Ryzen vs i7 (Mainstream); Threadripper vs i9 (HEDT); X299 vs X399/TRX40; Xeon vs Epyc

Well if the MSRP is sub $500 I'm happy. Initial demand will drive prices up, but still much better than people saying it was going to cost over $600. Cheaper and faster than the 7820X.

Intel 9th gen CPUs have mixture of hardware and software fixes for Spectre and Meltdown.

https://external-preview.redd.it/tU...pg?s=4d64b17445f5baf534407ed93501f6e17c9419ed

 

Does the mainstream have those fixes? We already knew the HEDT platform had the fixes.

Also, MSRP doesn't mean it will actually cost that, and this isn't just initial demand, it is LIMITED SUPPLY! We already know about the shortages and Intel even said they won't have them dealt with until next year. So we cannot know that the shortages on the new chips will not be due to supply issues, rather than pure demand making supply inadequate. That is likely the reason for the high MSRP anyways, allowing the price to limit sales, as well as trying not to too heavily cannibalize HEDT sales.

So, that price they gave does not have much, if any, bearing on what the price will be in the marketplace, as seen with all products recently in this sector.

 

Well I paid $529 preorder from BH Video which is a mere $41 above MSRP, which I would have paid in sales taxes anyways locally or through Amazon so I'll call it a wash.

 

Fair enough. And of course I wish you well and great joy in your hardware pick (trying to make clear I don't care what anyone buys, whether AMD or Intel, just that people are informed). Looking forward to benches and performance numbers. You putting this in a laptop or desktop or testing on both, then putting it in one or the other?

 

The 8700K will go to my laptop, and the 9900k to the desktop. I will however likely test the 9900K in the laptop for "research". May as well figure out how hot it gets for others. I seriously doubt anything but stock clocks will be used but we will see.

 

Is there a delivery date yet for your order? The Intel presentation said available starting Oct 19...

 

No delivery date yet. I ordered last night minutes after it went live on their website around 1150pm on Oct 7. I have a feeling they may be slower to get their chips as usual for them but we will see. I always have Newegg which I'm waiting for them to drop theirs online.

 

Intel Surrenders to Threadripper With New Skylake-X Refresh
Hardware Unboxed
Published on Oct 10, 2018

Spoiler: Click to See Video

 

im more interested in the latest rumour 13% IPC uplift. that would overthrow intel by a long shot, especially if it comes with 12 or 16 cores. if the IPC is even or even if its a tad lower by like 1-2% in single threaded CB15 it'll still be great. 16 cores at 4.5 or 4.6ghz hope that comes in a laptop.

 

Which rumour is this? The one that has been out for awhile saying 10-15% IPC over Zen+ and around 20% over zen 1?

To put this into context, Intel's current IPC advantage is 7% over Zen and 4% over Zen+. So, by powers of math, that would give AMD an IPC lead of around 6-11% over Intel, which is significant. At 10%, ignoring other factors like smt vs ht, etc., it would mean Intel running at 10% (or around 400-500mhz faster, roughly) would be at about the same performance, depending on other factors. With smt and ht factored in, it adds 100-300MHz that would be needed in addition to the speed to overcome the IPC performance, potentially.

Edit: also, the 4.5 boost was on early silicon. For comparison, 3.4 boost was on the original zen early silicon, with all cores boost on the final silicon being 300-400mhz higher. Now, I take it 5GHz is harder to hit, but that could place all core boost around 4.7, give or take a couple hundred MHz.

Edit 2: just saw the rumor of 13% in science etc. workloads. Now I've got context.



Sent from my SM-G900P using Tapatalk

 

the article on wccjunk. im assuming it'll be similar to that of zen+ vs zen, 3% ipc increase with remaining frequency boost. but they did mention msg tweeted was at same frequency, zen+ vs zen2 with 13%+ average ipc uplift for science/math related workload, which involves cache/ram heavily.

assuming ram speed / timing are the same for both system.. this would mean theres decent improvement in either IF or cache rework, or both, maybe not so much as the actual core itself. all speculation of course but i want it to be true.

 

So, here is the skinny. The original rumor came from Bits & Chips twitter account. The twitter account said that it was IPC, that the frequency was not told to him at which the chip was run, and it was in a Server setting meaning low ram speeds (we ALL know how Zen loves ram speed, unless they did something to divorce the IF speed from memory timings).

Here are the links so you can see the origins and other coverage that isn't WCCF. This is full 13% IPC, though, and combined with the prior rumor of someone at Radeon, the sample they saw, which is early silicon from around Sept. 13, ran 4.0GHz base and 4.5GHz Boost. Remember, this is very early silicon and expected release is around March or April for that chip, so literally 6-7 months from the time they were playing with the chip, which is why I brought up the speed of the Zen early silicon from the December before release, which was 3-4 months before that release. That means that speeds will likely change as they finalize things, but those speeds and that IPC already before volume, that is looking pretty nice.

Also, Bits and Chips mentions it was the same source that gave all the correct info they heard before Zen first gen was released.



Here are some links
https://mobile.twitter.com/BitsAndChipsEng/status/1052194745647165441
https://www.svethardware.cz/bits-and-chips-amd-zen-2-nabidne-13-ipc-oproti-zen/47780
https://www.overclock3d.net/news/cp...e_reportedly_offers_a_13_ipc_boost_over_zen/1
https://www.techpowerup.com/248642/amd-zen-2-offers-a-13-ipc-gain-over-zen-16-over-zen-1
https://wccftech.com/amd-zen-2-7nm-cpu-13-percent-ipc-increase-rumor/

 

bits&chips pretty good so this will keep my hope up. also have to note, server low ram speed but got 8 channel, so bandwidth isnt all that lacking. even if 13% ipc uplift is only for specfic workload i'll still be very happy, a lot of my stuff are cache/memory heavy related so it'll be a nice boost in terms of core + IPC if it does come with 16 cores. the dual channel memory on a 16c 3800x is a bit disappointing though. now just have to wait until clevo or some company willing to put ryzen into laptop that has socketed GPU as well.

 

You will need +10 cores if you shall see a gain over 9900K in laptops. And I doubt we will see it next year. Maybe not even in 2020.

 

@12nm you may be correct, not too sure yet @7nm. I would hold off on such silly statements until actual silicon is here. The reverse should be true of AMD fans, we do not know what it is until it is!

 

it has to be at least 12 cores, 16 is the preferred number but its all hopes and dream right now, not entirely impossible there are leaks/rumor around for 12/16c zen2 cpu.

ontop of at least 12 cores, need at least 4.5ghz. either 4.5ghz with similar intel IPC, or higher IPC and i wouldnt mind below 4.5ghz. but who will make laptop for it?

 

I can't see 10 cores come so it need to be 12 cores. But it will probably be same no tagged X cpu. Aka forget max clocks in laptops.

65w 8 cores R7 2700 run 3.35GHz or is it 3.6GHz on all cores boost? Double the cores up to 16 and put 4.5-4.6GHz boost on all cores (this in similar Ryzen laptops as we have today... It will fail. I talk about laptops and no X tagged cpu.

Edit. Todays Ryzen Threadripper 2920X 12 cores is 180w chips running max 3.60 GHz on all 12 cores. Put it down to 65w with coming 7nm tech will not give you the clocks you want. More power efficient aka go down from 180w to 65w and still run high clocks like 4.5-4.6GHz in laptops ... Nope wont happen.

 

since we are overclocking those tdp dont really matter. AMD has advantage here being new arch along with 7nm shrink, it'll be quite a bit more efficient than intel's cpu for sure at its most efficient frequency. I would dare say a 4.2ghz at 12 cores would be similar TDP of intels 9900k at 5ghz OC.

these are not really considered as over estimation i really think it is possible, but we have big problems, NO support. companies have to make the laptop, it must come with good bios, good cooling, AMD OC software must work. biggest problem i'd see is that throttlestop not work with AMD's CPU. I need TS real bad.

 

So, moving the discussion on soldered TIM here. This will give context. I am posting primarily videos from Der8auer, one from Gamers Nexus. It will show that the AMD implementation seems to be a thinner layer of Indium. The AMD Ryzen direct die video shows that going from soldered to direct die, Der8auer got 1C delta on max temp and 4C delta on average temp. This is with removing the IHS which acts like a shim and can increase temps by upwards of 4C. So, AMD did well with soldering the chips.

As for Intel, we have seen delidding and removing the solder on the chips, but keeping the IHS, caused a decrease of around 8C. Now, that was not going to direct die, so some will scream it is not a direct comparison, but removing the shim that is the IHS should, theoretically, allowed for a larger delta than keeping the IHS, meaning it already being half of what Intel's delta is for soldering is very telling.

Discuss....





https://overclocking.guide/the-truth-about-cpu-soldering/

 

So, another random thought. Since mindfactory showed AMD for September vastly outselling Intel, I wanted to look at the Amazon best sellers list for CPUs.

https://www.amazon.com/Best-Sellers.../pc/229189/ref=zg_bs_pg_1?_encoding=UTF8&pg=1


As can be seen, AMD has about half of the top 40 spots, 9 of the top 20 spots, and the only one with an HEDT chip on the top 20 list, and 3 HEDT chips on the top 40.
For Intel, the 8700K is a no show, but the 9700K is in number 8. This is likely due to availability on the 8700K currently and the price of the 9900K combined with the limitation in quantities, where one person claims Newegg said they were apportioned around 80 to 90 chips, but has orders for over 1000 currently.
There might be an addition of the 12-core 2920X to this list soon, as that comes in at $650 and is set to release at some point this month, supposedly. The 24-core may or may not make it into the top 40 on the basis of price, but who knows (likely grouped later with Intel's HEDT chips which is where the 2990WX is located).
Also, I have no idea how Giga's MB made it on the Best Sellers list for CPUs, but....

I don't know if they change this list based on IP geographic location or not. If they do, then this only holds true for that location. But very interesting list.

 

isnt titanium real bad at heat transfer? sand that sh*t off!!

 

They use it as a better barrier than the sealed silicon to prevent issues of electron migration, IIRC. That or making the gold stick to the silicon block, needed to help with the wetting of the indium. Been awhile since I read up on it.

 

Intel kills off the 10nm process
This is actually a good thing for the company
Oct 22, 2018 by Charlie Demerjian
http://forum.notebookreview.com/threads/intel-kills-off-the-10nm-process.825515/

 

Not a bad gaming rig, IMO, for $1K. I would swap out some of the corsair parts for other parts, but reasonable build.

 

It seems weird to pair a $150 CPU with a $400+ GPU, but I suppose it makes sense considering the non-Ti 1070 and Vega 56 aren't significantly cheaper.

 

This is to game at 1440p, not 1080p, hence why most would not think of a 2600 or 2600X as an option seeing the 1080p frame rates. Once you step above 1080p, which a 1070 Ti or 1080 would allow, the frame differences between Intel and AMD narrow considerably (and disappear at 4K, even with higher end cards than the 1070 Ti). It also will depend if we are talking getting 60fps, 120 or 144fps, etc. If you are targeting above 60 fps at 1440p, then you would not consider the 1070 Ti or a $1000 gaming build at all. So this is to match the build with what you want on performance, and for 1440p@60hz, this build is really something to consider. You are saving on the CPU to get a better GPU.

So, let's consider looking at the 8700K. You just took $200 from other components to get that CPU. If that came from the 1070, you would have to look at 1060s, 1050 Tis, 580s, or 570s. Those cards are fine in their own right, but you no longer are really looking at 1440p anymore. So, you got your 1080p frame rate up, but at what cost? You could maybe also scrimp on the ram at that point to guarantee a 1060 or 580 (which for that price on ram, you could get 3200MHz sticks, IIRC), but then you would have to drop to 2400 or 2666 and try to overclock the ram, which if you don't know what you are doing, that can be a chore, even if you use the DRAM calculator (should be seen as a guide, but knowledge on how to get your voltages right to make them work properly and stably is a different matter, as well as qualifying the machine so you don't have to worry about data corruption).

I understand why you have your hesitance on the pairing, but it actually is an amazing value when you look at it from that standpoint. Also, if OCing all cores, you then can get a 2600 to outperform an 8400 6-core from Intel at the same pricepoint, which cannot OC and has like an all core boost of around 3.8GHz. An all core OC of 4.0-4.1GHz on the AMD, even with the IPC being slightly lower (around 4%), still results in more performance. But you have to OC the machine and get the 3000 to 3200MHz ram running stable.

 

If I would have known, at the time, of my build I would have put in a Vega 56 instead. I rarely ever use 3D graphics so the cost savings over the 1080TI would have been nice. As far as power I doubt over idle there would be much of a difference.

 

https://arstechnica.com/gadgets/201...ns-48-cores-and-12-channel-memory-per-socket/

Intel officially announces chiplets on 14nm and a giant 5903 pin socket...

Aaaaaaaand more dodgy benchmark claims. It's getting ridiculous that one cherry picked scenario and the weasel words "up to" are enough to claim total stupid like '17x faster'


Will we see the EMIB based interposer counter to IF?

 

Vaporware anyone?

Edit this is a laugh. 12 sticks of memory per core. Then the latencies added by it's interposer or IF equivelant etc. etc. etc. No firm date etc. etc. etc….

 

This had been rumored for awhile and is Intel's only counter to Rome, a 64C/128T chip from AMD.

From this description, they were talking 48 cores, which means 2x24 cores, all when previous rumors had 2x28 cores, which may have been too hot or they needed too many for the unlocked 28 core CPU.

Either way, it is rumored to be bga, unless they changed that recently to a socket.

As to IMC, Intel is doing NUMA on single socket just as AMD is moving away from that with the uncore chip.

Either way, except for 12 channel memory, lower core count, and no mention of SMT/HT, everything confirms prior leaks on the topic. I wouldn't doubt this was leaked to combat tomorrow's AMD horizon event, where Rome details will be given.

https://www.anandtech.com/show/13535/intel-goes-for-48cores-cascade-ap

Sent from my SM-G900P using Tapatalk

 

Again, since there is absolutely 0 (NO) hardware it is vaporware. Leaks about possible silicon means nothing to me. Show me the hardware, where as Rome hardware does exist. NUMA or no, the interposer will add some latencies (extra pathways if nothing else).

And of course they are scared to death of Rome.

 

Aha, there's an AMD event that starts tomorrow/today that's why we saw this vapourware announcement from Spintel.

 

Intel won't let AMD have a nice day.

 

so before i can get 4ghz 32cores with chiller now i can do same cooling solution while double my cores at 4ghz no problem. thx amd.

intel only really got the upper hand now when it comes to avx2 and 512, i could still see them somewhat struggle in this area.

 

dang son

Spoiler

LOL

Spoiler

 

https://www.phoronix.com/scan.php?page=news_item&px=AMD-znver2-GCC-Patch

Sent from my SM-G900P using Tapatalk

 

thats nice and all but i just dont think it'll matter that much tbh, at least not for consumers. no doubt enterprise software might be using those. just look at avx has been with intel since forever and their dominance for last decade, we only have a handful of avx software its just sad, though no doubt enterprise will be using avx512 going forward, or they might not, if people smarten up and just use AMD instead.

spend less for more cores and beat intel cpu even when they are running avx512. amd can just go for avx and still do a decent job.

 

So, with the doubling of the floating point bus to 256-bit, it will likely greatly increase AVX workloads on AMD. Won't help with 512, but it is something.

Sent from my SM-G900P using Tapatalk

 

just a bit concerned about uncore being a separate die & before we got extremely low latency within ccx much better than intels, while during 2 ccx or 2 die its very slow due to IF's issue and ram speed. now with it completely being a die of it's own i can see latency go up, but if they can somehow improve on IF.. it'll be interesting to see how it turns out.

they did say with this design it is better ipc/frequency improvement than expected however iirc originally 7nm was suppose to be 1.35x performance now became 1.25x performance show TSMC's 7nm did have some of its own issues.

then finally theres the consumer side of zen2, wonder how likely it is they will offer us a "single ccx" die with all of it's uncore included in it, or would consumer be 2 chips made of I/O die and a core die. this will be interesting to see.

 

1.35x was from when the difference was between 14nm and 7nm performance, and now we have the difference between 12nm and 7nm performance as 1.25x?

 

14nm and 12nm was pretty much the samething tbh. it boosted the frequency by like 100mhz on average chips and out of 3800mhz in total thats like 2.5%. 1.35 down to 1.25 thats closer to 10%, there are issues for sure. even with the issue the .5x power envelope is sitll a big win.

i just hope intel doesnt go to TSMC for 7nm lol

 

That would equal the death of Intel fabs.

In any case, have to do some errands and on phone. So later today I will hop on to discuss my thoughts on the uncore as implemented, IF (the IF controller is on the uncore chip, meaning speed may be divorced from memory), arrangement of dies, the performance numbers, etc.

Also, I'll bring up cache talk a little, even though my knowledge on cache is not where it should be, the changes here could be quite important.

I also feel they overstate the impact of I/O double bandwidth and made a mistake by not doubling memory bandwidth through doubling channels, instead increasing supported speed to 3200 which gives an approximate 33% bandwidth improvement, likely due, in part, to equalized memory latency to each die, along with frequency increase.

I'll come to give more details to what I gleaned later. Here is AMDs video of the event:


Sent from my SM-G900P using Tapatalk

 

given that full working 8 core die will be used on Rome, we likely wont get a single 8 core die for consumer, not to mention it is made without imc/cache. ryzen 3700x w/e chip will be 2 core dies with an I/O die in center, up to 8 cores. for 16 cores.. 4 die or 2 die still unsure.

they probably willing to UP the latency and Lower latency for 1st/2nd die compare to what it was before, 1st and 2nd ccx.

ryzen 2 intra core latency was less than intels by about a good 8-10%, within a single ccx vs intel's uncore is also a good 8-10% so overall if a single die with it's I/O we expect to have possibly better IPC than intel. im hoping for something along the line of similar latency for all the core die yet come close to intel in terms of IPC

 

Next Horizon - Epyc is EPYC!
AdoredTV
Published on Nov 8, 2018
It's a bad week to be an Intel fanboy.
http://forum.notebookreview.com/thr...ga-polaris-gpus.799348/page-521#post-10818056

 

if anyone from AMD is listening, we want AMD in laptops. 8Core Zen2 v+ 2070 in XPS 15 body would be killer

 

I agree with Nvidia GPU being used, but stop with the worrying on 8 core CPU. Think about it, if they are talking half the power, they could get the performance of like the 2700X in a TDP envelope of 48W. Cannot guarantee the 25% performance uplift at that point, as AMD showed at next horizon 25% performance at isopower, which is the 95W TDP. So instead, shouldn't you yell at OEMs and ODMs to make laptops with it?

Sent from my SM-G900P using Tapatalk

 

Intel i9-9980XE Review: Disappointing Overclocker, But Good Stock
Gamers Nexus
Published on Nov 13, 2018
The Intel i9-9980XE didn't perform exactly as we expected. Our review of the CPU looks at stock and overclocking performance, including extensive thermals. Article: https://www.gamersnexus.net/hwreviews...
This review benchmarks the Intel i9-9980XE vs. AMD Threadripper 2990WX, Intel i9-7980XE, 9900K, and more. Testing includes a focus on overclocking and thermals, with additional testing for power consumption, Premiere rendering, Handbrake transcoding, Photoshop, gaming tests, and more.
http://forum.notebookreview.com/thr...-x299-xeon-1p-2p.804776/page-27#post-10819820

Intel i9- 9980XE Review - ANOTHER Skylake Refresh?
HardwareCanucks
Published on Nov 13, 2018
The Intel i9-9980XE is an expensive processor that's supposed to perform well in benchmarks against the AMD Threadripper 2 2950X. But at $2000 its VERY hard to recommend.
http://forum.notebookreview.com/thr...-x299-xeon-1p-2p.804776/page-27#post-10819820

 

It's been done, and unless something's changed in the last 10 months or so, @XMG says the ball is fully in AMD's court, and they're making no move to help support any Ryzen platform in a laptop.

http://forum.notebookreview.com/threads/clevo-ryzen-possible.801836/page-2#post-10469075

http://forum.notebookreview.com/threads/clevo-ryzen-possible.801836/page-49#post-10647375

 

Presently at 14nm and 12nm the chips do not really seem to have a great mobile equivalent. We have to wait and see where 7nm leads us and even if it is feasible being as they are using the 14 nm (or possibly the 12nm) I/O chip?