sadly rest of the results are don't put it so amd chip's way." so that it uses grammar and makes sense? I would like to understand your point.
A dingo ate my sig...
What?
Have you seen the real power consumption on Skylake-X?
uh... electricity is cheap dude.
You are wrong. https://img.purch.com/o/aHR0cD...
Ryzen 1700 uses 35W less than a 7700k and 1800X uses 25W more. In gaming a Ryzen uses around 15% less which is typically the upper end how much slower it is in games compared to a 7700k. E.g. it is as efficient (games) or tons more efficient (when all cores can be used) than a Intel i7
Intel however is certainly ignoring their own power envelope with their factory overclocked CPU and from all news, their Skylake-X are worse, even the low end chips, in their mad dash to beat AMD. I doubt this will change with Threadripper which uses the same dies as Ryzen.
It doesn't matter if it's AMD or Intel: they always ignore your mythical "power envelope", especially when they are behind like Intel now and AMD before or when they have to press out the last bit of performance from an aging architecture like Intel now or AMD with the 9590.
Is it possible to build an AMD Threadripper Hackintosh? The performance data looks very good, high performance, low power. Time to rip some threads!!!
Seems that if you max out the Ryzen on Linux or BSD, it can (under certain conditions) cause a reset:
https://hothardware.com/news/freebsd-programmers-report-ryzen-smt-bug-that-hangs-or-resets-machines
It's not all that surprising that gaming benchmarks don't scale as well to large numbers of cores. Videogame programming isn't a field in which performance can simply scale nearly linearly based on the number of hardware threads available. That's because the CPU is performing a huge number of very diverse tasks among all it's engine components, and there's a great deal of global coordination that occurs on a central database. It's essentially a heterogeneous workload, and those just don't scale as well.
Thinks like a 3D renderer or video encoding benchmark, in which you can divide up portions of the screen or encode successive frames on different threads - those sorts of things will scale nearly linearly with the number of threads, because it's a largely homogeneous workload.
So, it's not necessarily about synthetic vs read world benchmarks. It's really about how well the application in general scales to multiple threads. For some things, it's relatively easy. Others... not so much.
Irony: Agile development has too much intertia to be abandoned now.
Threadipper at $550 has more pci-e then Intel at X2 the cost.
For $599 you still only get 28 lanes with intel
Even on the desktop not high level you get more as well.
I consider myself a gamer, but I WILL NOT burn through the power that a GTX 1080 consumes.
So you wouldn't buy a $500 graphics card because it'd cost you $5/year in electricity. Got it.
Live today, because you never know what tomorrow brings
Are we still waiting for these mystery drivers/patches to make any new AMD CPU decent at games? What processor do you buy if you want raw grunt and be good at games? Hint: it's not AMD.
Just a double whammy there. The new Intel CPUs aren't compatible with the old motherboards
http://www.pcgamer.com/intels-...
It looks like they are practically driving people AMD's way. Nice to see the shakeup though it's been far too long.
"I WILL NOT burn through the power that a GTX 1080 consumes"
I bet you had no problem saying that while running a fucking GTX 780, at nearly DOUBLE the TDP of a fucking Threadripper or GTX 1080.
Still waiting on Serviscope_minor to wake up to fucking reality and realize that Jessica Price isn't going to fuck him.
He said that porn doesn't feel the same on an AMD chipset.
You mean his Ball Grid Array needs better polishing?
Browsing at +1 - no ACs, I ignore their posts. So refreshing!
Comment removed based on user account deletion
... at the cost of 30% of the performance of the chip...
Deja-vú...
Buldozer anyone?
*** Suerte a todos y Feliz dia!
Summary of the video: 16 core threadripper beats 10 core i9 in a highly parallel job. Wow, wow, wow.
Bleeding edge performance always consumes a lot of power. Doesn't matter if it's computers or cars.
If builders built buildings the way programmers wrote programs, then the first woodpecker would destroy civilization.
For home use then gaming performance counts, in business TPC-C may be more interesting.
If builders built buildings the way programmers wrote programs, then the first woodpecker would destroy civilization.
Yep. What I see is a 16-core AMD CPU losing out to the 10-core Intel CPU in all of the 3DMark tests and Tomb Raider, which is the only real test that was conducted.
AMD's Ryzen 1950X has 60% more cores and a 100MHz clockspeed advantage and it still can't keep up with Intel.
Zen/Ryzen has better performance per watt than any Intel chip currently offered and the next 6 months or more. You are either Intel fanboy/shill or you have been living under a rock. Tom's actually measures the wattage draw directly from the CPU, 4-core Intel draws more power than an 8 core AMD chip. http://www.tomshardware.co.uk/...
Zen/Ryzen has better performance per watt than anything Intel has currently or will offer for at least the next 6 months or longer. Example a 4-core 7700k Intel chip uses more power than a 8 core Ryzen 1700. It is basically impossible for Intel to get wattage down without a new lithography and arch, which isn't happening for more than 6 months or longer, cannonlake got pushed back until second half 2018, 10nm isn't going well.
Dammit forgot link http://www.tomshardware.co.uk/...
You get five installs of Windows 10 Pro. You won't have any problem activating on a newer PC.
What are you talking about? These benchmarks show a 35% performance gap. Other benchmarks (do a quick google) show a 16% power penalty. The Ryzen is more efficient based on the benchmarks we've seen so far.
Judging by their reaction and rushed 'counter' release, they where caught with their pants down. And this is a good thing. Intel have been resting comfortable at the top for a while now, so this will be good for all of us. Now while highly unlikely, if only AMD could pull some similar stunt against NVidia also..
But you have, it's blindingly obvious. Either that, or you're actually retarded.
Look at the percentage difference in power consumption. Then look at the difference in cost of acquisition. Now, think about how much electricity you can buy for that sum.
You'll find you'll be burning an awful lot of electricity before the Intel even hypothetically begins to pay for itself.
On the other hand AMD's GPU are still much more open-source friendly compared to Nvidia, and at the same time are still relevant when compared to Intel (even if not as power-efficient as Nvidia).
AMD has Linux devs on their payroll, is supporting 2 stack one of which (their long term goal) is opensource (runs a classical DRI/Mesa stack), while the other (eventually targetting for professionnals who need some weird features) leverages the same kernel driver.
The AMD opensource drivers are decent, offer support for most of the hardware (except the current DC/DAL delay) and is the official stack for older hardware.
Means that AMD graphic cards "just work" with the latest version of your favorite rolling distro (Opensuse tumbleweed in my case).
(BTW: Intel is also providing opensource drivers, but completely different than their windows stack)
Meanwhile, Nvidia only provides a blob which, while mostly decent, has several problems :
no way to fix it by 3rd parties for newer kernel (so you're stuck with which ever kernel version they decided to support)
and they only support a small subset of advanced features (optimus used to not work for quite some time),
also lots of problems on platforms which are not the standard desktop (sleep and mode-setting working badly on some laptops).
The only alternative is Nouveau which, while having achieved quite some progress, still needs to be done nearly enterily by reverse engineering Nvidia's drivers, all done only by volunteers with very little actual help from Nvidia.
Meaning often some basic stuff (reclocking) isn't working.
"Sufficiently advanced satire is indistinguishable from reality." - [Tips: 1DrYakQDKCQ6y52z6QbnkxHXAocMZJE61o ]
Software capabilities traditionally lag behind hardware capabilities. Look at how underutilized the multi-core capability of modern hardware is even today. Now consider that GPGPU is even newer than multi-core CPUs. Thus, logically, most 3D rendering software *isn't* designed to use the GPU to do its work. 3D rendering software designed to use the GPU is *just appearing*, and even where it exists, it can't be reasonably used for complex scenes, since those have traditionally been memory-limited. High-end 3D production was using multi-gigabyte assets (around 10 GB per frame) around the year 2000 already. An average graphics card is just getting there, but the industry has moved on already.
Ezekiel 23:20
That's because the CPU is performing a huge number of very diverse tasks among all it's engine components, and there's a great deal of global coordination that occurs on a central database.
Which is more likely to be just old architecture. In general, game worlds do exhibit at least some kind of locality or metrics amenable to distributing work. If nothing else, it should at least allow the emergence of new classes of games. For example, virtual worlds with smarter NPCs with actions based on actual reasoning or emulating more complex economies and such.
Ezekiel 23:20
That particular CPU will most likely beaten by the same 10-core Intel CPU at games. ;) After all, the Skylake-X parts that had already come out have already been found out to be underwhelming compared to other Intel CPU options for games.
Ezekiel 23:20
"...but AMD is doing it at the continued cost of a significantly larger chunk of electricity..."
Talk about "having a perpetual favorite" :)
Ezekiel 23:20
"...but AMD is doing it at the continued cost of a significantly larger chunk of electricity..."
Talk about "having a perpetual favorite" :)
Ezekiel 23:20
For rendering it comes down to how the scene is set up which is based on the viewport of the player as well as the action in the game. There is no way to predict how many points or triangles there will be on-screen at any given time and you can't run these calculations in parallel because there is only one player. It's not an architecture problem. It's a nature-of-the-problem problem that has been well known in gaming since at least Quake 3.
I know a lot of you don't think small power consumption issues are a big deal but I thought I'd highlight a few points:
x) We now operate in a space where the physics of chips well into the future is already known, planned, and targetted for production. 7nm and below is atoms-wide production that we have been theorizing about for over a decade and what you are seeing is the culmination of a lot of that work today. This requires high skill and high tech to just be able to prototype, let alone mass produce. There are only a few companies in the world left that have the money to do this and Intel, AMD, and ARM are all on the short list.
x) As we get smaller, heat exponentially gets to be a larger problem. Making things faster in smaller spaces trades off heat as a waste by-product and this has been ramping up significantly since the Pentium chips. I remember reading about the first ones being water and/or ice cooled (right here on /., iirc).
x) The current drive for chip production is probably going to data centers (think AWS). Data centers do not care what chip in the box. They need chips that run cool and use the least electricity possible because they have a lot of them working a lot of the time. I have seen several analysts say that Intel is so far ahead here you will never see AMD in data centers at this point. It is simply not affordable based on today's market rates for computation.
I hope AMD stays around for a long while. I have an FX-8350 right next to me. That being said, the chip industry is run completely on physics at this point and things are going to start getting weird.
Scenes are hierarchically bounded and subject to geometry metrics. I don't see how this is subject to the number of players. The problem until recently was the API limitations of the drivers. And even that is covering just a small part of what a complex modern game ought to be about.
Ezekiel 23:20
That 18 core Intel processor will probably cost $600 more than the 16 core Threadripper processor, though. I'm not sure if the 15% performance boost will be worth the price.
Because of optimization. You don't render all points in a space. You only render points in a viewport. That means part of the job of the game loop is constantly keeping track of which points in space you need to worry about for rendering purposes versus physics purposes. There are tons of algorithms to do this and it is one of the areas that people optimize pretty aggressively.
Once you have a point list, you can build a scene and render it based on a ton of other factors (z-order, shaders, shadows, light blending, etc.). All of this is completely dynamic and can't be pre-computed. The assets that are pre-computed are basically just hints that tell you how to do all of these real-time calculations faster. Also, a lot of optimization goes into the 3D models themselves to eliminate unnecessary triangles. The more triangles in a viewport, the more calculations you have to do, and the slower the engine will be.
Again, a lot of this can't be done in parallel because it doesn't depend on the number of players. It is a problem with the fact that you are multiplexing an entire universe into a single viewport and trying to optimize it so that it actually runs. Some things can (and are) run in threads. Like I said, they've been working on this for a long while now.
"Topping out at less than a third the power consumption of a 1080" yeah, except not. Normally I wouldn't even both correcting just another ransom person spouting completely wrong information. But you have been going on and on about it, talking about how you *refuse* to use something so power hungry and inefficient, and how you're so clever cause what you use is *of course* way better perf/watt.
That would be nice and all if it were true, but it's complete bullshit. TDP of a 960 is 120W, per Nvidias site. See for yourself: https://www.geforce.com/hardwa...
The 1080 TDP is 180W, Nvidia significantly improved the power usage of the 10 series compared to the previous 900 series cards for equal or greater performance levels, like the 1080 vs 980 ti. https://www.nvidia.com/en-us/g...
If you were actually right, your continued mentioning of it would be kind of annoying, but that's ok. But since you're actually not even close to right, it's just obnoxious to keep reading the same bullshit spewed out over and over again. So you aren't oddly principled about using only the best perf/watt components, you're just ignorant. Sorry.
I don't know where you have been, but graphics processors have been used for 3D rendering for a long time.
While no where near the power we have now, SGI was making dedicated 3D chips that were utilized not only in the creation of 3D scenes, but also in the final render. This was over 20 years ago. Professional houses have been using PC cards all the way back to the Voodoo 2 in 1999.
Now it would be almost unheard of, for any final rendering stage not to use the GPU.
Heck ILM has their own rendering plug-in with customized graphics drivers to try to cope with the rendering load.
No, a graphics card cannot handle all the textures, polygons and shaders needed to render a final scene, but they don't have to. They load in what is needed at the time, render their part, then load in the next part, only keeping the frame in the card's memory.
Actually it is very common on blockbuster movies for multiple cards to be working on one scene at the same time with each card rendering a section of the frame.
Microsoft, Apple, Google, Amazon what's the difference? All steal money from devs and control with walled gardens.
If you buy a retail license, it's transferable. If you buy oem, it's tied to hardware and cannot be transferred. Been this way for 20 years, AFAIK.
Why would you bother to reply, then? No one gives a shit that you don't give a shit. You added nothing of value. To fucking brag about chess? FFS.
Probably not actually... AMD has always had excellent branch prediction for instance, and they have a micro op cache just like Intel now.
The micro op cache probably does have a massive effect on synthetics... which is why it was ignored for bulldozer. Bulldozer was actually competitive when given real workloads in many cases. The fact is a Ryzen CPU is very similar in many ways to an Intel one... just different. The similarities are there mainly in order to remain competitive at ticking check boxes.
In the end the difference in size is probably due to shorter pipelines on Ryzen 7.... which is also why it doesn't clock as high but is more efficient at the same clock since there is less overhead due to pipelining as much as the Intel processor.
If you are running code that requires inter thread communication to be fast... it will be slow on all processors anyway, its something you design out of your code no matter what it is running on. The only place Ryzen really loses is when running with slow ram... the inter CXX links run slower, and AVX512 code like is probably in video encoders that is an extremely small portion of code in the wild.
Videogame programming isn't a field in which performance can simply scale nearly linearly based on the number of hardware threads available.
Actually, it is. But there is a limiting factor: videogame management are cheapass bitches who can't stomach the idea of paying more money than they pay themselves to bring in the kind of engineer who can design and develop high performance parallel code.
When all you have is a hammer, every problem starts to look like a thumb.
And that is exactly what you are seeing in game benchmarks... "BOOST".
The ThreadRipper probably still had plenty of cores on tap in those GPU benchmarks. While the Intel CPU was taking advantage of its longer, higher clocking and more wasteful pipelines to boost one core a bit faster to drive the GPU a tiny bit harder.
Lets not forget that ThreadRipper and EPYC have twice the IO and 50% more memory bandwidth than their competitor chips.
128 Lanes on EPYC... I'd laugh manically at this point but it goes without saying.
No prob. Just chat with MS, and they will help you.
Not true.
Intel is continuing their idiotic practice of using poor quality thermal goo between the chip and the package.
This is not only untrue, it's the opposite of the truth. AMD's choice to group cache around units of 4 cores kills inter-process communication between cores when more than 4 cores are in the same package, Software has already been written to partially get around this problem, but it remains a design deficiency.
Contribute to civilization: ari.aynrand.org/donate
This summary and article gave me the kind of information I can use. These "synthetic benchmarks" are actually my intended workload. Games don't benefit greatly from multi-cores over about 3 or 4 these days, and in the future I think they'll start to take advantage of the newer platforms.
My Phenom II 1055T has served me well, but I'm thinking something a bit more high end might be a good upgrade.
All that said, is there something I'm missing? Can someone suggest an even better upgrade? At home I'm mostly playing around in Unity3D/Blender3D and making videos.
System specs:
(Off-site backups and on-site backups to other machines, of course!)