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

I do not think anyone will be upset with the first 300w CPU. I can almost guaranty even at 4.0GHz that 28 core Intel is well better than 300 watt. Even at 205 base the 8180 for 2.5 GHz all core is 205w TDP. I think so long as there are acceptable cooling solutions that can be had then even 400w is fine.

 

We are already touching 250W using 8700k's @ 5.2-5.4Ghz

Im sure these 32/28 core chips can pull more!

 

The point is what is acceptable for stock speeds for power draw. I can see the point of wanting to stay at 250 on the TDP. Mainly as once you get too the 300 or 320w mark you are discounting on using just air to cool it.

As it is mine can see 320w on the package. Even with the Enermax 360 it is a challenge too cool. As it is my system seems to handle 320w on the package without issue but I think with Prime95 it could be a sever challenge for 350 w or 400w. Below is a 15 minute P95 run with my system @ 4.0 GHz, vcore 1.375 memory 3200 MHz.

 

I do have to add one caveat. The present TR generates its heat with only two contacts to the IHS. The TR2 supposedly uses the same die size contact to the IHS but now there are four contact patches of the four CCX modules. This for the same overall TDP should make heat transfer to the IHS much more efficient and thereby more efficient to remove. It will not scale 100% but I will bet on a 50% gain of efficiency.

Even if it were as low as 25% it means where the overall TDP of 200w can be handled by the cooler on a TR1 250w can be handled as overall for the TR2. This could be why they have selected 250w as a target as the heat transfer of the CCX configuration removes that as well as it does 180w on the TR1. The 320 of my setup might actually handle 385w of a TR2. Exciting days ahead.

 

I should also note where with the Enermax 360 they advertise it can handle 500w. In the TR1 configuration I can not see that being the case. Under the TR2 configuration it just may be more realistic.

 

maybe your thermal paste job / case ventilation / fans / ambient temp / whole setup just sucks

(just trolling you a bit here jealous of your specs )

 

I did not win the silicon lottery but I have 4.0GHz so I can not complain too loudly.

 

I5 8400 vs R7 2700 vs R5 2600X | Tested 15 Games
For Gamers
Published on Jun 9, 2018

Ryzen 1600X vs Ryzen 2700X vs Intel 8400 | Tested 13 Games
For Gamers
Published on May 14, 2018

GTX 1050Ti | (STOCK/OC) R5 2400G vs I5 8400 | Comparison
For Gamers
Published on Feb 16, 2018

25 Games Tested | Ryzen 5 2400G vs Intel I5 8400 | Comparison
For Gamers
Published on Feb 11, 2018

 

totally forgot about that, good observation.

also what kind of cooling you're using right now on TR1?

 

Intel's "5GHz" Computex ConJob
AdoredTV
Published on Jun 8, 2018
Con-tinuous BS from Intel's marketing department, this time at Computex.

Intel 'Forgot' to Say 28 Core Was Overclocked
CPU coverage ends at 15:45, after that it's about GPU's...
RedGamingTech
Published on Jun 8, 2018
Let's discuss the elephant in the room next, as some more reports have come out on Intel's 28 Core "5ghz" CPU. Apparently, during their conference at Computex 2018, Intel "forgot" to mention that the monster 28 core processor was overclocked. We have also learned a bit more about it, such as what cooling it was using to run at that clock speed...

Spoiler: AMD / Nvidia Details...

In a recent interview, AMD's David Wang has commited the company to annual GPU releases. We have seen the graphics card market stagnate somewhat over the last few years, but AMD wil be giving us a brand new product every year. This will undoubtedly mean refreshes for the upcoming architectures, but we can probably expect some new processes as well.

Finally, there is an annoying bug for Nvidia Pascal and Maxwell Geforce graphics cards owners, as there is a display port bug which unfortunately needs a firmware (AKA BIOS) update to fix. Given that the issue causes slow downs and black screens, it's a bit annoying you have to pull out a firmware fix to resolve it.

 

Enermax 360, got one even before I had the CPU. As far as TIM I just installed ThermalGrizzly. I had an LM install but it failed. The LM lost its bind ot both IHS and heat plate and when I pulled it apart it was just small beads. First time I have ever had LM fail.

I think I will be ok with 400w for heat the VRM's may not bee too happy, they get pretty warm with just the 320 package wattage now.

Edit; another point of fact. The original TR was released with all those AIO adapter rings. So there are quite a few of them out there with AIO's that do not cover the entire IHS. Since AMD has to make this a drop in replacement for TR1 they have to keep the TDP down.

 

My daughter turned on the air in the house, ambient 72f. Same 15 minute run of P95

 

OK, so how did you change your VID because even though I am running way lower CPU VDD, my VID is almost 200mV higher, which is why my CPU package is pulling almost 100W higher than yours.

 

Your vcore is apparently 1.550, where as mine is at 1.375. This for 4.0 GHz here. As mentioned I did not win the silicon lottery, it seems you did win though as even though your package is at 392w your temps are way cooler.

I have to ask what cooler for the CPU and VRM's?

 

Nope, VCore was 1.175. I just checked by going default and using Ryzenmaster to set the VCore. It changed the VID reported in software, but the VCore reported in the software was the same (they have the VCore generally reported elsewhere). Temps were exactly the same (Within 1C and the ambient was different by around 1F, so practically the same). When using HWInfo64 (my preferred), it shows the current as lower as well (in the 200s), while the package current is about the same because of using the VID to calculate the current rather than the Vcore. Now that I have that settled, unless it is reporting some other voltage, it is a software error of some sort.

 

My VDD node 1 and beyond mirror my Bios vcore. Is the hmonitor shot after 15 minutes of P95 also, I thought it was and your VRM is super cool is why I ask.

My VDD and VID mirror one another that is.

 

I'll run gooseberry with both hwinfo and hwmonitor tomorrow in both configurations and show you. Also, VDD node 0 is your SOC voltage.

Sent from my SM-G900P using Tapatalk

 

I can see your two values are different, mine are not. I am not sure what does this but it is this way right from bios default settings.

 

320W package!!! lolz love it so we can expect like 450W from.TR2 when overclocked? haha

over here i can barely screech by and avoid thermal throttling by a hair when running p95 avx at 120W+ on the cpu package... which of course is not too shabby for a 15 inch laptop i guess

Sent from my Xiaomi Mi Max 2 (Oxygen) using Tapatalk

 

Assuming we have the VRM's and cooling solution I could see someone taking a TR2 up to 600 Watts on the package if it overclocks as well as the 2700x.

The math for CBR15 as 3500 @4.GHz TR1 would be 3500*(4/4.35)*2 or 7613.

 

Evidently not with mine. Here are two images. I couldn't get the ram stable at normal clocks again, but got them close enough and roughly similar voltages. Aside from that, the one with the 1.55V VID is the one with my custom fan and pump profiles which favor sound a little more than stock on those.
BIOS

RyzenMaster


I do not know the cause, but as you can see, they are the same, just that different values being incorrectly reported to the software somewhere. Notice how low my CPU wattage is (203 to 208 max on CPU Core Power in HWInfo64 @3.95). Also inconsistencies on HEDT being enabled, etc. But, they are roughly in line.

Edit: Also notice the 227W on CPU Package @3.95. Now that is a nice chip! All cores, 1.181V (which is can do 1.175, but I wanted to do them the same and the Vdroop hits harder for my RyzenMaster OC and showed slight instability).

 

Now that we have the software issue figured out and better figures to work with on talking about the silicon, my 4GHz is around 1.225 or so, and my 4050 is 1.2875V stable with gooseberry, sandra sisoft, V-ray and Corona loops, and 1hr on the newer realbench. My block on the CPU is the XSPC TR4 block, and I'm using Koolance universal VRM blocks for both sets of VRM (hence the 40C), with 3x480s on the radiators and two swiftech 50x pumps. And, I did buy this chip from silicon lottery, top bin BEFORE they had the 4GHz binning (was one of the purchasers on day one and got one of the best chips at the 3.95 binning). So, some good stuff in this system.

 

man thatd be super sweet if TR2 also managed to do 4.35....

 

Only the 8-16 core variants, if released, MIGHT be able to hit that, like the Ryzen 2000 counterparts. The 32-core variant, that is a pipe dream. Expect roughly 4.2 for the average on the 16-core chips, just like the 8-core 2700X.

Edit: Also, notice how my 3.95GHz runs a max of 66C on Gooseberry in blender. Have you ever run that bench? I finish in 22-23 min.

 

I di not expect, but I could see the 32 core reaching 3.2 GHz on water. It would be a 480mm custom water loop (like the Intel that was on display) and the board needs to be a newer one with a HUGE VRM section.

 

They are working on the 32 core reaching 3.4GHz on air, but I would still recommend the 19 phase board by MSI or other boards forthcoming and a 480mm custom water loop to really deal with it. But, as you can see, the top bins for the TR1 16-core can reach 4GHz all core clock around the 250W package power for the TR2.

 

@4.0 GHz 1.375, 3200 memory, My bad, my router took a dive off the deep end and just got back on but I meant 4.2, just not 4.35.

Edit; I should note I got 19 minutes and change on the first run so I did it again just to be sure.

 

Definitely beats my scores. How tight are your memory timings at 3200?

 

16-18-18-18, the memory is supposedly 3467 but 3200 is 100% stable default some workers in P95 will stop eventually.

 

It comes out better than 13% faster so at this ram depth my IPC seems much deeper. There is one thing to check though to be sure. This is integer performance in Passmark. If it is low it is Windows fault, I just recently had to get mine back, here is mine, again pay attention to the integer score, if it is low it needs to be fixed.

Edit; When it is off it is like 33,000 only!

 

OK, what needs fixed in Windows, specifically? What are you talking about being off in Windows, as I am having the low Integer and floating point.

 

Ugh. This is a bad omen. No one yet has come up with what the problem is, M$ answer is to just reinstall Windows which does work, AT FIRST. Eventually though it WILL happen again. Constantly re-installing is not an option for me. So to my solution.

I use Macrium Reflect, the free version is fine but I am paid. IOMMU must be disabled. Start with a fresh windows install with Macrium and Passmark, once the performance is back create a back up. Do a few updates and program installations and make sure the performance remains good then create another backup. I did this for about a total of a dozen backups. if performance gets lost restore from the good latest backup and continue on again (mine were done incrementally so I had all backups just incase).

I keep my data all on a separate drive so there is not an issue there. a heads up here. This last time I apparently let it og to long and my 3 month old backup did not restore performance at first. I had to go to reset the system and all files. Then used the 3 month old Macrium image. I have all my user data backed up so that was not an issue.

Hope this helps.

Edit; I have seen a few posts elsewhere with Intel systems that have had this issue as well.

Edit 2; This upsets me to no end as the first level tech will inevitably have the user to reinstall Windows and since this fixes it the problem is never resolved. It is never escalated and just written off.

 

That is just All CAPS on M$ part.

 

Yeah job security. They will all tell you that you did something and just reinstall Windows. No I didn't and I can show where it just all of a sudden happens for no apparent reason at all. It is something on their end.

 

So, looks like I'll need to dig out my copy of macrium. I stopped using it because it is **** on doing multi-boot setups, or at least was not working well with my multiple OS installs for backing up the OSes (basically, unless you booted and installed the software on the OS, which you wouldn't want to necessarily do on bench OSes, you backup the drive as just a drive and not windows, even if you select the proper hard drive sections with the boot sectors to back it up). But, looks like I'll have to dig out all of my keys again and re-install EVERYTHING, after backing up the current setup. I'm wondering if that is how you cut 3 minutes off of the render on Gooseberry and a couple other oddities. How does your ram score look on performance test?

 

Well I use Macrium just for the C;\ drive, but I have a lot of drives. I got rid of using Grub by booting to the C:\ drive from bios each time, then F11 on reboot when I want Linux.

C:\ = Windows OS, 500GB EVO 960
D:\ = User Data, 500 GB EVO 960
F:\ = Macrium and user data backups, 3TB 7200 RPM HDD
l:\ = Spare 1TB SSD, SATA
120GB SSD for Linux, SATA



Edit; When I ran at 4.1 GHZ scores were higher.

 

@hmscott You talk and talk about that Intel screw their users with every new cpu release the year after(no compablity with cpu refresh on current socket/chipset). You have messed preached about this the last couple of years now. From your own mouth... Only AMD offer Cpu upgradeability.
This below is rumors and not officially but will you continue as usual, speak up loudly that Intel will screw the users who have Z370/X299 MB when the next processors and MB come this fall?
Intel’s Flagship Coffee Lake-S 8 Core CPUs Arriving in September on Z390 Refresh Platform – Skylake-X Here To Stay, Will Get 22 Core CPU Update This Year

 

Yeah, I need to someday get another couple SSDs to have the OS on each one. Currently, my SSD shares all my OSes and most programs except games, which are on one of the 8TB drives (2TB of games). But I definitely get doing it that way, which simplifies many things.

Here is my score for 3600 CL14 17-17-34-48 1T @3.95GHz on the CPU.


Also, do you have AIDA64? Wanted to compare the FP scores on that considering this software said my floating point was 1/6th of what it should be. I'm trying to rule out a driver or other software effecting this software specifically, although something must be up considering you also smashed my blender render by like 3 minutes.

 

Although you are speaking to him, and do have some points, your article doesn't stand for the proposition you think it does.

First, the Z390 or whatever from Intel changes to the Cannonlake-H chipset, from the Coffee chipset. This means, just like the release of the Haswell second chipset, the older boards will not support beyond this year's 8-core, which is basically just a type of refresh, but more, than what the 4790K was compared to the 4770K. But, as I said, that means that we could be looking at compatibility on this chipset with the 10nm chips for next year.

Now, it would be wrong to say Intel never allows forward compatibility, as every time they have stalled on die shrinks or new processes, they then have limited forward compatibility, like the with the 4790K and the broadwell series (which sucked), Skylake and Kaby on Z170 (because they couldn't get 10nm working for cannon), and then they popped out the z370 within 8 months of the 270 because that is the type of forward thinkers they are and they know by calling them roughly the same gen that consumers would revolt if having to get a new board AGAIN so quickly on the same line. But you do have a point, to a degree.

On the HEDT platform, they screwed consumers, and there is no two ways around this. The platform has always lasted for two years and was based on the promise of a two generation compatibility with the server industry as the socket was the same socket as the xeons. Here, that is only semi-true. Now, Intel said they were splitting the HEDT and server markets clearly with this gen last year, introducing the FPLGA 3647 pin monster for the server side. That would have been fine. But what we now run into is a further market segmentation with the news of a new "Ultra-Premium Desktop Market" for the new chips that will cost easily DOUBLE what AMD will likely charge for their 32-core variant, meaning they are dicking the HEDT out of getting the highest possible chips, as they thought they would be able to get, and making the UPDM (which ultra-premium is their way to appeal to the dipshit consumers that don't realize they were already dicked, and to appeal to nothing as all it is is a high end desktop) so cost prohibitive few will buy it, which is good considering how low the yields they have on monolithic dies. They SK-X refresh with 22 cores is to tell those that got the 18-core you can still upgrade, even though those on the X399 platform, so long as they have certain boards without crap VRM, can throw the 24 and 32 core variants in without a problem (my VRM stays around 40C while pulling 250W, meaning I could even push up to 300W easy on the chips). The X399 boards benefited from the Intel VRM debacle as they practically copied the X299 boards for X399 with some tweaks, then corrected some issues that were found in June on X299 by the time the X399 boards arrived in late July and throughout August. With that said, I am looking forward to the new 19 phase designs for both platforms.

But, you seem to gloss over a couple points in that article I would like to highlight:

"The reason Intel will stick to the X299 Skylake-X CPU family and not take an entire jump to Cascade Lake-X platform is due to price. The Cascade Lake-X platform is going to at least, twice as much expensive as the current Core-X processors and their accompanying platform."
"[F]irst is the chip itself which are basically repurposed Xeon Platinum and Xeon Gold parts. These chips cost north of $5000 US and the top variant comes in at $10,000 US (28 Core Xeon Platinum 8180). Considering the Intel Core i9-7980XE costs $2000 US for 18 cores, we can expect the 28 core model to be close to $5000-$6000 US easily. Even though there’s a 24 core part and more configurations on the way in the Cascade Lake-X, we expect the price to put a heavy toll on your wallets."
"Secondly, ... motherboard manufacturers are putting extra RnD in building the new LGA 3647 socketed motherboards. The ones Intel used as a demo on their Computex press conference stage was a prototype and not a final variant. The motherboard vendors we had a talk with revealed that they expect the launch of Cascade Lake-X motherboards to slip to 2019 as it will take a long time getting the stability on the processors correct. We are talking about over 200W TDP parts and Intel board vendors had a hard time getting 18 core parts (under 200W) stable on current X299 parts."
"Furthermore, to continue X299 as a more cost-effective HEDT solution, Intel is going to offer new configurations with a never before seen 22 core part." (This is crap as they already said these are repurposed Xeons, which cuts against the narrative of never before seen because this is a rebadge; also, notice the language meant to upsell the new platform by calling this "more cost-effective")
"The latest X299 processors will also be accompanied with new X299 refresh boards that push the TDP support up to 300W. AMD and their board partners are going to take a similar approach as their 2nd generation Ryzen Threadripper chips bump the TDP to 250W."

That last one means to run the old line (as it pointed out that the 18-core parts have a hard time running stable on current boards) and the new chips up to 22-cores, you need or should purchase a new board. That, once again, cuts against the promise of two-year cadence on those boards, mainly because Intel wasn't straight with MB mfrs last year for power requirements and this year they caught on for the refresh boards to support up to 22-cores, where the AMD counterparts theoretically will be able to support 32-core chips, at least at stock or low OCs.

So, no, this is evidence Intel is screwing their users, just not in the way you were using it which was to disprove the statement in regards to mainstream boards and chipsets. In other words, if I was him, I'd be yelling my analysis I just gave on Intel screwingg X299 users readily, although this, in part, cuts against the narrative as to the Z370 chipset. Please, let me know if I misstated anything here.

 

No AIDA64. I tried it but it was not too my liking. I think once you get the system performance back you will see a huge difference. Now with CB R15 it seems to make little difference but with things like SuperPi and some other apps it does. If you have an extra drive do a fresh install too it of Windows and you will see.

I have tried drivers and a bunch of things to find the issue to no avail. The only thing that was somewhat consistent was turning on the HPET timer to only and EVENTUALLY it would have the performance issue. If you find the issue and it is reversible I am sure it will make a lot of others happy, including me.

I've many of people say I was nuts just finding the consistent indicator of this issue let alone what was always a 100% reversal of it. TBH I noticed it as the desktop lost a lot of its snap and went looking for a reason and this is what I found. And yes it did return the snap back to my system.

Edit; this issue can rear its ugly head as effecting integer math only or both integer and floating point, it seems to always effect integer is why I say to specifically look for that. I mention this as it may be two different settings or whatever but with one of those being the primary trigger.

IE what ever effects FP may not have a noticeable affect to performance till the problem affecting integer performance finally hits.

 

Did a clean install from within windows on the Bench partition to test it, this is the result at 3.95GHz at 3600MHZ@CL14-17-17-17-34-48 1T


|I'm doing a Macrium and EaseUS backup, then will start installing things.

 

Nice to see the performance back. That fast memory does wonders, what memory is that?

Edit; I went too 1.4 on memory and 3600, 16,18,18,18 works but still not as fast as yours.

 

Definitely. On the regular OS, the performance came back using the build 1804 .esd to update it in place. So now just have to keep up with the backups.

For the memory, I have G.Skill 4133 CL19 21-21 2x8GB (two kits for 32GB, which are Samsung B-die). Here are the timings I put into the Ryzen Timing Calculator:



I run 1.4V on the ram, 1.0375V on SOC.

 

Expensive ram, wow.....

 

Bought it at two separate times and before the ram shortage got into full swing (beginning of the class period for the class action). So only about $200 per kit. Had an Intel CPU when I got them, but the MB caused me to be limited to 4000MHz, which because I could run 3733 so tight on timings, I switched between those speeds depending on bench.

Also, found the culprit for Passmark: It was the HPET. When turned on, with the way the program is designed, the latency hit causes a huge reduction in scores. Switched it off, it went back up. So, now they need to be informed and can straighten out the problem.

 

I turned it off and it did not go back up. Also HPET did not cause an immediate reaction either after reboot. As mentioned it was the one thing I found that turning it on as the primary timer EVENTUALLY would cause the issue but it could actually be caused on its own. Then again something could be broke here that does not allow me to actually turn it back off even though the setting is there.

I know others have reset HPET and it did not fix their systems as well. I should note though it has been a few months since trying to enable and disable HPET in windows so it may have been fixed. I will set a backup and try it but have little hope. The reason being Macrium on its won would not restore performance last time and I have to do a reset of windows entirely first and then the same Macrium backup image to get my programs back to restore performance.

 

So, HPET may be the solution to some who have the problem (takes a reboot between turning on and off at cmd line to take effect), but it also sounds like it could be caused by a number of other things. I will look into it further in the future, but because I already reloaded both windows OSes, unless I wanted to load the broken copy (I don't/lazy), I will address it if/when it arises in the future. I still have to go through settings and msconfig items that have built up over 9 months to improve performance as a maintenance, so not ready to troubleshoot one program further yet.

Sent from my SM-G900P using Tapatalk

 

That's interesting, I thought it was primarily Intel CPU's that had the Spectre / Meltdown patches / firmware installed and enabled that saw such a high performance hit on some benchmarks and mostly affected a few gaming results when forcing HPET enabled:

A Timely Discovery: Examining Our AMD 2nd Gen Ryzen Results
by Ian Cutress & Ryan Smith on April 25, 2018 11:15 AM EST
https://www.anandtech.com/show/12678/a-timely-discovery-examining-amd-2nd-gen-ryzen-results/4

It's interesting that this benchmark triggers the Spectre / Meltdown firmware - patches performance hit with HPET force-enabled, as if you check the charts at that link above, it was mostly particular games that saw a huge hit, for Intel...




But, then again the performance hit you all are seeing isn't really huge, but it is repeatable and noticeable.

It would be interesting to see which particular test(s) from that suite of tests is being hit the most - causing the swing in results. Since this suite is supposed to represent real world use, it would be helpful to see what real world use is affected.

 

An interesting summary perspective at where Intel is vs AMD as seen through the recent Computex 2018...

A History of Intel's Ryzen Fall
UFD Tech
Published on Jun 15, 2018
What do you think of Intel's fall from grace? Are you still rooting for the company? Ars Technica History of AMD: https://arstechnica.com/information...e-top-of-the-mountain-to-the-deepest-valleys/


Hardware Unboxed is doing an Intel / AMD comparison series of at least 2 videos, with this first one mostly focused on Intel and what they can do to improve their competitiveness against AMD:

What INTEL Needs to Fix...
Hardware Unboxed
Published on Jun 18, 2018

 

The problem I do not think is HPET, at least on its own. As a primary example I was getting a 13% performance increase before the issue cropped up over @ajc9988 in blender. My bet here is now his is even faster than mine. I also noticed this issue of a reinstall needed at build time back in 10/2017. It only happened 2 times before 2/2018 but at that time I went with Macrium to eliminate more re-installations.

What happened at the time is during an update it all of a sudden lost performance and I had to restart from the get go. Macrium eliminated that need. Also the paid version made cloning to the 2 960 EVO's painless.