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

Unfortunately, I couldn't find a clean spec sheet on the Gigabyte or Asus boards. We have info from back at Computex and a couple new screen shots, but not much more. Considering pre-sale on the 27th, pricing will be known soon.

@hmscott , @TANWare , @Papusan - Did you notice the supported speeds on those Asrock boards? 3600MT/s, which means they have QVL for that speed most likely. They didn't say if that was a single dimm or full quad channel, though. But wanted to point it out.

 

Hopefully quad channel. I will get 4 3600 dimms and hope for the best either now or down the line.

 

I'd expect nothing less! I'm going to re-purpose my 4133 in there and hope to get 3600-3733 at absurd timings (like CL 14@3733 which is what they currently run in dual channel with both channels populated).

MSI gets quite a few ram overclocking top scores, but so does Asrock. Asus is usually in the top 2 or 3 boards for overclocking. Gigabyte is a strong performer. But, until I see some reviews, I cannot jump, as usually one board is stronger in one area while another beats it in a second area. You have to look at whether you want the strongest memory performance, the VRM and heat performance, the strongest OC, BIOS/UEFI ease of use and features, if water cooling, whether there will be a compatible block (especially if buying a full coverage block for the board, and when it will be available), etc.

 

From a quick look over thw MSI Gaming looked like the nicest one.

 

MSI looks good, but I'm waiting to see performance. On Ryzen 7, I don't think they had bclk.

Sent from my SM-G900P using Tapatalk

 

See:
http://www.tomshardware.com/reviews/intel-core-i7-7820x-skylake-x,5127-9.html

While the i7-7820x Skylake platform may not offer the best overall performance (I'm ignoring gaming 'scores' as they're irrelevant to me), what I get from the above article is that even with the optimizations Ryzen has seen in the last few months - Intel is still the overall productivity choice. The specific Intel processor you need to choose varies depending on your workloads, of course.

AMD is close; but productivity doesn't reward 'close' - it rewards 'best', absolute, period.

AMD may be firing on all cylinders vs. Intel right now with the latest Intel offerings seemingly 'a mess' as some describe.

I see past all that though and what is left is that AMD is only competitive in very specific scenarios (price being the biggest one and depending on the actual workload/workflow; high core/thread performance), overall.

What is eye-opening for me is that the idle power spec's of AMD's platform make it unsuitable for my usage style (I leave most of my systems on 24/7/365).

With a minimum of ~46% decrease in idle power vs. a Ryzen 1600X (the 1700X and 1800X are worse...), the i7-7820X is very efficient when just taking care of Windows background processes for me so that I can sit down at any time and use the system as I need without waiting for the O/S 'caretaking' to take place first.

And just in case you think that under-volting can make a difference;

The popular opinion is that Intel is scrambling - and it may be (time will tell) - but right now? The products it offers still match the workflows and workloads of the majority of the world's pc users.

Thanks again to AMD (and their fans) for pushing Intel a little more than what we had become accustomed to.

 

This does not mean to carry on in another thread. Things and plans change to meet the market, enough said.

 

FYI...recently made recap's of Intel's long sorted history of "fair play" with AMD...

Intel - Anti-Competitive, Anti-Consumer, Anti-Technology.

Are we MAD at Intel?!

http://forum.notebookreview.com/threads/intel’s-core-x-i9-and-i7-series-x299-xeon-1p-2p.804776/page-13#post-10572105

 

AMD R3 1300X Review vs. 7350K & More | Intel's Response

 

Been posted 4 times, other than my attempt to point to a specific time in the video. It definitely leaves room for more cores on the future. They've also found out the socket is the exact same socket as EPYC. No mod on pins or connections. That means that it is only the MB artificially creating the quad channel memory and the fact only 64 lanes were laid that separate the lines.

To date, AMD said they will not unlock an EPYC 1P CPU, but there are plans to unlock future feature sets. Now, imagine they unlocked one 1P chip that is 16/24/32 cores and superO made a gaming board for it, where you could use riser cables for the pcie, etc. 128 lanes. Especially since Nvidia is following suit, doing an mcm design, which means they might remove the bs artificial cap on two cards for Nvidia SLI. Dreams...

Sent from my SM-G900P using Tapatalk

 

I'd like to see that running at 5.0GHz. And, I wouldn't even care how high the TDP would need to be. 500-600W, or even 750W, power draw for only the CPU would be just fine with me.

 

thats gonna be quite hard to cool lmao.

that aside, silicon lottery on skylake-x chips looking good, 4.8-5ghz 6-10c definitely doable, too bad they dont come in a laptop. its quite sad clevo didn't agree with what we and eurocom wanted, truly pathetic

 

Could probably top out at 4.0 or 4.1 same with the Ryzen 7 1700(x)/1800(x) on air. As for the 5GHz, it seems they are targeting this when they shift to 7nm see https://www.globalfoundries.com/sites/default/files/product-briefs/7lp-product-brief.pdf page 2

 

Also 5-core CCXs - 10-core 5GHz in 2 years? ^_^

 

I have tried and there is no way I can get excited about a CPU that maxes out at such a pathetic clock speed in spite of the other things one might find a reason to get excited about. That totally turns me off. Owning a CPU that runs the same for everyone else that owns the same CPU would be about as stimulating as watching paint dry.

 

seems unlikely for a first gen 7nm to hit 5ghz right off the bat, even the best of the best sandybridge 32nm was 2nd revision of 32nm and skylakex 14nm too. well people can still be wishful and hope for the best.

also wtf threadripper this better come soldered.


update: looks like its soldered good stuff

 

Typical OCer's tune. I used to be the same till I figured out that it is fast enough to get my stuff done. Like I said if it does it in a blink of an eye who cares when it is finished while my eye is still closed so long as it is done before it is opened.

 

heh for your usage maybe. i cramp my comp with stuff loaded when i use them its gotta be fast fast alright

 

Not in a laptop. Even with an AC blowing, you are not really getting that in laptops. Delid helps, which is the only way to get 4.8-5.0: binning, delid, liquid metal, and most times a custom water loop or better on cooling.

I care less about speed than performance. I still need to see scores for the max OC, but we've seen where the 6-cores wind up needing 4.7GHz on Intel to keep up to a 4.0GHz AMD chip in some tasks. So it isn't about the speed to me, but the final scores.

 

That does not win any contests. There is no such thing as "good enough" and having a belly button CPU is pointless if your primary reason for having a passion for PC is competing against other with the same interest. If it does not respond well to being pushed out of spec, then it cannot be taken seriously as an enthusiast product. Same deadly curse has haunted AMD on the GPU side of things for a long time. Fine for consumers and bubblegum gamers, but sucks for enthusiasts.

 

I have to see both or I am just not interested. And, the "some tasks" thing isn't very compelling. It leaves too much room for excuses and limits potential too much. I really hate the "yeah, but" approach to justifying anything.

 

In about 50% of games (they tested 30 games) it was faster. I don't want to rehash too much of this topic, but the point was that, depending on the software used by the end user, it was a push. Single threaded and unoptimized goes to Intel. Multi thread and optimized went to Ryzen. So, it isn't a yeah but, it is literally a coin flip.

Sent from my SM-G900P using Tapatalk

 

im sure with a heatsink as big as p570wm and a capable chip like skylake-x with delid will get there for 6-8c. 5ghz 6c and 4.8ghz 8c perhaps. but since there are no laptops out there that can hold these cpu, its all ghost talk, sucks

 

Anything is possible once they get their act together regarding Heatsink fit and finish. Most stock 870 DM2/3s go nuclear at stock clocks without the enthusiast touch from the end user or vendor. This niche group of enthusiasts is all for the taking for Clevo if they can pull their head out of their butt. They need to learn to walk again before they can run.

 

It's annoying because all it really takes to get a lot of them down to reasonable temps is a repaste from whatever they ship with to pretty much anything else.

 

True that Brother.. I better not say where my second hand t̶h̶e̶r̶m̶a̶l̶ ̶r̶e̶a̶c̶t̶o̶r̶ DM1 was originally purchased from then

 

Please stop talking about Clevo manufacturing failures, it has nothing to do with the topic of this thread, thanks.

 

It had to do with @ole!!! wanting X299 in a Clevo, so don't be a wet blanket thanks

 

AMD EPYC vs Intel Xeon Scalable 2S Architecture Ultimate Deep Dive

 

I am not sure if this was posted already but get out your salt shakers. http://techreport.com/news/32310/rumor-mysterious-slide-lists-full-intel-core-i9-specifications

If true, the war has only begun...

 

I can see the TB 3.0 as it is limited to 2 cores but TB 2.0 seems a bit much if for all cores. I think the turbo/OC without a delid on all cores will be 1-1.2 GHz over base. Still this will be admirable for those CPU's against the TR

 

shake shake!

tb 3.0 is the way to go! i'll be testing that as soon as i can get one, with E-ATX. my case S24T portable is already waiting lol

 

For now TB 3.0 will work great for allot of games and single/dual thread apps. As apps start to use all those cores TB 3.0 usefullness will diminish, but that is for much later on.

 

reflect macrium is one of them. uses multi core but for different tasks and stages of the backup/restore its pathetic. crank that ST to 5ghz and watch it fly.

 

It should be noted that they adopted the IBM process for 7nm, which IBM has already produced 5-5.5GHz processors. Further, that is what the goal for HPC systems is (servers, etc.). When you look at the current 14nm literature, you can see the goal was 3GHz, which they hit. Now, if the server side gets that, we may see 5GHz+, but that needs tempered with Papermaster's comments that we will not see drastic increases in speeds other than what is currently on the market, which is why software designers must start moving toward implementing better multithreading in their programs.

"The technology is key for 'an era of Moore’s law-plus where we’re getting new density advantages at each node and cost advantages as each new node matures, but mask costs are going up and chip frequencies are not going up, so how we put solutions together is critical to sustain the pace of development,' he said." http://www.guru3d.com/news-story/amd-cto-talks-about-moving-to-7nm.html

"In software, 'my call to action for the EDA community…is to redouble their efforts to take advantage of more CPU cores and parallelism…As the processing required for 7nm escalates…their algorithm optimization needs to take advantage of the very technology they are helping us manufacture,' he said, noting AMD’s new Epyc processors sports 32 dual-threaded cores." Id.

https://www.techspot.com/review/1450-core-i7-vs-ryzen-5-hexa-core/


Just a little to chew on considering my last comment on the 4.7GHz vs 4.0 being a push. It should be noted most benchmarks still favor single threaded IPC and some are not suited for testing MT tasks well (while others are). Because of this, Intel may hold the crown, but it is a case of software testing for specific tasks that the industry is currently evolving away from. We've been through the arguments on when to dive in, including a recommendation for many to wait until the real battle, which is Zen 2 vs Ice Lake (7nm vs 10nm+). So, this is why I'm slightly less concerned. If the HCC wind up too close to 4GHz from Intel (assuming that TR cannot go past 4GHz avg.), we could see a situation where there isn't enough at the single thread performance and not enough MT scaling to get a recommendation to get Intel, especially at a 70% cost premium.

 

Yes, I agree. It shows us that, and it reinforces the long standing concept that clock is king with a 4C/8T CPU (7700K) opening a can of whoop-ass on both hexacore processors. It's great for AMD being the underdog to not get totally destroyed by Intel like they used to. That being said, it also reinforces my opinion that it sucks that R5 1600 can only manage a 4.0GHz overclock and the biggest thing we see from this video is how badly Intel dropped the ball and gave us a dog CPU in the 7800X. Anyone that already has a 7700K shouldn't waste their money on either CPU. At the end of the day, not much to write home about in this comparison.

 

Well, you use what data you have. The larger point is that, after keeping temps under control, you see 4.5-4.7GHz usually on the 7900X without delid. After looking at that, expected OC when core counts go up, etc., we're seeing more that, comparing on core count, Intel's Skylake-X chips are not as good as they are made out to be. Who cares if you can get to mid or high 4.x GHz on a higher core count chip if the equivalent core count chip from the competitor beats it in multi-threaded tasks (the main purpose for the higher core count) while still stacking up well in many single threaded processes while clocked 20% slower, almost. That is my larger point.

Sent from my SM-G900P using Tapatalk

Edit: also, clock is king because of programming not keeping up with tech and Intel's artificial limitation on quad core for mainstream, as was seen with dual core to quad transition.

 

Yeah, I get all that but it's still super lame and it is only useful making excuses for it to be OK that Ryzen sucks at overclocking. It makes it a great consumer or business CPU and a crappy enthusiast CPU. And, I totally agree that the Skylake-X chips are junk, but that's beside the point. They are not a good option and I think they are probably junk because Intel miscalculated (underestimated) what they were up against with Ryzen. It looks to me like Intel rushed a half-baked product to market instead of kicking back, letting AMD enjoy a brief run in the sun, then releasing something that destroys it on purpose. A delayed but calculated attack with something much better would have been a lot more Intel-igent (pun intended). All of this is good in one way. I'm glad AMD isn't going down in history as a chronic producer of anemic trash CPUs. However, I remain disappointed in both of them and not going to make excuses for either brand. Both hexacore CPUs should have utterly annihilated 7700K and matched or beat it in clock speed. They should have both beaten 7700K even where thread count is not important. I think it sucks that neither one did, LOL. If Intel, AMD and NVIDIA would just knock it off with trying to meter and measure and give us only enough to get the job done and give us their best every time that would be better for everyone. If AMD is already doing that, shame on Intel because I do not believe they are doing that.

 

I find it amazing. Everyone agrees Intel for years has participated in anti trust behavior for years. The have kept AMD out of other OEM systems when they were a better CPU, they engaged in having programing benchmark code made to favor their own CPU even though it was slower to look good. They have participated in having code optimized specifically for their CPU's. The have tried to start proprietary standards to lock out the competition ( IA-64).

So we still fall into the trap of not advancing Moore's law. Even Intel has realized the future lies in multi core and multi thread. It no longer impresses me trying to soup up the CPU that drives yesterdays software. I am more impressed by the CPU's of the future for Moore's law, no matter who makes it.

In the end we need to fall out of that OC mentality as it is too singularly dimensional. Yes awards are nice, no one will dispute that however the reward of having sponsored better general computing to me far out weights that.

 

No way. I want require both. As I have said, and will continue to say, if I cannot use a product to differentiate what I have from the masses, there is no sport. Having a belly-button PC that runs exactly the same as every other PC with the same specs is just boring. If I can't overclock it, or only by a little bit, that product is ordinary and offers zero value to me. I can already have boring right now with a tablet or smartphone, or a BGA turdbook. I expect an enthusiast product to actually be that. It cannot be an enthusiast product if it doesn't overclock well.

 

I think the problem here is Intel is somewhat giving up on OC being the answer to beating the Ryzen crush. Although still no excuse for using tooth paste. I do not think this is an end to the OC era entirely but SkyLake-X is a definite boulder in the road. The problem is every process, be it 14nm or 7nm etc., hits a clock wall that only LN2 breaks to find yet another wall. With a good fabric they can keep on stitching in new cores with no foreseen wall yet.

 

You hit the nail directly on the head. Intel is taking away your sport, we are just the messenger and unfortunately we somewhat agree with the methodology for Moore's law sake. You need to go to Intel about your requirement for extreme OC'ng, by the way good luck with that and I truly mean that as I think it can be fun too.

 

I don't think so. Intel has made a lot of money and gained their lead over AMD because of the overclocking sport. I really do believe what happened is they got too lazy at the top and AMD caught them with their pants down. What could happen is that they will come back with something that casts AMD in a negative light again as selling junk like the pre-Ryzen era. I hope not, but that's how they roll. AMD doesn't have enough money for that, but hopefully Ryzen's success will fix their cash flow issues and allow them to produce some scary and deadly WMDs. All they've really succeeded in doing as of now is leveling the playing field on performance with a cheaper product. It can't end there or they'll be killed by Big Blue.

It needs to be a ton of everything. Core, threads, IPC and insane clock speeds. They can do that. I believe Intel (and NVIDIA) hold back some crazy stuff so they can meter it out and make more money. I believe AMD pulled out all the stops in 2017 and they may not have a way to compete with Intel if Intel launches a destroy campaign. That would be bad for everyone. What I want to see is a long drawn-out bloody battle that lasts years with them trying to outdo one another and non-stop bickering and feuding among enthusiasts about which brand is best. Like the good old days when AMD ATI made graphics cards that were as good or better than NVIDIA. Those were great times. I love that kind of fighting among friends. It was awesome and we need to get back to the guerrilla warfare.

 

I'll start at the bottom and move up. Intel has been slowly taking away the fun of it, using toothpaste that requires a delid, etc. Look how little actual difference between generations has made recently. Then you had the Kaby spikes and Intel's response was don't overclock the CPU you paid a premium for it to be overclockable.

As to the belly button comment, what do you think every unlocked CPU is. Everyone with that chip gets about the same, with the silicon lottery playing a role. Now, a better OCer can squeeze a lot out of a lemon, but it will still be a lemon. You get the same thing on Ryzen. They don't all get the same. The best get 4.2, almost all hit 3.9, which means you have the same exact range you get on Intel on the silicon lottery, just it is clocked lower. Now if with the IPC advantage AND the speed advantage at certain tasks only gives you a 50/50 on which CPU wins, then it is coming down to programming and optimizations.

This brings me to @TANWare 's point about Intel trying to hand out proprietary compilers to purposefully use a WAY slower instruction set if AMD was found instead of Intel. We see this in a lot of programs today, where the FX series is now able to beat SandyBridge handily all due to coding for better multithreading because of Ryzen. That means that the FX was ALWAYS a better chip, and that programmers were doing this BS to make single-thread programs or only optimizing for Intel on purpose, thereby holding us to quad core just like we saw with quad cores.

And that brings me to your original statement that the hexacore chips should have annihilated the quad core chip. They very well would if it were not for the code not being better at handling MT instructions. It's why we have seen the wall on games at six cores for awhile, with little scaling above that. On the speed, that is the laws of thermodynamics: more cores = more heat (although they now have shown lasers that can remove heat, which may be useful if they can figure out a process for that as we transition to light based information transmission over classical mediums).

Now, AMD designed the new chips by scaling back the R&D engineers, but focusing on talent. They had to find the best way to get into the game again without blowing the budget and going belly-up next year in bankruptcy. I feel they already accomplished this. Now, after done fighting for survival, you see them working on a better process than Intel's which may beat logic density of Cannonlake (Ice lake may be denser because of refinements in the process, so leaving it open; under traditional calculation on the density). Intel has been trying to go elsewhere for extra revenue due to lagging computer sales. Because of this, they have bled money all over the IoT market, all while taking their eye off of the ball.

We'll see where it goes...

 

My firm belief is that when AMD first was making TR and Epyc they were using Stepping 1 of the CCX's. I think there were major issues and it was not even as effective as Intel's ring fabric they were using. In fact it was so bad that Intel never even planned on over a 12 core to combat it. Epyc's were probably even worse so Xeon's were posed no threat at all but eventually they would be getting updates as well.

Stepping 2 though came in and the Epyc started to pose a real threat, and subsequently TR as well. This caused Intel to use a redesign of their fabric for better response above 12 cores and the cache changes as well. The rest is now current history unfolding.

 

Sort of. There are K and X CPUs, where you have to pay to play. I'm actually OK with that. Pay more, get more/pay less, get less, is a good business model.

What I am not OK with is a normalization, where you pay the same and get the same as the guy next door, and that's all there is to it because it's won't do more if you try to force the issue. Or, it does so little more that it's not worth the bother.

Remember 7970M versus 680M. AMD fanboys were happy to have an NVIDIA GPU killer. And, they did for a few minutes. Then once the overclocking and vBIOS mods started, 680M tore 7970M a new heiny hole and that was the end of it because 7970M sucked at overclocking. Then they started dropping like flies because of people trying to catch up with 680M and they weren't build to handle it.

 

Intel doesn't have anything up their sleeve. That is why the 14-18 cores came from pulling from the Xeon lineup, then having to move around FAB time to try to get them out quick enough that they weren't irrelevant, while shuffling coffeelake, all while cannonlake still isn't performing, meaning it is pushed even further into 2018, which should have been released last year, then when Kaby was. They are in panic mode because 10nm isn't what they hoped. If TSMC and GloFo beat them and have that process for years, with TSMC moving to 5nm around the time Intel makes it to 7nm, the pressure stays on and performance is there.

Not only that, AMD kicked up R&D spending this last quarter and are working desperately to succeed on 7nm, where the true war is fought against Intel. They needed cash flow to do it, and Ryzen is that cash flow (as Vega is...). Intel doesn't want their bottom line to hurt, but doesn't want to kill AMD because if they do, they can be broken up! Also, on SMT, Intel's HT doesn't come close!

Intel cannot really respond until Ice Lake, and considering IBM's process is used and the IPC on their products, that may just shift it all. But it still comes down to software developers getting on board! Look at Mafia III, the only game with "optimizations" for Ryzen where the Ryzen scores WENT DOWN while Intel's went up. Or look at the poo-stain Gears of War 4, which also shows Intel with a huge lead (which is also born out in the productivity suite Office, where a quad core is king still, which is a joke)! So, until those issues are addressed and software developers get off their butts, you'll see this crap with them saying Intel's better when it may not be.

 

Well, yes and no. If we measure the results and everything is tilted in favor of Intel on the software side and Intel continues to beat AMD, guess what... it's better based on results. Good intentions are nice and admirable, but the only thing that really matters at the end of the day is results. I don't agree that a CPU is actually better unless it produces better results. The reasons why don't matter if the results don't show it is the better product. It's good to teach kids that how you play the game matters. Integrity is important. But, if you play the game right you might lose. Losing always sucks. Doing the right thing will always be important and we need to sleep at night, but losing always sucks. We have to accept that sometimes things are just going to not end the way we want them to, and then go find something else to do.

 

This is backed up by them also discussing that features are still not unlocked on Epyc, but are built in (meaning more to come, but is like the memory AGESA issue).

Here's the thing, that is false even with Ryzen. Of the 8-core chips, the range on air/water go from scoring high 1500s in cinebench up to 1950 in cinebench 15. It isn't normalized when you get a jump of around 350 points between your stock and overclock, where many fall down in the early to mid 1700s, then having another bit in the 1800s. You have to get your ram setup properly, etc.

The way you say it here, it is like there is no difference between the guy who overclocks and doesn't, when every single MHz actually does help here.

 

Under that rationale, you can justify all sorts of messed up things and even force a regression of progressive (like Intel's been doing). This is why AMD has been working on the new APIs for years, Vulkan now becoming openCL and Vulkan aspects borrowed for use in DX12. They are sending out units to developers all over. Even with that and skyping to teach them how to use the architecture, etc., you can only do so much, especially as some would need to do a teardown or build a new engine to actually use Ryzen or Intel's HCC chips better. That means you will see a mix on implementation. But, my larger point is that sometimes you give to the underdog if it is the only way to get a good dogfight going. To call the small one dead before the fight is to literally just take it out back and put a bullet in it. We both want a show. There is one way to get it.

As to the research advantage, at 4%-10% improvement per year, they are out of runway. They hit the shrink wall and everyone is lapping them. All they have left is the entire server market, which is a huge amount of their income (the server market is $16B; the enterprise client PCs come next, and that is being challenged with the Ryzen Pro series very well, especially since Intel i3 cannot do security on vPro, whereas AMD can do the equivalent with dash). So the fight hasn't even begun for the large marketshare. HEDT gives notoriety. Records give notoriety and status. But HEDT is only a high margin, small market.