Review AMD 4th-Gen EPYC Genoa 9654, 9554, and 9374F Review: 96 Cores, Zen 4 and 5nm Disrupt the Data Center

[Admin]

We put AMD’s fourth-gen EPYC Genoa 9654, 9554, and 9374F through the benchmark paces against Intel’s Ice Lake to see which server chips come out the winner.

AMD 4th-Gen EPYC Genoa 9654, 9554, and 9374F Review: 96 Cores, Zen 4 and 5nm Disrupt the Data Center : Read more

 

[Roland Of Gilead]

Okay, for those more knowledgeable than me (I'm not really into server tech), how is it that Intel is so far behind in terms of core count with these systems? Looking at some of the benches (and I might as well be blind in both eyes and using a magnifying glass to scroll through the data!) it seems to me that if Intel were able to increase core count, that they would be comparable in performance to the AMD counterparts? What gives?

Intel have taken the performance crown with ADL and Raptor in the consumer market, but on the bigger scales can't get close.

 

[rasmusdf]

Well, basically AMD has mastered the art and technology of connecting lots of small chips together. While Intel still has to make chips as one big lump. Big chips are harder and more expensive to produce. Plus when combining small chips you can always and easily add more.

Intel might be competing right now in the consumer space - but they don't earn much profit on their expensive to produce CPUs.

Additionally - AMD is cheating a bit and is several chip nodes ahead in production process - while Intel is struggling to get past 10 nm, AMD is on what, 4 nm? Because of TSMCs impressive technology leadership and Intels stubborness.

 

[gdmaclew]

Strange that Tom's mentions that DDR5 support for these new Data Center CPUs is a "Con" but they don't mention it in yesterday's article of Intel's new CPUs using the same DDR5.

But then maybe not, knowing Tom's. It's just so obvious.

 

[SunMaster]

Intel have taken the performance crown with ADL and Raptor in the consumer market, but on the bigger scales can't get close.

Rumour has it that AMD prioritize less on the consumer market, and more on the server marked. So if "one core to rule them all" it means Zen4 is a core designed primarily for server chips. Whether Alder Lake took the "performance crown" or not is at best debateable, as is Raptor Lake vs Zen4 if power consumption is taken into consideration.

Intel does not use anything equivalent of chiplets (yet). The die size of raptor lake 13900 is about 257 square mm. Each Zen4 ccd is only 70 square mm (two in a 7950x). That gives AMD a tremendous advantage in manufacturing and cost.

See

View: https://www.youtube.com/watch?v=oMcsW-myRCU&t=2s
for some info/estimates on yields and cost of manufacturing.

 

[bitbucket]

You have forgotten the face of your father.

Most likely multiple reasons.
1) Intel struggled for a long time trying to reach the 10nm process

This delayed entire product lines for a couple of years and ultimately led to Intel outsourcing some production to TSMC which wasn't struggling to shrink the fabrication process
AMD had already been using TSMC as AMD had sold off their manufacturing facilities years before

2) AMD moved to a chiplet strategy long before Intel, which I don't believe has a product for sale using chiplets yet, not sure though

Large monolitic, high core-count CPUs are harder to make than smaller lower core-count CPUs

• - An example is AMD putting two 8-core chiplets in a package (plus IO die) for a product that has 16 cores
• - Intel has recently countered this by going with heterogeneous cores in their CPUs; a mix of bigger/faster and smaller/slower cores
• - I don't believe that the heterogeneous core strategy has been implemented in servers products yet

 

[InvalidError]

The DIMM slot fragility issue could easily be solved or at least greatly improved by molding DIMM slots in pairs for lateral stability and sturdiness.

 

[Roland Of Gilead]

You have forgotten the face of your father.

Hile Gunslinger

 

[Roland Of Gilead]

Whether Alder Lake took the "performance crown" or not is at best debateable, as is Raptor Lake vs Zen4 if power consumption is taken into consideration.

Fair point

 

[GustavoVanni]

It is just me or do you guys also think that AMD can fit 24 CCDs in the same package in the not so distant future?

Sure there's some small SMDs in the way, but it should be doable.

Just imagine one of those with 192 ZEN4 cores ou 256+ ZEN4c cores.

Maybe with ZEN5?

 

[-Fran-]

Intel's only bastion seems to be accelerators and burn as much money as they can on adoption, even worse than AVX512.

I just looked at the numbers of OpenSSL and I just laughed... Intel is SO screwed for general purpose machines. Their new stuff was needed in the market last year.

Regards.

 

[tamalero]

Rumour has it that AMD prioritize less on the consumer market, and more on the server marked. So if "one core to rule them all" it means Zen4 is a core designed primarily for server chips. Whether Alder Lake took the "performance crown" or not is at best debateable, as is Raptor Lake vs Zen4 if power consumption is taken into consideration.

Intel does not use anything equivalent of chiplets (yet). The die size of raptor lake 13900 is about 257 square mm. Each Zen4 ccd is only 70 square mm (two in a 7950x). That gives AMD a tremendous advantage in manufacturing and cost.

See

View: https://www.youtube.com/watch?v=oMcsW-myRCU&t=2s
for some info/estimates on yields and cost of manufacturing.

Sounds exactly like when they were doing the ATHLON.

They concentrated on OPTERON first.

 

[The Historical Fidelity]

Rumour has it that AMD prioritize less on the consumer market, and more on the server marked. So if "one core to rule them all" it means Zen4 is a core designed primarily for server chips. Whether Alder Lake took the "performance crown" or not is at best debateable, as is Raptor Lake vs Zen4 if power consumption is taken into consideration.

Intel does not use anything equivalent of chiplets (yet). The die size of raptor lake 13900 is about 257 square mm. Each Zen4 ccd is only 70 square mm (two in a 7950x). That gives AMD a tremendous advantage in manufacturing and cost.

See

View: https://www.youtube.com/watch?v=oMcsW-myRCU&t=2s
for some info/estimates on yields and cost of manufacturing.

From a business standpoint, I think what AMD is doing is smart, spend majority of R&D on the market segment that yields the majority of profit IE servers and then glue a 2x L3 cache on top of the server chiplets and yield a gaming optimized consumer processor.

 

[prtskg]

Last quarter Intel sold $4.2B worth of Xeon without any profit and yet lost market share. Seems like this quarter will be worse. Things should improve for Intel from 2023 as Sapphire rapids is to be launched on Jan10, I think.

 

[Kamen Rider Blade]

Strange that Tom's mentions that DDR5 support for these new Data Center CPUs is a "Con" but they don't mention it in yesterday's article of Intel's new CPUs using the same DDR5.

But then maybe not, knowing Tom's. It's just so obvious.

Like many have said before, different/easier "Requirements" for Intel to get positive coverage, they have it easier at Tom's Hardware.
If AMD had same requirement, it's a negative.
If Intel has same requirement, it's a positive.

God forbid we have the same standards for both.

 

[Deleted member 14196]

Like many have said before, different/easier "Requirements" for Intel to get positive coverage, they have it easier at Tom's Hardware.
If AMD had same requirement, it's a negative.
If Intel has same requirement, it's a positive.

God forbid we have the same standards for both.

This. So true

I am so sick of the Intel bias. And no matter how much Tom’s hardware pushes it I will never buy them ever.

The fact that sites like this pushing intel so hard really turns me off and they’ve made it so that I will never buy Nvidia or Intel products ever again

we all have thread ripper’s now at work. There’s no need for any Intel hardware. Every workstation can run tons of virtual machines super fast Intel have nothing to touch it for the price performance and value

 

[Kamen Rider Blade]

This. So true

I am so sick of the Intel bias. And no matter how much Tom’s hardware pushes it I will never buy them ever.

The fact that sites like this pushing intel so hard really turns me off and they’ve made it so that I will never buy Nvidia or Intel products ever again

we all have thread ripper’s now at work. There’s no need for any Intel hardware. Every workstation can run tons of virtual machines super fast Intel have nothing to touch it for the price performance and value

I'm the same way, I'm done with nVIDIA & Intel.
Every machine I buy & recommend will be "All AMD" from now on.

I'm sick & tired of the BS from Intel & nVIDIA.

Especially Intel's convicted Monopolists actions that nearly bankrupted AMD.

Now it's time for Intel to suffer, so much that they need to be bled of money until they have to sell off their Foundary Manufacturing into it's own company.

Then suffer through at least one decade, preferably 2-3 decades of rebuilding Intel from the ground up as a design only firm.

Intel needs to wipe off that typical "Intel Arrogance" and learn humility, humbleness, and honor.

 

[scorched_675]

Okay, for those more knowledgeable than me (I'm not really into server tech), how is it that Intel is so far behind in terms of core count with these systems? Looking at some of the benches (and I might as well be blind in both eyes and using a magnifying glass to scroll through the data!) it seems to me that if Intel were able to increase core count, that they would be comparable in performance to the AMD counterparts? What gives?

Intel have taken the performance crown with ADL and Raptor in the consumer market, but on the bigger scales can't get close.

Hi Roland.. I am brand new here, and I first off, I am biased as I work at AMD as a customer facing, not chip design, but I will try to answer as honestly as possible my opinion.
First AMD and Intel have approached x86 in two different ways. Initially, it could be said that we extracted more cores using our chiplet tech, while intel extracted more IPC, instructions per cycle, and many times with higher baseline frequency. We were heavily criticized by Intel for using this chiplet function. But, then they attempted to glue to 8000s together to get the AP chips. And now Intel's new sapphire rapids that is coming is pretty much chiplet style and we expect them to have a decent jump in cores and massive jump in performance over their prior generation due to this and their IPC improvements. Rumors are it will get to 60 cores maybe a bit more on some special chips. And we hear the follow on procs will finally start to have the core counts. But just remember cores are not everything, its just one of the factors that contributes to the total computational capability of the processors.

Lucky for us, as that Genoa is 4th gen Zen, and you might have seen our gen 4c announced, which will bring an even larger density up to 128 cores per socket in Q1 23'. The IPC probably won't be that different, but due to core counts it keeps us ahead in the game. Where our 4th gen zen cores really shines is power execution efficiency. Not only do we have the fastest supercomputer with our 3rd gen Milan and MI250, we also own #1 in the green 500. We even have calculators that can show how much carbon we can prevent by using us over the competition. And finally we have a good Zen roadmap to keep us going for a very long time but we expect Intel will show up with some big performance numbers even without the core count.
Just my 2 cents, probably not worth that much though.

Cheers,
Scorched

 

[scorched_675]

Sounds exactly like when they were doing the ATHLON.

They concentrated on OPTERON first.

Hi there. A new AMD'r here. My unofficial opinion is that we're not concentrating on either consumer or enterprise. The Zen core's themselves are designed in such a way as to be leveraged by both teams. Take for example our launch of the Ryzen 7000 series, those too are Zen 4's, ddr4 and pcie 5. The only way I can describe what is happening is that Zen is an overlaying term for some underlying functionality that allow each division to tune to what the market needs. In the end, this allowing us to release Ryzen 7000 and EPYC 9000 at the same time because they are both Zen 4. Plus we have RDNA3 just announced and CDNA 3 for our MI series coming, so Lisa Su is making sure we are hitting on all cylinders.
Cheers!

 

[TerryLaze]

Strange that Tom's mentions that DDR5 support for these new Data Center CPUs is a "Con" but they don't mention it in yesterday's article of Intel's new CPUs using the same DDR5.

But then maybe not, knowing Tom's. It's just so obvious.

Being able to read is a disappearing artform it seems.
DDR5 is a pro, having to pay much more for it is a con.

Pros

• +
  First 5nm x86 data center processor
• +
  Unmatched core density
• +
  Performance in both heavily- and lightly-threaded workloads
• +
  Leading-edge connectivity, like PCIe 5.0 and DDR5
• +
  Support for AVX-512, VNNI, BFloat 16
• +
  Support for the CXL interface

Cons

• -
  Early adopter tax for DDR5

 

[PaulAlcorn]

Like many have said before, different/easier "Requirements" for Intel to get positive coverage, they have it easier at Tom's Hardware.
If AMD had same requirement, it's a negative.
If Intel has same requirement, it's a positive.

God forbid we have the same standards for both.

Can you please point me to the Tom's Hardware review of the Intel server chip that uses DDR5? I can't seem to find this biased review.

SPOILER ALERT: It doesn't exist.

 

[PaulAlcorn]

Strange that Tom's mentions that DDR5 support for these new Data Center CPUs is a "Con" but they don't mention it in yesterday's article of Intel's new CPUs using the same DDR5.

But then maybe not, knowing Tom's. It's just so obvious.

Can you please point me to the Tom's Hardware review of the Intel server chip that uses DDR5? I can't seem to find this biased review.

SPOILER ALERT: It doesn't exist.

 

[Roland Of Gilead]

Hi Roland.. I am brand new here, and I first off, I am biased as I work at AMD as a customer facing, not chip design, but I will try to answer as honestly as possible my opinion.
First AMD and Intel have approached x86 in two different ways. Initially, it could be said that we extracted more cores using our chiplet tech, while intel extracted more IPC, instructions per cycle, and many times with higher baseline frequency. We were heavily criticized by Intel for using this chiplet function. But, then they attempted to glue to 8000s together to get the AP chips. And now Intel's new sapphire rapids that is coming is pretty much chiplet style and we expect them to have a decent jump in cores and massive jump in performance over their prior generation due to this and their IPC improvements. Rumors are it will get to 60 cores maybe a bit more on some special chips. And we hear the follow on procs will finally start to have the core counts. But just remember cores are not everything, its just one of the factors that contributes to the total computational capability of the processors.

Lucky for us, as that Genoa is 4th gen Zen, and you might have seen our gen 4c announced, which will bring an even larger density up to 128 cores per socket in Q1 23'. The IPC probably won't be that different, but due to core counts it keeps us ahead in the game. Where our 4th gen zen cores really shines is power execution efficiency. Not only do we have the fastest supercomputer with our 3rd gen Milan and MI250, we also own #1 in the green 500. We even have calculators that can show how much carbon we can prevent by using us over the competition. And finally we have a good Zen roadmap to keep us going for a very long time but we expect Intel will show up with some big performance numbers even without the core count.
Just my 2 cents, probably not worth that much though.

Cheers,
Scorched

Hey there,

Welcome to the forums.

Thanks for the info. Much appreciated.

 

[JarredWaltonGPU]

Strange that Tom's mentions that DDR5 support for these new Data Center CPUs is a "Con" but they don't mention it in yesterday's article of Intel's new CPUs using the same DDR5.

But then maybe not, knowing Tom's. It's just so obvious.

This is an extremely biased (blind even) take, for several reasons.

1. Paul gave these new EPYC chips a 4.5/5 star score. That's about as high as it gets; 5-star reviews have to be "perfect" and that almost never happens.
2. Intel's consumer chips (which is what you're referencing) support both DDR4 and DDR5. If you want maximum performance and are willing to pay for it, go DDR5. If you want a good midrange or budget option, stick with DDR4. AMD does not offer this flexibility with Zen 4. For the record, Raptor Lake also got 4.5/5 stars. I don't see you pointing out the list of power and cooling requirements as being unreasonable.
3. As Paul notes, we haven't seen Intel's competing solution, Sapphire Rapids, which has been delays about a year now. Comparing server and consumer reviews to try for a biased hot-take is also misguided at best, intentionally stupid and misleading at worst.
4. We always need to come up with something (often multiple things) to list as a con. Con doesn't mean "THIS IS THE WORST THING EVER" but rather "This is something to consider." Saying DDR5 costs more is the only con listed, and you're acting as though we slammed the product for it. Again, intentionally biased or misguided: take your pick.

Here's the RPL vs. Genoa review summaries side by side. And you'll notice that it doesn't say "DDR5 Support" but rather "Early adopter tax for DDR5." Again, not the end of the world, but you will spend a lot more money on a Genoa system's memory compared to something that still uses DDR4. And that's okay, because for servers the cost of the hardware often pales in comparison to the software and other aspects of ownership (TCO).

I bet when the competing Intel Sapphire Rapids chips arrive next year, and DDR5 is required and costs a lot of money, that will still get called out in the articles. But until then, try pulling your head out of the sand and exercising some common sense. A high rating with only a single "Con" should immediately tell you the product is good. The consumer Ryzen parts got a 4/5 star rating, largely because the DDR5, chip, and mobo costs become far more significant on the consumer side. 🤔

 

[Kamen Rider Blade]

This is an extremely biased (blind even) take, for several reasons.

It's not like Tom's Hardware didn't gain the reptuation over the internet for Intel bias over the decades w/o reason, you're known globally as "Pro Intel" through & through amongst the tech viewership. Even I've known about it 20+ years ago when I was just a teenager reading Tom's Hardware and reading your sites articles back then.

1. Paul gave these new EPYC chips a 4.5/5 star score. That's about as high as it gets; 5-star reviews have to be "perfect" and that almost never happens.

   
   Yet this Genoa review was so close, yet you had to "Come Up with a reason arbitrarily" to prevent it from doing so....
   Sus as eff.
   
   

2. Intel's consumer chips (which is what you're referencing) support both DDR4 and DDR5. If you want maximum performance and are willing to pay for it, go DDR5. If you want a good midrange or budget option, stick with DDR4. AMD does not offer this flexibility with Zen 4. For the record, Raptor Lake also got 4.5/5 stars. I don't see you pointing out the list of power and cooling requirements as being unreasonable.

   
   Yet you didn't add in "- Early adopter tax for DDR5" to RaptorLake's review for some reason, even though it would be just as applicable.
   Conspicuously missing from Intel's review for some reason.
   Just because DDR4 is an option, doesn't change the fact about DDR5 either since DDR5 was a huge selling point.
   It's not like consumers wanting a new CPU wants to gimp their CPU performance with last gen DDR4 when DDR5 was a option.
   
   

3. As Paul notes, we haven't seen Intel's competing solution, Sapphire Rapids, which has been delays about a year now. Comparing server and consumer reviews to try for a biased hot-take is also misguided at best, intentionally stupid and misleading at worst.

   
   Is it? Because from a average consumer/reader/viewer, it doesn't seem misguided from our PoV.
   The same fact should've equally applied to a Intel review, but it ONLY got brought up for a AMD review, just to give it a "CON".
   You literally had to search for a "-Con" to be applied for the sake of it.
   Do you understand how that makes you look/sound to the average consumer/reader/viewer?
   
   

4. We always need to come up with something (often multiple things) to list as a con. Con doesn't mean "THIS IS THE WORST THING EVER" but rather "This is something to consider." Saying DDR5 costs more is the only con listed, and you're acting as though we slammed the product for it. Again, intentionally biased or misguided: take your pick.

   • Ok, that's fine if you want to come up with Something to list as a con, as long as it's applied evenly to both companies products when it's applicable.
   • But to ONLY have it show up on AMD's review, yet not on Intel for some reason, you can see how from our PoV how that's biased?
   • JaredWalton, I know you've worked for Tom's Hardware for quite a while, but how that makes you look "Biased" is REALLY bad. That shows Favoritism towards a company because you didn't apply it to their review when it was equally pertinent.

Here's the RPL vs. Genoa review summaries side by side. And you'll notice that it doesn't say "DDR5 Support" but rather "Early adopter tax for DDR5." Again, not the end of the world, but you will spend a lot more money on a Genoa system's memory compared to something that still uses DDR4. And that's okay, because for servers the cost of the hardware often pales in comparison to the software and other aspects of ownership (TCO).

It's not a matter of being the "End of the World", it's a matter of even handed analysis and applying it to both.
When RaptorLake came out, the fact that the "Early Adopter tax for DDR5" was equally as important to a consumer comparing platforms at that time shouldn't be lost on you.
Yet it was carefully omitted from the "Cons:" section.
Can you start to see how we, the readership sees that as biased?
You might want to consider applying it evenly to BOTH SIDES, especially when it's technically relevant to both sides.

I bet when the competing Intel Sapphire Rapids chips arrive next year, and DDR5 is required and costs a lot of money, that will still get called out in the articles. But until then, try pulling your head out of the sand and exercising some common sense. A high rating with only a single "Con" should immediately tell you the product is good. The consumer Ryzen parts got a 4/5 star rating, largely because the DDR5, chip, and mobo costs become far more significant on the consumer side. 🤔

We know it's quite good, but other less discerning or informed readership / viewership might not.
That's why they come here, because you guys have been around since the early 1990's.
But they might not understand your PoV, and when you do things like this, it literally gives ammo to those who are already weary of your reputation because of what you did.

View attachment 153

Then why isn't the "-Early adopter tax for DDR5" applied to the Raptorlake review as a "Con"?
It's not like DDR5 wasn't more expensive for Intel consumers?
It's not like DDR5 wasn't a Early Adopter tax for new Raptor Lake customers?
Just because DDR4 option existed, doesn't mean DDR5's "Early Adopter Tax went away"!