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

OK fighting has to STOP! I haven't deleted content yet but continue with any more fighting off topic posts and watch pages of this thread disappear, that is when you guys eventually get back from the forum bans!

 

It seems this Passmark active CPU is being misrepresented in the media. Most people do not realize that users will load and run this in the early life of the system. Both to see what they have and maybe tweak a bit. After the early days most will not run this again or at least not very often.

So what this then gives us is a peak at new systems entering the market space, from OEM to home built. While these numbers may eventually effect total market share, that will take quite some time. It does show though AMD is having an impact.

So again when OEM AMD systems hit the shelves market saturation will start to increase. The smart investor out there knows these numbers are just a preliminary look at what AMD may be capable of. The true battle for market share though has yet to begin.

http://www.mcvuk.com/news/read/amd-is-finally-gaining-ground-on-intel/0184305

 

I like how the article's author tempers the data shown.

Of course AMD is making sales. Market Share though is another story. Passmark is not in the running to indicate anything useful there (imo).

 

Yeah, he tempers it but most do not. Even so the majority of people will not understand about early benchmarking and then those people tend not to continue benchmarking forever.

 

what do u mean by temper? with the benchmark results

btw did you saw the link i linked you to AMD's system for lower 4k storage performance?

 

By temper, I think they might be referring to the premise that the writer of the article is not exaggerating the numbers but is trying to remain objective (unlike many other outlets).

 

" http://forum.notebookreview.com/thr...399-xeon-vs-epyc.805695/page-57#post-10557140">

As far as temper, we mean he is not taking the results as an absolute and immediate change to the market share. Unlike others that look at these numbers as the absolute real time value of market share.

 

ole!!!, in this thread? I think I missed that.

I'd be interested in that link too...

 

http://www.tweaktown.com/articles/8073/amd-ryzen-ssd-storage-performance-preview/index3.html

http://www.tweaktown.com/articles/8073/amd-ryzen-ssd-storage-performance-preview/index4.html

its mostly 4k random performance thats dropping off for w/e reason might be lack of driver or support. this, and also lower frequency/ipc on the cpu are the biggest reasons i can't go AMD. no point of getting optane storage and put it onto AMD system only to lose a chunk of the performance im paying for optane.

 

Dude, those are day of release test results from March 2nd, months ago now, a lot has changed since then to improve Ryzen and motherboard performance.

Maybe find some newer tests, with the most recent BIOS updates.

Wow.

 

I agree with hmscott here. Contact then and ask if they could do a them and now comparison, consisting AGESA 1.0.0.6 is out, plus driver and software compatibility. You'd be surprised what you might get.

Also, you could frame it as wanting to have comparison points for TR, to see if changes to I/O in stepping had any impact.

Sent from my SM-G900P using Tapatalk

 

https://www.semiwiki.com/forum/cont...alfoundries-discloses-7nm-process-detail.html
Upcoming process comparison.

Sent from my SM-G900P using Tapatalk

 

@hmscott @TANWare

if wish to be that detailed, that is the correct way of doing things. however reviewers dont often get samples and/or update. that rig used for testing may no longer be in the reviewer's possession as it is passed around for other testing.

now since @ajc9988 suggestion certainly involves a lot more work, i'll go the easier method. the old amd chipset/cpu even with optimized bios, has rather poor performance on storage devices versus older intel platform. ryzen is definitely better but 4k still lacking. looks like we both aren't going to ask them to update bios and retest it, i guess the discussion would end here with the conclusion that, intel platform has stronger storage performance in 4k vs ryzen.

i mean, clearly as a trend we can see faster cpu would result in better storage performance of the same type of storage device, assuming the driver/software are the same. date this all the way back from sandy to haswell for intel, storage performance hasnt changed much, and some newer IRST driver actually produce LESS performance in single drive device rather than raid arrays. lets not kid ourselves and think bios update can do all the magic here, this could very well be a windows problem not being optimized or simply due to ryzen's design.

nothing to frame here tbh, this is what historically AMD platform have shown as far as when comes to storage.

 

also, if you guys are really down for TR, we can compare storage performance once i finish my intel rig certainly by that time you'll have the latest bios and drivers.

edit: i rechecked the test rig and thought 4k performance reduction might be because of the CCX latency issue and if its somehow related to ram, the ram speed is 3000mhz so maybe problem lies else where. the performance for AMD's chipset x370 prob just isnt that good compare to intel's chipset.

now assuming if it really is the chipset issue, if we were to take the storage devices connected directly via CPU lanes rather than through chipset, we might see closer results, all these are just speculating and from logic afa i can understand things.

 

First, ccx latency has to do with memory and infinity fabric and is the same as Intel mesh. So not related. Learn what the ccx latency is and means.

Second, you do have a point on the chipset. But, the chipset isn't made by AMD, supposedly. Now, if IF is used for the connections to the chipset, then IF latency is in play.

Third, programs have been optimized since then to take advantage of the ccx structure. This means if it is scheduling, that may have been resolved.

Sent from my SM-G900P using Tapatalk

 

LoL, you are using raid then forget 4K, your stripes are guarantied to much larger than 8K so 4K W/R is useless. If you are using a 128K Stripe you would only worry about 64K on a dual drive Raid0 or 32K on a four drive, ETC..

With an Optane direct to PCIe I think things are different there as well.

 

i simply using it before you can talk about ccx as an argument. i dont need to learn any afa i can tell lol. now we dont want amd fanboys to coming up with excuses haha. btw from the way you talked i can see how conversation eventually get heated up. donno if its your personality of simply due to it being your 2nd language.

naw, i brought up raid as an example that driver dont always help out performance, should be clear that some driver does better than other in a targeted area, such as power saving, efficiency, consistency etc. the same can be said for firmware or bios, i am simply bringing that up for him and show him that not bios isnt all that magical.

however on the other note, i'd be interested to see intel's VROC and how it fairs on against traditional raid 0 performance. afaik 4k performance dont stack in raid unless help with writecache but i wonder if theres something similar to writecache option in VROC vs raid via chipset and IRST.

 

First, English is my first language, not second. The errors in my writing often arise to autocorrect owning me.

Second, as far as software optimizations go, CCX can still be better utilized if not addressed. NUMA aware software usually has the easiest time, but as we saw with ROTR and others, changing the size and ordering of the schedule can give large performance boost. This means this can still be an issue.

This, you mischaracterize the CCX hit on latency between the cores of different CCXs. It, at 3200MHz ram, is just 10-20ns slower than the mesh. We hammered that out forever.

Third, it is when a person improperly characterized things consistently that I take issue with. Even here, you are trying to rule out what amounts to software optimizations by pointing to the latency between CCX, which is relatively fixed and is an architectural feature. In other words, you conflate a physical feature for something that is handled in software. Considering if properly scheduled, you'd have less cross-CCX talk and better utilization of the cache structure (which even Intel skylake-X needs optimized on), there is more than what you lead on.

Sent from my SM-G900P using Tapatalk

 

thats what i made you think didnt i? you are assuming things that i did and jumping to conclusion. i think you already went down that route with tillerofearth so i'll leave your argument here. if you didn't like the example, simply say its a poor example and i'll try to revisit it. when i think too much and then try to put into words which im terrible at, realize it'll become a story, sometimes i just give up and say the first thing i typed out . honestly speaking, i use to care and want to prove im right all the time, now i just dont care anymore lol. I still believe what i learned is correct and if you dont agree with it, thats your opinion. quite frankly if you're wrong, then i dont need to let you know where its wrong and you can stay that way, if im wrong, i'll correct my basic understanding and re understand it when the time comes.

now back onto topic, i gave you the reason why bios, firmware/ software /driver isnt at all perfect and can't fix everything. AMD platform is one of those reason, bios wont do much however the chipset itself might be the factoring point here why its 4k performance is much poorer than intel's.

 

No, I'm trying to clarify what you said, this time because of ambiguity. I have to assume you mean plainly what was said, although here I should have asked for clarification, even though your conclusion is factually incorrect, except as to the chipset.

When talking about the inter-ccx communication, we are talking multiple things. One is IF and the latency surrounding it. Second is firmware changes, which one change increased support of higher memory speeds, thereby increasing IF through put, which can effect the results. Software can change how the program uses the CPU during the read and write, especially in utilizing the cache system, which effects reported speeds to a degree. Drivers are obvious as well, and one cannot say that the older drivers work better if it doesn't recognize our utilize the scheduling on the new hardware.

This is why I try to explicate so that it won't become simplistic in one saying things as an absolute.

Sent from my SM-G900P using Tapatalk

 

ole!!!, thanks for digging that up. I did miss that.

Your points are spot on and understood by most here - but I already see you're not worried...

Can't be stated simpler than that. But the graphs are pretty too.

 

W/o more recent tests to comfirm we can't say for sure Intel is faster. TBH, the 4K W/R Q4 is not so distinctly off from each other that you will notice this real time on a day by day basis. As I had mentioned also forcing raid0 on a larger stripe forces the elimination of smaller W/R's so the supposed difference is reduced significantly.

 

This also shows that @ole!!! 's thinking is wrong:
https://www.pcper.com/reviews/Stora...e-RAID-Tested-Why-So-Snappy/Latency-Distribut

Edit: specifically, for the price expense of large sized optanes, which may take up a slot of the PCIe slots or that you use as a cache drive, you could get a raid card supporting NVMe M.2, then populate it with moderate priced NVMe drives, have them in raid, and still have 48 more lanes to use for whatever you want.

Edit 2: "One additional point those with a keen eye would have noted is that in both cases (writes and reads, but reads less noticable due to the log scale), shifting from a single SSD to a RAID results in a ~6μs additional delay to *all* IO requests. This is the overhead cost of Intel's RAID implementation, and it represents things like the time taken to translate the IO addresses to the array. This added delay does have an impact at very low queue depths, but it is almost immediately outweighed by the increased 'acceleration' of the array as the queue depth climbs just a single point to overcome the effect."

 

well, the problem is, hitting QDepth of 4+ constantly isnt as easily as one would think, doesnt happen much in consumer's workload, also what I care the most about is QDepth 1-2.

im glad you bring up articles of a reputable site, though i dont see how my thinking is incorrect.

the article shows, raid 0 will reduce your QDepth 1-2 latency due to overhead for 4k read/write but will increase QDepth 3-4 onwards you start to notice the difference. in simple terms, you only feel faster once you hit QDepth 3-4 versus a single NVMe device. when you're in QDepth 1-2, its actually slower, where by optane technology can cover up that difference even with raid versus traditional NVMe. how I replied to Tanware just above as I mostly care about QDepth 1-2, raid via chipset will increase the latency due to overhead & hitting QD3-4 constantly isn't common nor easy.

however I wonder what will happen raiding with VROC, which btw is only available on Intel's latest x299, i am extremely interested to find out about its raid performance along with turbo boost 3.0 to see where it can take us. according to Allyn he mentioned the overhead would simply disappear that means if we were to raid it would benefit more than raid via chipset.

 

I mentioned QD-4 as those links used the benchmark set at QD-4.

 

Why do you bring up chipset? I said they are designing pcie raid cards holding as many as 4 nvme drives. As such, it isn't the chipset, rather the [card's] controller, dealing directly with sending the information to the cpu, thereby not dealing with the chipset. Granted, that is an additional expense, but it is a technical detail you ignored in my statement.

Sent from my SM-G900P using Tapatalk

 

the on on pcper? they also got QD1 results which raid doesnt benefit it anymore than a single drive

just what are you talking about, i am talking about chipset AMD vs intel, what about you.

 

I'm talking a raid card, which removes the chipset from the equation. As I said, it has been speculated AMD is buying the chipset from a third party. Meanwhile, if speed is the concern, and we are talking about pcie for the larger Intel Optane drives, at least initially, other than the cache drives, let's put it on equal footing. That is the raid controller on the pcie card which will support 4nvme drives. Raid isn't processed by the chipset or cpu, meaning you are left with great speeds and limited to the control of the raid on the card for latency (other than other system latencies including pcie).

Sent from my SM-G900P using Tapatalk

 

They tested not at QD-1 but QD-4

http://www.tweaktown.com/image.php?...yzen-ssd-storage-performance-preview_full.png

 

and why would you be talking about raid card? im talking about raid via chipset and raid with VROC.

CDM and other benchmark shows QD1 results as well for 4k.. ATTo is using QD4 as it has no way of showing anything less than QD2.. also ATTO tests sequential performance only and those 4k you're looking at on ATTO is block size for sequential data, not 4k random data.

 

Fighting has to stop. Name calling, slandering etc.., stop it all.

We will end this with the 2 points I can get from this.

1.) As of release the Ryzen is slower. Apparently the x370 is not mature as of the release with SATA.
2.) with the chipset out of the equation and a proper raid pcie card things do equal out a bit more.

 

Get out your salt shakers, without any real news even Wccftech is being left to rumors;
http://wccftech.com/amd-ryzen-threadripper-1950x-cpu-performance-benchmarks-leak/

Just a heads up, the geekbench shows as an engineering sample (I think the one used in the early gaming demo's) and is a stepping 1, not stepping 2 and still shows as Summit Ridge. Also southbridge as "AMD X370 51"

 

Only 3.5-4 weeks on details, 5 weeks to release. Then the answers will come. I'm betting board info in 2-4 weeks, including siggraph.

Sent from my SM-G900P using Tapatalk

 

a new update boys! this is what happens if you have a bunch of competent people who knows what they're doing, working and pour their experiences together and you get just pure knowledge and facts and how things work. love it



@unclewebb though they guy didnt mention throttlestop to detect throttle he said coretemp, which is also made by you!! without your core temp and set affinity to higher than other software we would be throttling nonstop.

 

dont even bother with geekbench its so intel sided tbh. i just ignore that article all together and wait for the reviews themselves. wccftech click bait typical article.

 

Someone is posting Ryzen 1800x scores and spoofing geekbench by making it look like it's TR.
TR is UNRELEASED as of yet.
Until we get a final release, everything is up in the air.

 

So, in other words, he was correct on just about every point and had to teach reviewers how to use their hardware and push it (with settings I learned awhile ago) to replicate the results found by him. This would make the world's #6 ranked overclocker the smartest guy in the room, mostly, compared to other reviewers. They also came to the same conclusion as him, the heatsink is rubbish. I love how he pointed to the M8E (same board I currently have; I've ran 4000MHz ram on it in all four slots (hard to do with these boards without binning)) to show how cool the VRM can and should be. They definitely hold up.

I did appreciate the tip of the hat to the Silicon Lottery group considering he is with CaseKing and they are competitor binners. Class!!!

 

X299 VRM Heat - Der8auer Collaboration

prime95 small FFT FTW

Answered: Intel used the toothpaste to "pre-gimp" the CPU so it would throttle first, so the under designed consumer hardware would survive, and the VRM's would run cool.

Also note the BIOS power setting, from Auto to 140%. At Auto it's so power limited you won't get full performance, and the VRM's might be safe - but, if a tuner gets a hold of the X299 and tries to get the top performance, the power section is under designed for what the CPU's will allow (delidded).

All those built-in limit's are chickenshot methods of protecting a poor design, cheap design, and stealing the top off the value of the i9 CPU's at least from 10 core and higher.

It would be good to see if the lower count core CPU's are also performance limited at defaults, and if removing the limits starts that same too high load for the design.

 

Well, if nothing else, AMD solders, so it is whether they designed it to cool properly (I think X399 will benefit from this) and fully open it up. Time will tell.

Sent from my SM-G900P using Tapatalk

Edit: after reviewing the entire video, he brings up a great point I get to stress in stability testing: ALWAYS USE MULTIPLE PROGRAMS TO STRESS TEST!!! Notice when he said OCCT would fail and P95 would chug along. That is very important for people to understand that different programs load the CPU differently.

So better reviews in the future coming up!

Also a class act on the collaboration.

Finally, as they mentioned, and many of us laptop overclockers know, switching out thermal pads, adjusting contact, etc., can give better results. Now, if you plan on a water chiller (like me) or phase change (like @Papusan), you'll need to cool then down. I plan on a water cold plate to try to get them around 25C or lower for cleaner power delivery, but even keeping them in the 60s-70s will be fine.

 

From what I've seen, only peripherally, the X399 boards might have weak power designs as well, maybe worse than x299, so I'll be watching carefully for reviewers like der8auer and OC3D covering X399.

Maybe now more reviewers will check power temps and design more carefully moving forward.

 

I agree on that point, but it isn't that they are weak, it is the cooling is weak. Now you can use a doubler, add VRM to spread the heat and load, but what matters is the power delivery (referring to connectors, like two 8 pins) and the cooling. If they have to redesign a heatsink for the VRM on an already launched product, they will likely address it on the unreleased one coming up. (I also added thoughts on the edit to the post you just quoted).

Now, AMD is known for ramping up current needs above stated clocks. It just grows exponentially, historically, of course variant by part. So, it will be interesting to see actual CPU draw on TR.

But, we'll know more in a month! and always watching for reviews!

Sent from my SM-G900P using Tapatalk

 

no, in other words he is incorrect about 300w on an 8 pin power cable. also he did not mention how his power cable got to 65+ C in this 2nd video. swallow the facts man.

 

@hmscott the toothpaste is very disappointing i think many that are getting skylake-x would agree with me however all these high power consumption in benchmark test we gotta keep in mind that, we dont get computer for those purposes at all. if in the range of 350w to 400w with the usage Im using at which still has AVX workload, all threads say to 4.7-4.8ghz, I will only be hitting that wattage maybe once a year, or almost never.

they are clearly doing worst case scenario not like i dont understand how it works, but thats targeting high end enthusiasts that like to bench with heavy benchmarks, i dont do those i find them pointless. I may do cinebench or XTU once or twice to test temp and performance to see if throttling but thats about it, to see if im getting the performance i should be getting.

if delided, this cpu will run a bit hotter than broadwell, thats about it.

 

In these tests the Kabylake-X 7740x / 7640x vs Kabylake 7700k / 7600k show the new X299 systems drawing way more power to deliver the same (or less!!) performance as the x270. 20% more at idle, and 40% more at load!

Intel Krappy Lake-X: Core i7-7740X & Core i5-7640X

 

We already know 300W is able to be carried on a single cord. It actually can carry more than that. But power delivery from it, even reaching that, is a different story. Have you seen what the patterns look like on current as it goes above 25C on the VRM, then moving up toward the absolute limit? There is a reason video cards with 250W have two to three 8 pin when I've seen 180W pulled from the PCIe in death defying acts of heroism. If you can not understand what happens and why more connectors are preferred, maybe ask why, research, and find out!

Everything except the cable was replicated. No one gave a darn about the cable because the VRM are throttling! Priorities!

Sent from my SM-G900P using Tapatalk

Edit: also, with 350-400W on the power cable, the heat measured makes more sense. But this is measured by either a temp gun, a camera that picks up the heat differential for later analyzing, or direct temp probes. I'm doubting the other reviewers were focused on testing the cord when they needed help drawing the current by the CPU.

 

Yeah, we already knew the TDP was raised on the lower end skewed CPU's. One of the first complainst against x299.

 

But we didn't know that that was for the same or worse performance with the higher power draw...

Sent from my SM-G900P using Tapatalk

 

Well, since no extra pci-e lanes, no extra clocks and a worse TDP, it should be expected the same or worse performance.

 

naw, you need to be more detailed than just that, but once again i dont fee like arguing with you anymore so i'll leave it be.

on the other hand, 7900x all sold out and i miss it by 1 day now all gone wtf silicon lottery. doesnt look like anything above 4.8ghz. waiting for a 5ghz 7900x

 

What kind of cooling do you plan for that 400W+ blast furnace?