The Notebook Review forums were hosted by TechTarget, who shut down them down on January 31, 2022. This static read-only archive was pulled by NBR forum users between January 20 and January 31, 2022, in an effort to make sure that the valuable technical information that had been posted on the forums is preserved. For current discussions, many NBR forum users moved over to NotebookTalk.net after the shutdown.
Published on Jun 19, 2017
Alienware is unleashing a full Ryzen Desktop with the Threadripper CPU taking center stage! Don't know what Ryzen is or the Threadripper? Check out the discussion now with Eddy Goyanes.
Want to learn even more about The Alienware Area-51 Threadripper Edition and our other E3 announcements? Find out more here:
http://bit.ly/2sKsLXO
The Intel i9 7900X is a Monster Processor You Shouldn't Buy
Intel i9 & X299 5 Reasons Not To Buy!
Intel's "12K" Blunder & PCIe Lane x8 v. x16 SLI Benchmark
Still not a lot of good reviews...it looks like it's gonna take a while for Intel to get CPU's in reviewers hands...and a while longer to get the firmware up to speed.
That is the issue, the CPU needs further benchmarks from the standard suites and compared to the new Xeon's. Remember to watch too, as those Xeon's and Epyc's can be either 1 or 2 CPU config's and I think the Xeon supports even further than 2 on some.
The difference is connoted, in part, with the series 2xxx for 2P, 4xxx for 4P, and 8xxx for 8P Xeon systems. Granted, the 4 and 8 also have higher speeds and can be used in lower processor setups, but you are paying a large premium for features not used. This is why AND focused on the 2XXXv4 processors, not the higher ones from Intel. But all 2XXX will be in 1P or 2P systems for the most part. This makes up the majority of systems.
But, with the new system with skylake-SP, any chip is usable in the stack, if I remember correctly.
I like the attitude change in this post! Thank you!
For a videographer; I'll agree with you. That is not my realm of expertise - even if I'm familiar with some of their workflows.
I also wholeheartedly agree that sometime in the future; multiple GPU's (along with more cores and even more sockets) will be the defacto standard (probably even for smart phones by then...).
The only issue with this post that I must point out? At the very high end of 'pro-sumer' and above? Nobody buys multiple workstations/servers with what they 'believe' the market will move towards (at least not in any quantity).
Rather; they buy for maximizing their current workflow over the expected lifecycle of the hardware and will consider new workflow and hardware/software possibilities when that time has arrived (and usually 'proven' by others for a few months - if not a year +).
Yeah; the abundance of cores being announced in current/future platforms, the abundance of PCI lanes, sockets and RAM will have a drastic effect on programmers and the programs they can create for us. That is in the medium future though. Right now? Anything over 8 cores that comes with lower clocks (stock) is not optimized for 99.9999% of consumer/pro-sumer/gamer workloads.
Contrary to what you may believe; the CPU + RAM is what runs the rest of the platform. A faster, more powerful CPU will make the most of every other component used in the platform. GPU's today are still too hot, too power hungry and too big (physically) to be used 'everywhere' compute is needed across various devices the masses use. iGPU's on the other hand (and SoC's) are the best balance (particularly for mobile devices) and that is why programmers - even for the 1% of users - program way below what is available and has been available for years now (i.e. HCCP).
When do GPU's get 'real' - when they have a minimum of 8GB vRAM inside. If/when GPU's deliver at reasonable prices with 16GB and higher, I'll reconsider using them across my workstations (DT's). Anything below that is like running a Ferrari with a 60 MPH speed limiter installed.
I think this is one of the cases where the hardware makers thought they were waiting on devs/consumers to optimize for/ask for multicore and devs/consumers weren't asking for it because they thought it didn't exist for other reasons. It took manufacturers jumping in first, and even if they're not being fully utilized, people buying into it is what it will take for programmers to really embrace it. So while right now you're right about it, the direction hardware goes (and to an extent what people buy) should at least in part be influenced by what will be possible in the future.
Epyc, in its current configuration, is not supposed to compete with an 8180. This is their first enty in the 1P and 2P server market so there is a big discrepancy between those two CPU's. I will wait for better comparison of the new Platinum Xeon's but my thinking is they need a bit more aggressive pricing. But again we shall see.
Also, don't assume or interpret what I said. You twisted my words in places and are going back on the ignore list.
Sent from my SM-G900P using Tapatalk
Edit: Also, ignoring my edits showing people using 3-4 cards currently are in use by people out there, the main reasons held back is worries about support or not knowing it was possible at all except for larger operations that regularly use that many cards, but are restricted by PCIe lanes. Considering that, I fully expect to see 4+ cards used by these companies on TR in the future.
As to CPU power, I directly said it was important and that ram was important. When quoting from Davinci Resolve with Black Magic, it directly says DO NOT SKIMP ON THEM. But, Considering GPU acceleration is more important and that ThreadRipper has already been shown to be strong against Intel's offerings, you have to balance the benefits. This was previously explained. Also, when GPUs reduce processing time drastically, overall they are not "too hot, too power hungry and too big" for workstations being used today. And no, iGPU's are not the best balance. Plus, a 7900X DOES NOT HAVE ONE, IIRC, making you look disingenuous.
As to the amount of vram, see the post with the link to Black Magic showing amount of ram for GPUs with Davinci Resolve. Plus, Vega has the Frontier Edition on sale with 16GB of HBM2 for $1200 or $1700 for CLC. You have Nvidia offering the 1080 and above with 8GB+ of vram. So, you are just full of it. You do not need that for Photoshop. This is professional video editing software that need and use that, and 8GB and above will allow for content of 4K and over. So, please, just stop!
I disagree. They're pricing is set just about right. You get price breaks when buying (if really doing large servers, no one pays retail). We'll see what it actually sells for soon....
I think the pricing may have been right for the old guard but not for the new one an Platinum Xeon's, but again we shall see as right now that is an opinion and guess only and the current prices are conjecture only too.
Dual AMD EPYC 7601 Set Monero Mining 2P World Record in Docker and Linux
Published on Jun 22, 2017
We take two of the newest generation of AMD EPYC 7601 CPUs and see just how fast they are eating spare CPU cycles for Docker and Ubuntu Linux based crypto mining (Monero). We set a world record by almost a 2x margin over previous generation dual socket configurations.
Please do note, there are additional optimizations possible but we are using our scale out images based on Wolf's CPU miner.
This result is not indicative of general purpose workloads on AMD EPYC. It happens to use exactly 2MB L3 cache per thread which is the exact CPU core to L3 cache ratio on EPYC. This example is not hitting AMD's Infinity Fabric in a substantial manner. We are also on pre-production firmware so this performance may increase as systems become available.
Says who exactly?
Are you the authority on the subject matter?
I don't think so.
Also, Sandra barely included updates for what is still an unreleased product.
It had good deal more time to optimize for Intel architecture from the get go (giving it the advantage).
Oh and, not even at $1?
Vs $12 000?
Are you being serious?
Who in their right mind would want to spend 12 000 times more cash on 50% more performance (assuming that number turns out to be accurate in the first place).
One would need to be seriously deluded to do that.
Also, cost efficiency would like a word with you.
Do you honestly think that businesses would want to spend $12 000 when they could get same performance for $8000?
Seriously, even if EPYC ends up with 50% performance of Xeon Platinum (which we don't know)... it's still 3x lower cost.
Various companies would be getting 1.5x performance for the price of just 1 Intel system and far more cores to boot.
You are also forgetting that the energy efficiency of the CPUs is much higher with the EPYC chips than Xeons, or so the information to date has said, meaning that the cost of running is lower, as well as a decent footprint on size. His hyperbole is maddening. Time will bear that out for certainty.
u guys gotta know though, not like intel not giving out discounts to their customers i bet they are and trying to keep as many as possible. what they did with skylake-xeon early releases to amazon and google i think is one of the trick they did, release it half a year to 8 month early to their biggest clients so they dont lose them to IBM or AMD.
also giving discount and incentive to not buy AMD isnt that bad practice? though they were fined $1 bill for something happened years ago, its barely pocket change for them and might attempt it again.
i really do wish people just bite the bullet and buy AMD this time just to show intel, the way they milk money isnt the right way to treat customers and hopefully they'll learn it and not have to repeat the same milking strategy for at least another 5-10 yrs lol.
While impressive SiSandra to me shows nothing to workload. But more and more info leaking in. I am ready by 6/30 to make a decision one way or another, to wait a bit more for the unknown or to head off to settling on a known land.
i love the arguments in here with exception that people LOVE to point out simple and silly mistakes and make assumptions, just part of human nature when thing heats up. keeping it civil and share opinions and go down to the simplest level and finding facts/opinions which two people disagreeing is the first step toward moving forward in debate imho.
the take away of my message is, you're all wrong and im right. skylake-x 10c 5ghz HERE I COME BABY
All I know if your workflows are what you have said here.
Why I'm explaining, in excruciating detail, specifically video editing is it was a low hanging fruit to disprove what you said, which was Intel wins in every way that matters.
What I take issue with is when you mischaracterize my statement to say I said something I did not say. Further, it is your constant cheerleading, instead of true discourse, that I find the most painful, due to your intellectual dishonesty.
Ok, now this HAS TO STOP. The incessant fighting that is. No more name calling, insults, rude comments or otherwise! Next time warnings and from there some temporary bans are in order.
Good points or not I had to do mass deletions because of the ongoing battle!
So, here is a video where a reviewer performed the video render on Ryzen blender for the new Intel Skylake-X chips. The TR scored 13.04 seconds at computex, but we do not know the system configuration. The Ryzen tested in December last year got 36 seconds, although people can now reach 24.3 seconds (posted in the AMD Ryzen thread, linked from another forum) with the improvements in AGESA, memory speed, using higher than 3.4GHz, and other elements of support.
http://www.legitreviews.com/amd-ryzen-threadripper-runs-blender-benchmark-computex_195121
Another comparison with the new X chips for benchmarks, this time from PCWorld. On POV-ray, I cannot tell if they used the optimized POV that came out after Ryzen launch for Ryzen (which lowered the score slightly) or retested the intel 6000 series, which received a large boost in performance after the update (which may make the generational shift in score less than shown). Need more info, but currently watching it, so...
EDIT: I also want to point out with the WinRAR test that Ryzen is great at decompression, but bad at compression. If you do more compression than decompression, especially in significant amounts, Intel is much better for that workload. Very important to know your work. They also say the mesh is likely the reason for lower performance in this test for 7900X (latency hit?).
Edit 2: Cinebench, 2.5GHz, Single thread IPC comparison:
Disclaimer: this is one test. Depending on optimizations of the task, the IPC may vary by task. This can be affected by the use of cache optimizations, scheduling, etc. But, it is still an important measure. And yes, Instructions Per Cycle can vary between tasks depending on how it is written to utilize certain aspects of the architecture, as well as how it works around the latency of the architecture. But, this is one that AMD has consistently used to show performance, so the IPC variance in other tasks may be greater (negative, although certain tasks may be positive from this), so I wanted to put that out there.
If there is no loss of clock speed and 100% scaling the 10 cores at 17.9 seconds would be at 14 cores 12.8 seconds. There most likely is some scaling loss and speed decrease so Intel probably has it about right were as a 16 code Intel will be close to a TR but it needs 18 cores to be faster.
Also, notice the improvement from the demo to release to platform being stable for Ryzen. Some of the software optimization, including blender adding Ryzen R7 optimizations, may have helped TR already. But, we don't know if the demo was running TR below final speeds or if bios/uefi advancements will help, nor the ram speed used. But I do agree, Intel may have done 18 to guarantee dominance, but at the cost of energy efficiency.
Further, AMD got added to the DoD research because of the efficiency found with Ryzen (exascale computing). So, we have to also found out what the performance over TR is with the energy used to obtain the performance.
I'd also like to point out that without the 10 core and Ryzen 1800x overclocked, Ryzen hits 1600 in cinebench, the 7900X hits 2100. Now the 10 core has 25% more cores. If we add 25% to the 1800X, we are in the 2000 range, thereby nipping on the heels of the 10 core. Of course scaling, the stepping revision, etc. May help TR, but we cannot know that yet. So, should know in the next month and a half.
Overclocked we have seen where the 7900x can get ti 2400+ on R15. I have seen though some concerns with the x299 being able to carry out the power to push high speeds on the higher core count CPU's sowe will see in the end where it all gets us. As a side nore too, more core counts means larger acutal chip and more cooling required from the same cooling solutions.
intel frequency king, its what i need a 10 core 5ghz would be so good, even at 4.7 4.8 is already so good tbh with that many cores. concern is deliding and CLU going bad after a yr etc.
right now just waiting for more reviews from VROC and turbo boost max 3.0 with 2 cores instead of 1 before jumping in, and waiting to see if 14-18c will be soldered. what do you guys think boys.
Delid, that sounds right (might be 4.6-4.7 after, I don't know your heat target on the chip (70 or if higher is acceptable).
As to the HCC, I'm not betting on it. This keeps warranty claims down AND would allow for differentiation between server chips and HEDT. But, who knows if Intel will bend to public ridicule (ha).
Now, I could bring up GloFo 7nm target of over 5GHz (to give perspective, the 14nm process used had a goal of 3GHz and reaches 4 on mainstream, but the server chips they were talking about hit 3-3.2 on boost for the better chips, while being in mid to high 2GHz range for all core clocks).
But, it sounds like you've mostly made up your mind this round and that will be a great build (even though I still hold 12 core has a better value on Intel this round).
Problems? See this thread at archive.org.
