Intel Internal Memo Reveals that even Intel is Impressed by AMD's Progress

[Origami Cactus]

1 hour ago, Tedny said:

So, AMD have mobile chips or something? 

They have had Ryzen mobile chips for 2 gens now. The 3000u series processors are actually pretty competitive, but unfortunately they are zen+ not 2.

[Origami Cactus]

Just now, Tedny said:

That laptop makers will buy and push* 

Yeah, the 3rd gen ryzen laptops are actually being sold, and at great prices.

[MadDuke]

1 hour ago, Tedny said:

So, AMD have mobile chips or something? 

Yep. Altough outside of the GPU department Intel kicks ass still. Especially the new 10nm chips (probably, makes sense)

What I think most people didn't realize is just how much price of their laptop is connected with the CPUs in Intel cases.  Even an 8550u is 400 bucks or something. 

 

I think AMD (probably tied to manufacturing but I don't know) missed  their mark a bit with pushing the 7nm Ryzen/Navi  APU refresh for laptops for this "back to school" season. But hopefully next year. 

 

Other than dedicated GPU laptops for workstation and/or gaming use.  I think most people just want a speedy laptop that could start something GPU wise if necessary.  Most of my friends who don't do any of those things don't even own a desktop anymore.  They also buy ~500USD laptops and that is a really big market IMO.

 

15 hours ago, leadeater said:

@MadDuke I fixed the formatting a bit for you, added proper quotation and spoiler for source link as it was LONG. A little bit more personal commentary/opinion is generally expected i.e. expanding on why you find this interesting.

Thank you. Didn't know about spoilers. (Haven't been active in some forum for half a decade or more).  

 

 

[RejZoR]

2 hours ago, Tedny said:

So, AMD have mobile chips or something? 

I have Ryzen 5 2500U and it's pretty damn capable. 4 cores, 8 threads, 3.6GHz turbo.

[CarlBar]

8 hours ago, Neftex said:

single core intel, correct. but multicore also intel because xeons

 

If your including Xeons. Then EPYC still puts intel down. Which is the problem. it isn't just about cost. At any given performance level you can find an AMD part that is competitive on raw performance with intel. the fact that they're significantly cheaper is just pure gravy at that point.

 

Which completely ignores that fact that Server CPU's aren't allways the best choice either for someone. Even AMD adds a fairly sharp markup to their server parts. Whilst i'd have to double check the benchmarks to pricing comparison from memory i'm pretty sure Intel's HEDT processors offer equal performance for lower cost than AMD Server parts, (despite AMD having the performance edge in many situations with Threadripper at a significantly lower cost as well).

 

Then again you don't seem to understand what server parts are for, a lot of Xeons actually get their butts kicked by the Intel HEDT lineup. Yet companies still use them, ditto for EPYC compared to AMD's HEDT lineup.

 

Server parts aren't really about maximising performance. They're about scalability, specialised support from the CPU manufacturer and also specialised complementary hardware/software from other vendors that only works with server grade hardware. You only pick server grade stuff when you need that kind of thing. performance is rarely the primary metric in those use cases, (which is why the markup exists, cost to performance isn't the main thing people care about there), and only the highest end SKU's actually offer any performance edge over consumer and HEDT parts. And even then they can end up performing worse depending on the application.

[Neftex]

6 hours ago, cj09beira said:

its not as simple as that though, look at the Rx 470/ Rx 570 it has had the price performance crown for a long time, with much better performance than its competition but sales wise it doesn't seem to sell much better than other amd cards and not anywhere near what the 1050 ti does for example, so there is a good amount that is dependent more on brand perception than actual price/ performance

is the price/performance difference as significant as in cpu segment? didnt think so, also as mentioned mining was a thing back then

[Neftex]

35 minutes ago, CarlBar said:

 

If your including Xeons. Then EPYC still puts intel down. Which is the problem. it isn't just about cost. At any given performance level you can find an AMD part that is competitive on raw performance with intel. the fact that they're significantly cheaper is just pure gravy at that point.

 

Which completely ignores that fact that Server CPU's aren't allways the best choice either for someone. Even AMD adds a fairly sharp markup to their server parts. Whilst i'd have to double check the benchmarks to pricing comparison from memory i'm pretty sure Intel's HEDT processors offer equal performance for lower cost than AMD Server parts, (despite AMD having the performance edge in many situations with Threadripper at a significantly lower cost as well).

 

Then again you don't seem to understand what server parts are for, a lot of Xeons actually get their butts kicked by the Intel HEDT lineup. Yet companies still use them, ditto for EPYC compared to AMD's HEDT lineup.

 

Server parts aren't really about maximising performance. They're about scalability, specialised support from the CPU manufacturer and also specialised complementary hardware/software from other vendors that only works with server grade hardware. You only pick server grade stuff when you need that kind of thing. performance is rarely the primary metric in those use cases, (which is why the markup exists, cost to performance isn't the main thing people care about there), and only the highest end SKU's actually offer any performance edge over consumer and HEDT parts. And even then they can end up performing worse depending on the application.

1) can u show me benchmarks of epyc outperforming the 8180m, im sure those cases arent many

2) his point is performance crown sells the rest of the stack, my point is price/performance is what sold the ryzens. do you disagree?

3) price is not a concern of "performance crown sells the rest of the stack" argument (which i disagree with)

4) server parts are about performance, scalability is part of the performance

[justpoet]

I just want to see ECC and similar "reliability" tech on "normal" platforms again as the norm, since with the core count and raw performance numbers coming from "normal" platforms now, they're far more likely to be used as workstations/SBSs than "workstation" or "server" only products are.  That will also translate into a better experience for everybody (stability, security, etc, are all helped by these techs).  Thankfully, with Ryzen, AMD has had support for ECC, but it looks like only a couple motherboards will allow use of that ECC compatibility.

ASUS Pro WS X570-ACE
ASRock X470D4U
 

[Brooksie359]

12 hours ago, Neftex said:

here we go again. ryzen wasnt top end product. intel pretty much held all performance crowns regardless. only reason ryzen was successful was that it was significantly cheaper

 

tldr: it wasnt some performance crown that sold ryzens. it was price/performance

 

unfortunately we dont see that big price/performance difference in GPUs and that is exactly why we dont see amd adoption there. and the new gpus dont seem to change much

That's a bunch of crap. When ryzen was released Intel had what a 10 core as the max on a consumer platform? Ryzen came out at 8 cores and threadripper had 16 cores so in multithreaded workloads they were top of their class. If you wanted the best multithreaded performance on the mainstream desktop platform ryzen was it. If you wanted the best multithreaded performance on the enthusiasts platform threadripper was it.

[cj09beira]

2 hours ago, Neftex said:

is the price/performance difference as significant as in cpu segment? didnt think so, also as mentioned mining was a thing back then

its not half the price for the same performance but its like 40% faster for the same price (it used to be a bit more expensive, something like 10%) which is huge (ps vs 1050 ti)

2 hours ago, justpoet said:

I just want to see ECC and similar "reliability" tech on "normal" platforms again as the norm, since with the core count and raw performance numbers coming from "normal" platforms now, they're far more likely to be used as workstations/SBSs than "workstation" or "server" only products are.  That will also translate into a better experience for everybody (stability, security, etc, are all helped by these techs).  Thankfully, with Ryzen, AMD has had support for ECC, but it looks like only a couple motherboards will allow use of that ECC compatibility.

ASUS Pro WS X570-ACE
ASRock X470D4U
 

there are some that work just fine with it, its just not advertised, you have to ask around to know which ones work

seen the asus Ws board page, looks pretty good, its like a mix of the old tuf with enterprise features

[corsairian]

Should have kept my stock with AMD. I sold it when it was up. Shouldn't have lol. Guess I can always buy more.

[mr moose]

5 hours ago, Neftex said:

 

2) his point is performance crown sells the rest of the stack, my point is price/performance is what sold the ryzens. do you disagree?

No in the absolute sense you want it to be, The price benefit to AMD is just gravy as Carl pointed out. 

5 hours ago, Neftex said:

3) price is not a concern of "performance crown sells the rest of the stack" argument (which i disagree with)

Price is always a concern.

5 hours ago, Neftex said:

4) server parts are about performance, scalability is part of the performance

Server parts are about reliability,  doing lots of tasks without fail rather than a few tasks really fast.   There is a reason why server grade parts don't just get upgraded when anew faster part comes out and why you'll see a lot of xeons locked to lower core clocks.

 

 

[leadeater]

8 hours ago, Neftex said:

1) can u show me benchmarks of epyc outperforming the 8180m, im sure those cases arent many

Sure there's actually plenty, but that's getting pretty off topic plus you should have checked benchmarks and reviews before saying this. You don't even compare a 7601 to an 8180 anyway, people like to but these actually occupy very different product segments in the server market. Why would I pay a huge price premium for an 8 socket compatible processor and compare it to a 2 socket processor let alone actually use 8180 in a 2 socket configuration. Xeon Gold 6152 is almost a third the price, only 4 less cores, much less power and still support 4 sockets to boot. Can buy two complete servers for what it would cost for one 8180 and have 32 more cores.

 

8 hours ago, Neftex said:

4) server parts are about performance, scalability is part of the performance

Server parts have and always will be about use case and best fit for purpose. Performance is often traded and offset to account for other more important factors. Performance profiles also don't align to high model more expensive part is better either.

 

100 lower performance servers are better than 80 high performance ones in a general VMware vSphere cluster for VM hosting, of non specialty servers i.e. standard application and web servers not database servers.

 

But all of this is why Xeons aren't even relevant.

[Unixsystem]

13 hours ago, Princess Luna said:

Intel's secret sauce? Is that a joke?

They did have a secret sauce: their superior process node.

 

The problem is that no matter how good your sauce is, you need to change it up every once in a while or it gets a bit stale. But Intel keeps overcooking their new sauce and meanwhile the other restaurant down the street is cranking out fancy new sauces like a secret sauce factory. 

[MadDuke]

I think people are forgetting that price/performance is a huge factor, but so is performance/watt.  

 

When you have 2 products from different companies where one has a small performance advantage in most tasks, for a slightly higher consumption, but also costs 300% more. 

If hunting only raw CPU performance, then you really go (unless tied down by some previous business decision) with the just slightly slower ones.  The differences are so negligible that rack space is not a factor in this comparison.

 

Ace in AMD sleeve in this space was also the fact that it freaking offered 128 PCIe lanes. So if a datacenter is not going after only pure CPU performance but other stuff like GPUs....  Then the rack space becomes another factor for AMD.   Not to mention that the newest, 7nm 64 core 128 thread beast can actually complement and drive those GPUs without breaking a sweat.  I think Google Stadia is a good example of that (from recent times).

 

 

 

 

[SupremeGOAT]

"Impressed"? that's a weird way to say "Scared Shitless"

[CarlBar]

2 hours ago, leadeater said:

Sure there's actually plenty, but that's getting pretty off topic plus you should have checked benchmarks and reviews before saying this. You don't even compare a 7601 to an 8180 anyway, people like to but these actually occupy very different product segments in the server market. Why would I pay a huge price premium for an 8 socket compatible processor and compare it to a 2 socket processor let alone actually use 8180 in a 2 socket configuration. Xeon Gold 6152 is almost a third the price, only 4 less cores, much less power and still support 4 sockets to boot. Can buy two complete servers for what it would cost for one 8180 and have 32 more cores.

 

Server parts have and always will be about use case and best fit for purpose. Performance is often traded and offset to account for other more important factors. Performance profiles also don't align to high model more expensive part is better either.

 

100 lower performance servers are better than 80 high performance ones in a general VMware vSphere cluster for VM hosting, of non specialty servers i.e. standard application and web servers not database servers.

 

But all of this is why Xeons aren't even relevant.

 

Exactly. In fact if i recall isn't the only practical difference on a silicon level between a 2990WX and the top end EPYC skew the disabling of some bits of silicon to disable certain features?  Certainly AMD is capable of producing a 32 core EPYC Zen1 based solution with clocks very similar to a TR part on demand for any customer that wants it and is willing to pay for the custom part. And yet you don't see that. And thats because pure raw flops per CPU core isn't as big a deal as people tend to think. Thats not to say server don't care about performance, but what they define as performance isn't "how many flops can it do for X money's". They care about meeting a certain level of capability with very specific software. And given how different softwares have different ideal operating environments in terms of core count, memory bandwidth, (also storage, entworking, and tons of other similar non-CPU system aspects), and capacity requirements, and different abilities to work with multi-socket vs multi-rack vs multi-cabinet setups the ideal configuration for a particular application varies hugely to the point where the choice of hardware configuration is going to be influance far more by what software your going to be using than the theoretical capabilities of a given CPU.

 

It's when you get to consumer and HEDT uses where you can have a huge variety of different softwares interacting in a huge variety of ways that you start to get to a more clear cut delineation between things because suddenly you need a hardware setup that works well across a wide variety of use cases, and where raw multi-purpose CPU performance becomes a more important performance metric.

[Mihle]

I think server CPUs is usually better binned when it comes to perf/watt, and has lower clocks to get better perf/watt.

[Neftex]

6 hours ago, leadeater said:

Sure there's actually plenty, but that's getting pretty off topic plus you should have checked benchmarks and reviews before saying this. You don't even compare a 7601 to an 8180 anyway, people like to but these actually occupy very different product segments in the server market. Why would I pay a huge price premium for an 8 socket compatible processor and compare it to a 2 socket processor let alone actually use 8180 in a 2 socket configuration. Xeon Gold 6152 is almost a third the price, only 4 less cores, much less power and still support 4 sockets to boot. Can buy two complete servers for what it would cost for one 8180 and have 32 more cores

youre right, you can compare it to different config making intels performance even better. so wheres the AMD performance crown?

 

again, im not shitting on ryzen, epyc whatever. im just saying performance crown isnt the reason it sold so well

[leadeater]

3 hours ago, Neftex said:

youre right, you can compare it to different config making intels performance even better. so wheres the AMD performance crown?

In the many benchmarks where they are faster than the 8180 by up to 50%, think there some that is even more than that. Like I said you could go look before making a statement that may not be correct if you haven't done the required research to make it.

 

I was going to post them on the earlier comment but their were getting to be over 10 source links so I gave up as it was pointlessly long and easily found information. Here is just one example, https://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/21.

 

But the above bechmarks and many like those are largely irrelevant because these use cases are not that common in businesses like this, I'm part of a team that manages large VMware clusters and C-Ray/POV-Ray means exactly nothing here. Not just that but for the scientific workloads the data set completely changes the performance characteristics, there really is no simple or accurate way of saying one server CPU architecture is better than another because you could be comparing a drag car to an F1 car and wondering why both suck at rally.

 

And specifically EPYC did not sell well, which again why EPYC and Xeons are not relevant here.

[Neftex]

2 hours ago, leadeater said:

In the many benchmarks where they are faster than the 8180 by up to 50%, think there some that is even more than that. Like I said you could go look before making a statement that may not be correct if you haven't done the required research to make it.

 

I was going to post them on the earlier comment but their were getting to be over 10 source links so I gave up as it was pointlessly long and easily found information. Here is just one example, https://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/21.

thats 8176 not 8180 - 400mhz difference on each of the 28 cores, and you link to the FP benchmark? as you yourself pointed out there are other intel configurations than 8180 and those easily beat epyc on FP...

[GloriousPain]

On 6/27/2019 at 12:08 AM, Neftex said:

tldr: it wasnt some performance crown that sold ryzens. it was price/performance

I mean it depends; for gamers this is true; for multithreaded workloads Ryzen has been a smash hit taking most performance crowns to be honest. Especially before the 8000 series by intel... When it was say the 1700-1800x versus the 7700k, Ryzen was undeniably the king of performance in basically every multithreaded workload. The 8700k and 2700(x) were a bit more even though... with 9900k taking back the performance crown but with a hefty pricing increase.  And similar things can be said for threadripper vs Intel's HEDT line ups.

 

Saying Ryzen never had any crown and only value... really is only true for gaming/single core performance. Otherwise intel and ryzen have traded workstation crowns for a little bit now. 

[Neftex]

3 minutes ago, GloriousPain said:

I mean it depends; for gamers this is true; for multithreaded workloads Ryzen has been a smash hit taking most performance crowns to be honest. Especially before the 8000 series by intel... When it was say the 1700-1800x versus the 7700k, Ryzen was undeniably the king of performance in basically every multithreaded workload. The 8700k and 2700(x) were a bit more even though... with 9900k taking back the performance crown but with a hefty pricing increase.  And similar things can be said for threadripper vs Intel's HEDT line ups.

 

Saying Ryzen never had any crown and only value... really is only true for gaming/single core performance. Otherwise intel and ryzen have traded workstation crowns for a little bit now. 

you compare 4 core with 8 core in multithread ofc it looks great for the 8 core. if you want multithread then look at all the options, intel clearly had other better performing options but that performance crown didnt matter, why? ofc because of the PRICE/PERFORMANCE - my whole point

[GloriousPain]

Just now, Neftex said:

you compare 4 core with 8 core in multithread ofc it looks great for the 8 core.

I mean if you wanna compare their main stream platforms it's only fair. Intel didn't make above 4 cores; infact the general consensus is that the Ryzen 1700(and above obviously) is the only reason the 8700k (and sucessors) have more than 4 cores... It really was intel admitted defeat. 

 

2 minutes ago, Neftex said:

if you want multithread then look at all the options, intel clearly had other better performing options

I mean if you wanna look at say Skylake-E then you have to compare it to Threadripper, which AGAIN held the title... And if you're like "But intel has their Xeon with even more cores" then again look at EPYC.  

 

3 minutes ago, Neftex said:

that performance crown didnt matter, why?

It did matter, just not to gamers... But to people hosting small servers, video production, cad work, etc. it mattered greatly. Again general consensus is that it's the reason why intel quit doing Quad cores as their upper limit of the mainstream platform. 

 

There's a reason why from 2006 until 2017... Intel kept Quad core as the mainstream limit; Yet the moment Ryzen gained popularity due to it's performance crown. They upped it to 6 cores with the 8700k. And when AMD's Zen+ 2700 repeated the trend. Intel AGAIN upped their mainstream core limit to 8.  They upped their core counts TWICE in a row... After not doing it in ~10 generations. There's a reason for it, it's because they lost their crown and wanted it back. It mattered heavily. 

 

9 minutes ago, Neftex said:

 because of the PRICE/PERFORMANCE - my whole point

I mean overall Ryzen had the Price/Performance in the bag. That's obvious AMD's profit margins are anorexic in comparison to Intel's but to argue that Ryzen hasn't held any crowns or made any differences since it's release is just down right ignorant.

[Drak3]

5 minutes ago, GloriousPain said:

infact the general consensus is that the Ryzen 1700(and above obviously) is the only reason the 8700k (and sucessors) have more than 4 cores...

Which is blatantly wrong as Coffee Lake roadmaps have shown hexacores prior to anyone knowing anything about Ryzen outside of AMD.