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

That is certainly interesting, I was under the impression that they were just losing market share to other Xeon/Quadro based workstations (like MSI's), but the secure processor might be a good selling point.

 

They made sure to focus on security for these chips, that way to push it as a selling point. Literally, why do you think the bs articles about VM support on Ryzen Rx series came out recently, when that was mostly resolved with Linux kernel updates months ago, or the error that was fixed with a microcode update months ago (FMA3 instruction set)? It was to try to prevent adoption of Epyc on fears of VM support.

Now, I know the errors the articles pointed to were resolved (primarily the articles were on investor sites like selling alpha), but I cannot say for sure every instance has been resolved. But the security is a wonderful selling point.

Sent from my SM-G900P using Tapatalk

 

http://www.tweaktown.com/news/58180...al-cpu-against-intel-56c-112t-cpus/index.html

So, Intel Xeon platinum 2P against an Epyc 2P...

Sent from my SM-G900P using Tapatalk

 

I was just going to comment on that...
Are those scores taken with latest sisandra or were those previous scores?
Because, the new sisandra version included support for Ryzen and beyond and scores have increased over a double.
I'm just wondering if we're seeing that older version of results, or is this up to date?

These are the scores with new sisoft sandra version that were supposedly taken a short while ago:
http://ranker.sisoftware.net/show_r...d5e3daeed8e0d4f280bd8dabceab96a680f3cef6&l=en

And these were the scores taken before sisandra update
http://ranker.sisoftware.net/show_d...f5c8f8deb68bbb9de5d8e8ceabcef3c3e596ab93&l=en


Reading sisandra is confusing me with all different values.
Could someone elaborate on how these scores relate to each other including the newest ones posted?

 

That is not fair, look at the cost. TBH I was looking for a higher score though in R15.

http://www.tweaktown.com/news/57293/intels-new-xeon-rocks-28c-56t-costs-over-12-000/index.html

 

if only we got the microcode can unlock the xeons, 28 cores at 4ghz thanks intel

 

Yeah, it's actually quite ridiculous.
Also, the Platinum is running at much higher clocks than EPYC (700 mhz difference in base clock speeds - or 26% differential in Intel favor).

Adding to that, that multimedia performance is eerily similar to the older Sisandra version results:
http://ranker.sisoftware.net/show_d...f5c8f8deb68bbb9de5d8e8ceabcef3c3e596ab93&l=en

Vs the updated sisandra results:
http://ranker.sisoftware.net/show_r...d5e3daeed8e0d4f280bd8dabceab96a680f3cef6&l=en

So, Intel still has multimedia advantage if the 'new' scores are taken into account, but the difference is not that large.
And if this is the case, then someone is not posting proper results (or we are getting incorrect ones) in order to make EPYC look bad vs the Platinum.

But I cannot help and wonder if AMD raised it's clocks to 3 GhZ or above, their results would improve
dramatically.
That and Sisandra including more updates, and the fact that EPYC costs far less?.

Tweaktown hasn't exactly been 'reliable' in terms of news (same for Wccftech).

 

Looking at the comment section of that Wccftech article posting EPYC benchmarks is making me even more suspicious of these 'scores' representing proper numbers.
Everything from clock speed to results (in synthetics no less) 'reek' of old numbers that we discussed here a few pages back and saw updates to Sisandra for example which gave EPYC over twice the uplift in results.

And we still don't have EPYC officially out for proper testing.

 

The reason that they compared them was to do flagship versus flagship, rather than price point versus price point. So, you can pay 1/3 and get how much performance?

I do agree. It popped up in my feed and I posted to share, not the veracity. We had similar conversations surrounding other articles posted from TT awhile back (or was it WCCFTech?). Either way, this is why it is good to post for discussion, because some days some of us are too busy to dig into the numbers, but we can guarantee someone here will and will give the additional information needed to get to the truth of the matter.

BTW, thank you for doing the leg work since I couldn't at the moment.

 

I do not think Epyc was meant to undercut Xeon in the same way it has the other Intel line. Before all the CPU's seemed geared towards 50% or so of the price of the competitive part. It seems AMD here wanted to price the same cost of the Intel part would yeild a 20% to 50% performance improvement. AMD does not have 4P or the like yet.

So what we need for the 7601 is a $4,000 Xeon and compare those etc.. Comparing it to 1 in a million $12,000 chips is insane and AMD should be expected to be handed a smack down.

 

Well, you have to remember, 4P and 8P are a minority of cases.

Sent from my SM-G900P using Tapatalk

 

Some more benchmarks (i can't seem to find a price on the 8168 or others);
http://wccftech.com/amd-epyc-7601-benchmarks-analysis/

 

Intel's new Xeon rocks 28C/56T, costs over $12,000 Intel launches its next-gen Xeon Platinum and Gold Series processors, packing 28C/56T
http://www.tweaktown.com/news/57293/intels-new-xeon-rocks-28c-56t-costs-over-12-000/index.html

 

I saw the 8180, the others I can't find. There are numbers for 8168 and you can even find the benchmarks of some others. I think Intel is aware of the comparisons and is keeping the price list a secret so others will find it harder to compare. Prior to the gold and platinum series prices were easy to find.

 

That depends on whether we can consider this as such.
Take into account low clock speeds on epyc, coupled with old results, and the fact that except multimedia, the numbers are within 10-20% of each other.

Not to mention epyc is still not released

 

If believable here is the 8176 2P more in line with the 7601 Epyc 2P results and the CPU supposedly is available at $3222. If true AMD may need to adjust their price structure. Even the 8168 is only $3814.

http://wccftech.com/intel-28-core-xeon-platinum-8176-8168-xeon-gold-6161-6142-benchmarks/

 

4 Epyc are, the for flagship chips.

So, are you basing that on cinebench scores alone? Also, is that Epyc price MSRP? If so, you are comparing a reseller price to an msrp. Just trying to give context.

Further, there are even higher memory bandwidth scores now, compared to the 250GB/s score I cited awhile back and you still have 50% more PCIe lanes, which is worth something.

Sent from my SM-G900P using Tapatalk

 

I only have Cinebench, and as I mention they MAY have too reconsider. We do need allot more info but this is what we have to start, and it does not look good for now. AMD is welcome to have others benchmark their chips to justify to consumers the price base, but I would suggest doing it without making adjustments to the other scores as they feel they are not fair.

 

Woah, woah, woah. It is limited numbers on INTEL's chips, not AMD. All we have on the Skylake-SP chips has been leaked and show cinebench. We have more comparison with v4 Intel chips, but that is a generation old. Intel's chips are not on sale yet. But you are blaming AMD for lack of comparison data. Do you see the problem here?

Edit: also, the adjustment you mentioned has been explained AND they did give straight up numbers on Intel's 2P flagship, which showed them beating those chips in a 2P setup. Further, pulling it from the site itself still showed an advantage to AMD. So, that is a fair critique, if you also ignore the other data.

 

No, what I am saying is the data we have does not support their numbers. Also if you do not like the benchmark just do not use it. Intel did not pull the data to use from Epyc. There are plenty of third party suites and benchmarks that could have been used.

As far as these newer chips, I agree but again being as they are the standard you are up against then that is the market you need to target pricing against, not the old V4 chips. So again they MAY need to restructure pricing.

 

The data is out. We have numbers other than AMD's now. Those can be compared to what is public. Then we have articles comparing to old numbers before Epyc optimizations. But we can compare what we have without marketing materials or relying on journalists. What we cannot do is compare to unreleased products with leaks.

As to pricing, WHY BID AGAINST ONESELF? There is no reason, and will be no reason, to even consider re-pricing UNTIL Intel releases a product to justify it. You are calling for it too early is what I'm saying.

Sent from my SM-G900P using Tapatalk

 

One other thing about those Sisandra multimedia results is that avx512 benefits Intel a lot.
However, other people also suggested that the difference is too large (to the point of being impossible).
They suggested that avx512 would give 30% advantage to Intel (and even updated Sandra for Ryzen apparently still gives Intel double the score in relation to EPYC - which is better than 4.5x advantage seen with early benchmarks, but still too high - especially since EPYC cores were in multimedia mostly underutilized and running on 1.8 GhZ it would seem)... but for now, it's really only used in synthetic benchmarks and I don't know of any industry-wise applications that might be using it (anyone know of any?).

Bearing that in mind, Platinum still has a clock speed advantage, so AMD could easily overtake it for a small bump in speed (for example to clock it at 3 GhZ across all cores - it might be manageable).

At that point, Intel would probably lose any edge it has and would be 3x overpriced (and even if Platinum is faster than EPYC by 10-20%... would companies really pay 3x more for such a small increase as opposed to simply OC EPYC themselves?
Newsflash - AMD usually releases unlocked products, while Intel not so much

 

Btw... this article seems to be more 'balanced' in terms of reviews as it does mention the possibility of lack of optimization for EPYC or the benchmarks being taken with an older version:

https://segmentnext.com/2017/06/27/amd-epyc-7601-with-64-cores/

Is it too difficult to ask for 'balanced' reviews of (on top of everything else) unreleased products?

 

If there were an independent review body, maybe, but even then I'm sure someone would figure out how to game it to distort opinions right up to release.

 

SiSandra has never been a good judge for bench marking between systems. I had always used it to look maybe at an upgrade and its differences but that is about it.

Cinebench is a good mark for if the threads are severely optimized and heavy, that is about it. By no means a gold standard.

Most enterprise do not purchase a server with the idea of overclocking it. It is looked at as it is what it is. Which is fine by the way.

 

Wouldn't make business sense to buy something based on theoretical performance after modification, it's got to be out of the box TBH.

 

Correct me if I'm wrong, but even Cinebench requires optimizations to support new hardware.

 

Is it the M/B makers fault or Intel's? Like der8auer say's 'probably 50/50'.

But there is also one other at fault too: whoever buys this and keeps it when it is not fit for purpose (yeah; the consumer). And I hold them most responsible, imo.

Myself; if I was in the market for an upgrade over my existing platforms and was considering an x299 based solution; these issues would probably not affect me (I don't O/C...).

But if I was buying for a specific goal/outcome and it wasn't reached?

RETURNED w/FULL REFUND.

gen 2 may be the X299 boards to get... but manufacturers whining about other manufacturers not giving them enough time to 'design it right' is not what I'm interested in.

I'm buying for a specific outcome (productivity, reliability and dependability) and anything except a fix on those aspects is just an excuse.

And I don't pay for excuses... and neither do my clients pay me for excuses either.

Jumping first is fine... but make sure where you jump to is not going to hurt to bad... (always buy with full refund and no restocking charges...).

 

I'm glad that AMD is doing well and I certainly hope it continues on that trajectory since it needs a good year or two (maybe more) with Intel on the sidelines.

However, these 'issues' with Intel's X299 boards and cpu's could be resolved with microcode updates much like early woes with Ryzen we knew would be/were (performance increases, RAM support, stability, etc. - and new updates are still coming if I'm not mistaken).

If this is down to the hardware level that can't be fixed with microcode updates, then its a different matter, and it would further put a bad rep on Intel for having (essentially) overpriced hardware that doesn't work properly.

Granted, who is to say that EPYC won't stumble on similar issues?
Though right now, we don't seem to be getting indications of EPYC suffering from such issues. Or do they?
Anyone have any information on that?

Then again, EPYC is basically several Ryzens tied together with infinity fabric - and AMD was able to resolve most of the issues with bios/microcode updates... I don't think it suffered from hardware issues permanently baked into the design - still, EPYC is a different configuration and will likely need optimizations and patches for infinity fabric connection.

On a side-note... I wonder how much more performance could OEM's and AMD squeeze out of Ryzen with further microcode/bios updates.

I certainly hope Asus ROG laptops with desktop Ryzen's get the same treatment in terms of microcode/bios updates and that we as users will be able to update them.

Do OEM's still lock the UEFI/BIOS these days on new laptops?
Or is there any word on having an unlocked UEFI on Asus ROG with Ryzen and RX 580?

 

I'm going to chalk it up to initial release hiccups until I hear otherwise. People further down the chain, approaching and including consumers, will always find stuff that even the most dedicated QA and devs can miss. Identifying them and fixing subsequent releases of the same products is normal to me.

 

About the unlocked bios, that i probably a wait and see game. My concern too is power available on the x399 platform as it too will have a huge power draw. This especially if the 1998x is true and overclock-able. In the end we shall see and maybe this is not a good idea to jump in to be the first with a TR/x399.

 

look at the power consumption for 10c overclocked cpu easily 400-450w. assuming i can cool it off, finally can get rid of my space heater.

 

You can't upgrade the failed design physical power delivery on x299 motherboards with a BIOS update

 

u'll be surprised sometimes what bios update can do. since its firmware/code control hardware down to the basic level, it does make some difference.

now from my understanding, the board vrm is junk, either it is junk or simply because board makers dont have enough time to test them as chip was rushed. i dont understand the specifics of vrms but i'd make an assumption that the vrm used for overclocking 7700k which consume 130-140w maybe used for x299 and that prob just doesn't work considering how much poewr and heat is being generated by x299 + skylake-x. if they simply used that idea and just design x299 with that in mind then its going to have issue.

we didn't hear much problem about broadwell-E though, except that its overpriced

 

would be funny if 14c-18c HCC comes with soldered to IHS. buy a 14c and disable 4C to overclock LMAO

 

It looks like a rushed job, someone copy / pasted the power delivery section from another board, and didn't pick one scaled up enough to handle the higher power demands of the high core count i9's.

It's also likely Intel releasing info for the CPU's they had vs. the ones that will deliver later with even higher power requirements.

For the currently shipping boards to all fail to deliver enough power within a safe temperature range with the VRM's suggests Intel really low balled the power spec.

It's also the MB makers not having enough time to test out their "vrm heat shrouds" to make sure they weren't running too hot.

The single 8-pin power delivery in particular isn't going to be fixable with a BIOS update.

The other problems are also all physical, there are no updates that can be done other than pre-throttle the CPU's and power delivery to "safe" levels to make sure the boards and wires on the PSU's don't catch fire.

As de8auer said, there isn't a single x299 board on the market that he can run safely, and that means the same for all of us too.

We are going to need to wait for the 2nd generation X299 boards.

 

if u look at his 8 pin power delivery, u'll see that only 6 wires for the 8 pins. so that in itself got some issue.

 

"His" 8 pin power delivery? Don't you mean the PSU power delivery?

Contact him if you think you caught a problem he didn't mention.

It's also possible someone on his team shot or edited in B-roll that's not appropriate

 

To put this in context, you are questioning the statement of an overclocker which is in the top 10 overclockers in the world (currently ranked #6)...

http://hwbot.org/user/der8auer/

Sent from my SM-G900P using Tapatalk

 

thats right, i question it. if i see things dont make sense, i question. i dont care about what others think of it, your thought process is correct, but thats just how i am. though dont take in too much what i say cause people get offended easily for no reason. i always had laptops so this would be my first time building desktop, which makes your argument sound even more correct .

 

It's cool, I'm just thinking that what was filmed wasn't necessarily him putting anything through the paces, especially with him mentioning removal of the VRM shrouds.

Now, the majority of the heat was being trapped by badly designed shrouds. Then you have the power limits and thermal limits of the VRM, as well as concerns on power delivery smoothness.

Now, I do agree part is firmware/microcode related. The weekend before the pre-order, many removed the TDP limit for better performance. If Intel was not forthcoming on the power requirements, this may have made the entire design go well beyond the initial limits thought to be needed. You can tune some of that moving forward, but it is limited.

As to problems of power delivery (single 8 pin), selection of VRM, and the shroud/heatsink bad design, it is mixed (most cannot be changed, but the VRM cooler market may pick up, as well backplates).

But I do enjoy giving some ribbing!

Sent from my SM-G900P using Tapatalk

 

It's called "shooting from the hip", or a "knee jerk reaction"...

Spoiler: ...for ole!!!

It's something you can control, it will help your reputation, and keep people from Ignoring you.

If people have to question everything you say / question, as to whether it's real / important enough to pursue or not, pretty soon people will ignore what you say - like when I Ignore your posts for long periods of time.

Give things a little research, ponder them a bit yourself, and see if you can come up with reasonable reasons for something as is rather than questioning it immediately without due thought and consideration.

It's true that some things are going to be "controversial", or "too small or rare to deal with by most people". Those are the most important to get right the first time, with plenty of research to back up your claim(s).

In those cases there are people that stick to the unpopular issue until resolved instead of passing on it, and that's a good thing - but those are real issues that have a real chance for large consequence.

In this case de8auer said he tested a bunch of motherboards, measuring power and temperature in several places on the motherboards and PSU to see where the problem points might be.

It's pretty unlikely he didn't use proper procedures or equipment for his testing, but if you have a question about it ask him directly, don't post doubt publicly until you have something to back it up.

That's why people get upset at your questions, it's your inappropriately interruptive delivery impromptu to real thought or research.

Learn to hold back until you really have something of value to add.

And, above all else be considerate and polite

Malvin & Jim - Remember you told me to tell you when you are acting rudely and insensitively... you're doing it right now.

 

I especially liked it when der8auer suggested water-cooling the VRM's

If the board makers had more time for testing more of them might have implemented full water cooling jackets for their motherboards like with x270.

 

I'm planning to water cool mine on the X399 board. No reason not to chill that beast, if nothing else just for smooth power delivery by trying to keep them under 25C. But, on air, it should properly be lower than he was getting, even inside a case in summer, as he put it.

Sent from my SM-G900P using Tapatalk

 

EPYC Enables New VM Boot Security

 

AMD EPYC and Supermicro Ultra 2U System Windows Server 2012 R2 Compatibility

 

All this mess. Blame Intel rushing this out due hard smash from AMD. Maybe put the blame on AMD? LoL
And same is happening on Z370 mainstream. A short lived revision!! As mainstream can't compete with Ryzen. Already confirmed. Intel living hard times now. What a mess.

 

It's only a 'mess' if one has dipped their toes into it...

Intel isn't living hard times (yet). But they are having to think on their feet a bit, eh?

I'd like to think AMD is feeling a bit of the same; there is no way they could have predicted what Intel's response would have been to them (at least not with any accuracy).

An informal poll of my tech supply contacts do not mirror what I'm gathering from this forum and others (i.e. AMD/Ryzen enthusiasm); Ryzen isn't selling as much as it is being talked about right now.

People may be more than willing to wait this out and see who is still standing in the end? Either way; the only 'hard smash' from AMD that Intel would care about is loss of sales. I don't see this happening in my circles of business acquaintances.

 

Your circle of business acquaintances is smaller than Amazon's quarterly sales statistics, most likely, in regards to mainstream systems, which is what data is available so far except for servers purchased for test beds except for hardware partners that have been testing for months.

Amazon saw AMD CPUs taking 29% of CPU sales, significantly up from first quarter. I'll take that number from Linus over what your hardware partners have said to date.

Further, the new SL-X AMD Kaby-X have been on presale since the 19th and on sale since the 26th, far too early for accurate numbers except early adopters, especially since it has no comparative AMD product yet. (Although it can be argued the 6 & 8 core chips should be compared). We also don't have the data points on Intel 6&8 core sales.

Because of all this, it's improper to say either is being effected yet except what we have hard data on, the mainstream platforms, which Intel's market dominance had been hit.

Now, issues with the platform is growing pains, but is completely fair criticism, just like Ryzen was fair to criticize the first couple months.

But a jump of over 20% on sales of CPUs from a major distributor for home builders is significant. But, home builders is also smaller percentage than from OEMs, and the OEMs only announced Ryzen products at computex. So even the numbers from Amazon need discounted appropriately.

Just trying to clear the hype and provide context on both sides. We won't get Q2 earnings until July/August, so no one should say clearly what is happening yet (unless you are an insider).



Sent from my SM-G900P using Tapatalk

 

You may have misunderstood me (yeah; re-reading my original post I'm a little vague - sorry).

No doubt Amazon's stat's are a better gauge of sales!

My point being that 'most' business users are being much more cautious than 'home users'. After all; performance, stability and price are #1 there...

But this sharing and discussing of these finer points is good to see.

(Thank you for unblocking my posts and being able to read/respond to them as you see fit).