Linus plays with a 64 core epyc CPU.

[repman244]

No it won't. Seriously. Not possible.

Did you just ignore the link I posted? It's basic math which proves you are wrong.
Not to mention you haven't provided any data to back your claims.

 

[Sybaris_Caesar]

Can someone explain why Crysis ran so poorly? It was like slideshow. Then he flipped the switch and it ran better (comparatively to the first) but still slow af.

 

[lexluthermiester]

$7219

Unless you're rich, you're not buying one...

 

[EzioAs]

Can someone explain why Crysis ran so poorly? It was like slideshow. Then he flipped the switch and it ran better (comparatively to the first) but still slow af.

CPU rendered. He said it in the video

 

[lexluthermiester]

CPU rendered. He said it in the video

That was funny and yet impressive!

 

[HD64G]

Why would temperatures be too high? Are Ryzen 3700X temperatures too high?

With brief 7nm advantage AMD clearly wins in number of cores per socket.
The cores itself however are way slower than I hoped for.

What this means is that when someone is looking for an ~8, 16 or 24-core server, Intel is still competitive - despite an obvious technological and node handicap.

I really hoped AMD will show a more flexible lineup this year to put more pressure on Intel. ;-)


No it won't. Seriously. Not possible.


But these are the same dies that land in 3700X and 3900X, right?
So why aren't there any faster 8- or 12-core EPYCs? What is stopping them?

AMD Rome Second Generation EPYC Review: 2x 64-core Benchmarked

www.anandtech.com

So, you were saying?

 

[Hellfire]

Why would temperatures be too high? Are Ryzen 3700X temperatures too high?

Because most servers, especially the 1u Or 2u servers are passively cooled, 1u servers are only 44.45 mm in height and don't typically have active cooling.

You have to kick the notion of what you'd see in a PC as servers and pcs do not work the same in how they're set up. It's the same with Intel, you don't get desktop cpu clock speeds on their sercers either and Intel desktop CPUs are known for higher clock speeds than AMD are.

 

[biffzinker]

CPU rendered. He said it in the video

Here is what was mentioned in the video.

GitHub - google/swiftshader: SwiftShader is a high-performance CPU-based implementation of the Vulkan graphics API. Its goal is to provide hardware independence for advanced 3D graphics.

SwiftShader is a high-performance CPU-based implementation of the Vulkan graphics API. Its goal is to provide hardware independence for advanced 3D graphics. - google/swiftshader

github.com

 

[notb]

Did you just ignore the link I posted? It's basic math which proves you are wrong.
Not to mention you haven't provided any data to back your claims.

You haven't either.
Servers are bought for a particular systems (projects). They work for the time planned. And afterwards, they're "infinitely" recycled for other systems for as long as it's economically sensible - which is usually a long time.

No one throws away expensive hardware. And no one replaces production systems with every new gear - even if it's much faster. Hardware cost is just part of the equation. Implementation can cost way more. You can't just move to a faster CPU every year. It's not a home PC built for gaming.

AMD Rome Second Generation EPYC Review: 2x 64-core Benchmarked

www.anandtech.com

So, you were saying?

So you can neither write longer posts nor read them...
The question was: if Zen2 EPYC and Ryzen are built from the same die, why is there such a huge difference in clocks?
The argument about power curve is true for larger CPUs but not in general. Low core CPUs (8-16) will be used not in large cloud datacenters, but in standalone systems.
These systems could handle higher power consumption and heat. And would be much faster as a result.

On the Blue side you can get CPUs like Xeon Gold 6244 (8C/16T 3.6/4.4GHz) and Xeon Gold 6242 (16C/32T 2.8/3.9GHz). These clocks are very close to what you can get in the consumer LGA2066 or even LGA1151.

So the question stands: why isn't there an 8 or 16 core EPYC clocked like the 3700X?
I know they would pull way more power and produce way more heat. But servers can handle it. And the extra performance would be welcome.

Because most servers, especially the 1u Or 2u servers are passively cooled, 1u servers are only 44.45 mm in height and don't typically have active cooling.

Whaaaaat...??? :-/
Seriously... I don't expect you to know how a 1U server looks. But it's not that difficult to find online...

 

[TheMadDutchDude]

I have a 1U server here and can assure you it is anything but passively cooled. Eight super high pitched, screaming Deltas (doubled up to 16) beg to differ.

 

[notb]

I have a 1U server here and can assure you it is anything but passively cooled. Eight super high pitched, screaming Deltas (doubled up to 16) beg to differ.

Seriously... it's still 100-200W for mainstream server CPUs in 1S config. I'd expect everyone on this forum to know it's not easy to do passively - even in a large ATX case.
Shame on 2 senior members who liked that comment...

 

[Hellfire]

Seriously... it's still 100-200W for mainstream server CPUs in 1S config. I'd expect everyone on this forum to know it's not easy to do passively - even in a large ATX case.
Shame on 2 senior members who liked that comment...

Funny our HPE servers at work are passively cooled..

 

[TheMadDutchDude]

No active airflow?? No way. XD

 

[Hellfire]

No active airflow?? No way. XD

Not on the racks no, the room is chilled but the two servers I'm thinking about don't have any fans in them. (Except maybe the PSU)

However, back to the original point, i'll concede poor choice of works but cooling and temps in servers are a big deal, they don't have as much room for airflow like desktop PC's and servers need to be efficient in both cooling & power usage, this is why they don't push their clock speeds as they have no need to really, it's about efficiency, a Ryzen 3900x etc is hardly about efficiency.

 

[lexluthermiester]

I have a 1U server here and can assure you it is anything but passively cooled. Eight super high pitched, screaming Deltas (doubled up to 16) beg to differ.

I have never seen a 1U or 2U blade that didn't have active and often very noisy cooling..

 

[phill]

I wish, I will be getting my hands on the servers I am decommissioning though, Pair of HP Proliant DL360 G9's with a pair of Xeon E5-2640 v3 in it. so going to steal them ;-)

I imagine these are coming in useful at Intel HQ right now

View attachment 132886

I've been rather lucky with servers from work, I mean I do have one or two...

Spoiler: One or two...

I can tell you with hand on heart, none of these are passively cooled at all. I have some quad cores in there, 135w models, jesus the fans can be far from silent and god damn do they pull the power as well.. I've seen 500w+ from my wattage meter....

With help from a crazy mate of mine (@TheMadDutchDude), I have a pair of these...

I do have another server that will take the same versions of CPU, V3 and V4, but I'm waiting for prices to drop, since I need a few things for that one... I might put the quads in but we'll see
He said there was a surprise as well so when I saw this...

Spoiler: Just a little overkill....

No one is replacing servers before the end of amortization period or breakdown. That's not cheaper in any way.

As for the CPUs: the 64-core is clearly stealing the show for large computing clusters and so on. AMD will have an advantage in this - not very big but highly prestigious - niche.

But I feel like this launch is a bit of a let down in the lower part of the lineup.
Where are the high-frequency models? There's not a single 8-16 core model that gets even close to Ryzen 3000 speeds. That's just bizarre. :-/

I'm sure there are some reasons to change as we've gone from X58 model of servers to X299, massive step, good efficiencies and so on, but still not a patch on AMD, not now Oh and it was a little bit tongue in cheek as well

Not on the racks no, the room is chilled but the two servers I'm thinking about don't have any fans in them. (Except maybe the PSU)

However, back to the original point, i'll concede poor choice of works but cooling and temps in servers are a big deal, they don't have as much room for airflow like desktop PC's and servers need to be efficient in both cooling & power usage, this is why they don't push their clock speeds as they have no need to really, it's about efficiency, a Ryzen 3900x etc is hardly about efficiency.

I have never seen a 1U or 2U blade that didn't have active and often very noisy cooling..

From what I see from our server room, massive cooling units, 18C set and then things like this sat inside the servers to pull in the cool air....

They can be anything but quiet... As mentioned above, I've seen the wattage go from 100w + to over 300w + when these kick in full speed and my word you know it!!

I enjoy normal home PC hardware and even more so now, server grade stuff It's all fun to play with

 

[HD64G]

So you can neither write longer posts nor read them...
The question was: if Zen2 EPYC and Ryzen are built from the same die, why is there such a huge difference in clocks?
The argument about power curve is true for larger CPUs but not in general. Low core CPUs (8-16) will be used not in large cloud datacenters, but in standalone systems.
These systems could handle higher power consumption and heat. And would be much faster as a result.

On the Blue side you can get CPUs like Xeon Gold 6244 (8C/16T 3.6/4.4GHz) and Xeon Gold 6242 (16C/32T 2.8/3.9GHz). These clocks are very close to what you can get in the consumer LGA2066 or even LGA1151.

So the question stands: why isn't there an 8 or 16 core EPYC clocked like the 3700X?
I know they would pull way more power and produce way more heat. But servers can handle it. And the extra performance would be welcome.

Can you explain to me then why all -except for you- who their job is messing with servers prefer more cores per chip, better efficiency, security and higher vfm instead of clocks? And AMD beats heavily Intel in all those aspects, don't you think?

 

[GreiverBlade]

$7219

still cheaper than a dual 24/48(48/96) Xeon Platinum 8168 that doesn't reach the score of that 7k 64/128 Epyc, and the price is very interesting for its intended target "customer"

That price alone gives headaches and nausea all together but when you compare it to the intel rival, its a cheap offer.
In order to match performance Intel has to throw 2x Xeon Platinum 8280 (2x 10000$). Not to mention the power requirements... 225W vs 2x205W.
And more likely is a 225W vs 450+W because of the different standards in measurement by AMD and Intel.

EXACTLY! basically that price put in perspective it's ~3k cheaper ... (even cheaper the dual 8168 seen in the result for R20, similarly priced to one of the 2 needed to ... not even reach the same score )

as for TDP... aherm... we already know the hotter one (true hotter one, the one that measure the TDP at base clock instead of full boost like his concurrent )

That was nice.

Also this was worthy of a screen cap.

View attachment 132887

well .... the right part is Intel iirc

 

[lexluthermiester]

They can be anything but quiet... As mentioned above, I've seen the wattage go from 100w + to over 300w + when these kick in full speed and my word you know it!!

So much for passive cooling!
(no diss on anyone intended, just think it's funny)

 

[phill]

So much for passive cooling!
(no diss on anyone intended, just think it's funny)

Well you know mate

I think though from my limited knowledge, that Blade Servers are different to these full sized beasts.. These are monsters and I think the Blades are perhaps as long but less than 1/2 maybe even a 1/3 or 1/4 of the size
None of the servers I have at home are passive, by a long shot

 

[notb]

Funny our HPE servers at work are passively cooled..

I'll be gentle and call that unlikely.

Yes, low-power servers exist. I've seen 1U units with a Pentium G4560 and 4 HDDs. A setup like that could work without fans in an ATX case.
But rack units are extremely cramped and cooling them passively is almost impossible.

And of course having a passively cooled server really goes against the whole idea of security and continuous operation.

Can you explain to me then why all -except for you- who their job is messing with servers prefer more cores per chip, better efficiency, security and higher vfm instead of clocks? And AMD beats heavily Intel in all those aspects, don't you think?

This really shows you don't know much about enterprise servers. There's a huge demand for low-core machines.
And you must be really confused why AMD offers an 8-core EPYC. It's like a next gen PlayStation. What's the point?

There are a few decent reasons why low-core servers make sense, but there's really just one that needs to be mentioned here: per-core licensing.
Sometimes you buy an 8-core system because more cores would drastically increase software cost, so the additional performance isn't worth it. End of story.

 

[HD64G]

This really shows you don't know much about enterprise servers. There's a huge demand for low-core machines.
And you must be really confused why AMD offers an 8-core EPYC. It's like a next gen PlayStation. What's the point?

There are a few decent reasons why low-core servers make sense, but there's really just one that needs to be mentioned here: per-core licensing.
Sometimes you buy an 8-core system because more cores would drastically increase software cost, so the additional performance isn't worth it. End of story.

You still seem to ignore IPC performance as well as price difference and power consumption costs in the long run even for a product running almost full load 24/7. Zen2 CPUs are better in every metric by far and you try to convince anyone that Intel is a better choice even for server usage where the gap is much bigger than in desktop usage.

 

[Hellfire]

I'll be gentle and call that unlikely.

Yes, low-power servers exist. I've seen 1U units with a Pentium G4560 and 4 HDDs. A setup like that could work without fans in an ATX case.
But rack units are extremely cramped and cooling them passively is almost impossible.

And of course having a passively cooled server really goes against the whole idea of security and continuous operation.

I checked today, you're right, they have two tiny fans in the back but our Xeon D-1513N's have literally near no cooling on the server. and none on the CPU direct

 

[biffzinker]

I checked today, you're right, they have two tiny fans in the back but our Xeon D-1513N's have literally near no cooling on the server. and none on the CPU direct

I looked up the Xeon D-1513N on Intel's Ark page, and the rated TDP is only 35 watts. Being passively cooled wasn't necessarily wrong.

Intel® Xeon® Processor D-1513N (6M Cache, 1.60 GHz) Product Specifications

Intel® Xeon® Processor D-1513N (6M Cache, 1.60 GHz) quick reference guide including specifications, features, pricing, compatibility, design documentation, ordering codes, spec codes and more.

ark.intel.com

 

[remixedcat]

Why would anyone? Get computational tasks done faster.

This. Media encoding specially. I could use this EPYC but at that price I'd have to sell a kidney.