• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD to unveil Zen 4 CPUs at CES 2022

Status
Not open for further replies.
Bencher said:
They haven't managed to reproduce it, as far as I understand, on alderlake. From the official page of hertzbleed.

Hertzbleed Attack

www.hertzbleed.com www.hertzbleed.com

And yes, I know the server market is a really slow moving one, which is actually my point. By the time AMD can actually get any significant foothold, Intel has the time to have an answer ready. Doesn't mean their answer will be a good one, I'm just saying they have the time. Whether they fail or not, is up to them.

We don't know much about zen 4 right now, so going by what we already have, which is Zen 3 vs Golden Cove, the Golden Cove is faster and more efficient, while the Zen 3 due to the chiplets is way cheaper to create. I would assume, if there is one market that can afford to pay for big monolithic chips, it's the server market, so Intel will have the edge there, until of course a zen 4 epyc shows up.
Click to expand...

That's not how I read it, I read that they've tested and reproduced it on various chips upto 11th gen core, not that they've tested on 12th gen and failed to reproduce it, that would be big news and Intel at the very least would be shouting from the rooftops so to speak. They're not so I think it's safe to assume it affects 12th gen just like everything else.

The slowness comes down to verification/trust, even if Sapphire Rapids was released today, and was both faster and more efficient than Milan/Milan-X, it would still take time to make any change because of that verification side of things. So even if we accept the argument that Golden Cove is more efficient than Zen 3 (I personally don't but I've seen the way this conversation has gone) Sapphire Rapids needs to be out in time to compete with Zen 3 EPYCs, and it's not a given that it will be, it looks more likely it will barely be out before Zen 4 EPYCs if at all.

EDIT: Actually just checked out the AMD release about Hertzbleed, interestingly they give specific models that are affected by it and missing from the list are both Desktop 5000 series Ryzen and EPYC 3rd Gen.
 
SKILL said:
That's not how I read it, I read that they've tested and reproduced it on various chips upto 11th gen core, not that they've tested on 12th gen and failed to reproduce it, that would be big news and Intel at the very least would be shouting from the rooftops so to speak. They're not so I think it's safe to assume it affects 12th gen just like everything else.

The slowness comes down to verification/trust, even if Sapphire Rapids was released today, and was both faster and more efficient than Milan/Milan-X, it would still take time to make any change because of that verification side of things. So even if we accept the argument that Golden Cove is more efficient than Zen 3 (I personally don't but I've seen the way this conversation has gone) Sapphire Rapids needs to be out in time to compete with Zen 3 EPYCs, and it's not a given that it will be, it looks more likely it will barely be out before Zen 4 EPYCs if at all.

EDIT: Actually just checked out the AMD release about Hertzbleed, interestingly they give specific models that are affected by it and missing from the list are both Desktop 5000 series Ryzen and EPYC 3rd Gen.
Click to expand...
I guess time will only tell regarding hertzbleed. Problem is, as far as I understand, even a patch won't do much.

Regarding zen 3, it's just a fact that is less efficient, it's just physics. The core is much smaller than the golden cove. What ruins the efficiency on desktop alderlake are the E cores. A 16 GC core sku would be absolute bonkers in terms of efficiency and performance, but it would cost a pretty penny cause of the ridiculous die size. I can disable the e cores from my 12900 and compare it to a 5800x at same wattage, the difference both in performance and efficiency would be huge. You need 8 Zen 3 cores to match 6 GC cores at same wattage in MT performance and efficiency.
 
64 core 5995WX scores around 87,000 in R23 at 280 watts. (4.3 Watts per core)

I look forward to Sapphire Rapids beating that and by the time it comes out it will be up against a 96 core Zen 4 or a 128 core Zen 4C.

PS: i don't think it can beat the 5995WX
 
humbug said:
64 core 5995WX scores around 87,000 in R23 at 280 watts. (4.3 Watts per core)

I look forward to Sapphire Rapids beating that and by the time it comes out it will be up against a 96 core Zen 4 or a 128 core Zen 4C.

PS: i don't think it can beat the 5995WX
Click to expand...

Forgot to add, the Golden Cove core has 10 to 15% higher IPC vs Zen 3, at the same clock speed Sapphire Rapids would need 56 Golden Cove cores to match the 5995WX. it would be a quarter the size of a 300mm $20,000 wafer....
 
humbug said:
Forgot to add, the Golden Cove core has 10 to 15% higher IPC vs Zen 3, at the same clock speed Sapphire Rapids would need 56 Golden Cove cores to match the 5995WX. it would be a quarter the size of a 300mm $20,000 wafer....
Click to expand...
In R23? It's more like 25% IPC difference.

I know that the 5800x at 4.9ghz gets almost 17k in CBR23. 8 GC cores need around 4 ghz for the same performance
 
humbug said:
It doesn't do 4.9Ghz MT R23, it would melt, 4.7Ghz is about the limit for Zen 3 in a hard load.
Click to expand...
Ι've seen people do it. Still, that's not the point, the point is there is a 25% IPC difference in CBR23, not 10-15% like you claimed. I mean it's easy to test, screenshot a CBR23 run at 4ghz and ill do the same with E cores off ;)
 
humbug said:
12900K all core 5Ghz R23 P cores only: 19,500 (8 cores)

5800X all core 4.7Ghz 16,500.

19,500 / 16,500 = 1.18
5.0 / 4.7 = 1.06
19.500 X 0.06 = 1,170 - 19,500 = 18,330
18,330 / 16,500 = 1.11 is +11%
Click to expand...
Your numbers are wrong though. At 5ghz its get around 21k.

Also I doubt the 5800x gets 16500 at 4.7 ghz. I've seen 16800 @ 4.9 ghz, which means at 4.7 it should be around 16.100

EG1. I just tested it. It gets 20.460 at 4.9 ghz and 20.880 at 5.0ghz. The 5800x gets 16.800 @ 4.9 ghz and 16.100 @ 4.7 ghz. So the difference at same clockspeeds is 22%
 
Last edited:
Bencher said:
And quoting from the actual report

"When it comes to price and perfromance all three clouds were in a statistical dead heat"

So yeah, intel is in trouble, they are in a dead heat with amd. Very objective view of reality
Click to expand...

They're comparing price and performance of the cloud services aren't they? There's no Intel/AMD comparison implied in their comparison.
 
Bencher said:
And quoting from the actual report

"When it comes to price and perfromance all three clouds were in a statistical dead heat"

So yeah, intel is in trouble, they are in a dead heat with amd. Very objective view of reality
Click to expand...

Either I've mis read or you have! I think its referring to the cloud service pricing , not cost of the servers to the provider
 
Now lets talk about AMD vs Intel: 1) In past years, we saw Intel lead the pack in overall performance, with AMD competing on price-for-performance metrics. This year, both the overall performance leader and the price-for-performance leader were AMD-based instances.

2) AMD EPYC Milan and Rome, the multi-core x86 and x64 microprocessors, blew away the competing Cascade Lake and Ice Lake Xeon CPUs from Intel between the three cloud platforms in various tests on performance conducted by database company CockroachDB.

3) Furthermore, Intel Xeon CPUs have been having a hard time keeping up with their release schedule with the more recent Sapphire Rapids 4th Gen Xeon family delayed from a 2020 launch to the 2nd quarter of 2023 while AMD will have their 128 core EPYC Bergamo lineup ready for launch around the same time.
 
Status
Not open for further replies.
Back
Top Bottom