Intel Offers More Cascade Lake-AP Performance Numbers

[Athan Immortal]

Source article

 

Anandtech have condensed most of Intel's performance slides into an easier to read table.

 

 

There's no power figures with these either, and as the article notes

Quote

When we are this far away from a product launch, all benchmark numbers should be taken with a grain of salt. This goes doubly so for vendor supplied benchmarks.

We're still waiting for reviewers to get their hands on Zen2 Epyc and these new Intel chips, but I did think it was interesting to re-watch the AMD event where they showcased Epyc2 against 2 of the top Xeons (in a favourable benchmark). Of course in both cases they are vendor numbers.

 

As a side note, this slide of partners is lacking Amazon Web Services. AMD made a big point of showing their new partnership for Epyc2, I'm not sure it would ever be an either or scenario though.

[Princess Luna]

Boring, 10nm or gtfo Intel, going back to the gluing cpus together tactic it bashed AMD so badly due to Infinity Fabric is admitting defeat.

[Athan Immortal]

2 minutes ago, Princess Cadence said:

Boring, 10nm or gtfo Intel, going back to the gluing cpus together tactic it bashed AMD so badly due to Infinity Fabric is admitting defeat.

Watching Lisa take a shot at Intel for that during the AMD Next Horizon event was quite vindicating. As time has gone on, the true reasoning of scaling and yields behind AMD's method has become clear.

 

They're knocking it out the park at the moment. I do worry for them once Intel does get it's 10nm and beyond sorted out, but hopefully they've got the momentum now to keep going and not hit another brick wall like they did with Bulldozer.

 

I really like the scalability of AMDs approach.

[Sauron]

Oh wow, 96 cores are faster than 64 last gen cores in some tasks (according to these very trustworthy tests by intel themselves)... who'd have thought? Let's see how they fare against 128 zen 2 cores...

[Athan Immortal]

Just now, Sauron said:

Oh wow, 96 cores are faster than 64 last gen cores in some tasks (according to these very trustworthy tests by intel themselves)... who'd have thought? Let's see how they fare against 128 zen 2 cores...

I'm honestly more excited about technology and to see this battle over the next few years than I have been since Sandybridge was rocking the performance boat.

[Sauron]

2 minutes ago, Athan Immortal said:

I'm honestly more excited about technology and to see this battle over the next few years than I have been since Sandybridge was rocking the performance boat.

Yep, it's about time.

[leadeater]

4 minutes ago, Sauron said:

Oh wow, 96 cores are faster than 64 last gen cores in some tasks (according to these very trustworthy tests by intel themselves)... who'd have thought? Let's see how they fare against 128 zen 2 cores...

Oh wow it's so so much worse than I was thinking, I assumed it was 1 Cascade Lake-AP vs 2 EPYC 7601.

[Sauron]

6 minutes ago, leadeater said:

Oh wow it's so so much worse than I was thinking, I assumed it was 1 Cascade Lake-AP vs 2 EPYC 7601.

They wouldn't want to face epyc on its own turf (aka in a situation where cascade lake is bound by dual socket problems because of its design and the epyc chip isn't). I suspect the results would be a lot closer in that case, despite the core advantage for intel.

[leadeater]

10 minutes ago, Sauron said:

They wouldn't want to face epyc on its own turf (aka in a situation where cascade lake is bound by dual socket problems because of its design and the epyc chip isn't). I suspect the results would be a lot closer in that case, despite the core advantage for intel.

Honestly right now it would have been more insightful if Intel had compared the performance to it's own existing products, but that would allow us to actually figure out it's performance, including compared to AMD. Plenty of reviews for 8180 vs 7601, all we need is a Cascade Lake-AP vs 1 and 2 socket 8180 and we will know everything we need.

[The Benjamins]

Based on these numbers only linpack and YASK is better on Intel, but the rest Rome should win.

And Rome is shipping this year. Lol

[Taf the Ghost]

As a reminder, you can buy a 2S Xeon Platinum 8168 server already. This system should be slightly worse performance in a number of areas than that system.

[leadeater]

1 minute ago, The Benjamins said:

Based on these numbers only linpack is better on Intel, but the rest Rome should win.

And Rome is shipping this year. Lol

I'd hold on to that assessment, Linpack uses AVX and Rome is double throughput so... interesting times ahead.

[Taf the Ghost]

1 minute ago, leadeater said:

I'd hold on to that assessment, Linpack uses AVX and Rome is double throughput so... interesting times ahead.

The 48 cores with Cascade Lake-AP against the 2x 32c/64t Epyc 7601. The normal use 50% improvement points to that double AVX performance. Charlie at Semi-Accurate has been talking up Rome slaughtering Intel for 2 years. He's not wrong if they normalize AVX performance levels and come with 128c/256t per 2U server.

[leadeater]

1 minute ago, huilun02 said:

Ah more misleading crap from Intel again

[GoldenLag]

1 minute ago, leadeater said:

*sort news by searchword "Intel"

[leadeater]

1 minute ago, Taf the Ghost said:

The 48 cores with Cascade Lake-AP against the 2x 32c/64t Epyc 7601. The normal use 50% improvement points to that double AVX performance. Charlie at Semi-Accurate has been talking up Rome slaughtering Intel for 2 years. He's not wrong if they normalize AVX performance levels and come with 128c/256t per 2U server.

Given how AVX1 performs Intel vs AMD, total slaughter by AMD I can have relative trust in them when they say they have improved AVX2 and doubled it's throughput capabilities (throughput != performance). Even if it's only on par the price would make anything from Intel a joke.

[Taf the Ghost]

1 minute ago, leadeater said:

We kind of rightly gave Intel hell over the "Glued-together" statement, but the 2nd point, "Poor Track Record", was the actual attack. And, it mattered a lot.

[leadeater]

1 minute ago, Taf the Ghost said:

We kind of rightly gave Intel hell over the "Glued-together" statement, but the 2nd point, "Poor Track Record", was the actual attack. And, it mattered a lot.

14nm supply says what? 

 

Edit:

Intel was like, fuck it lets dig our own grave with that slide. Hindsight is a bitch.

[Taf the Ghost]

Just now, leadeater said:

Given how AVX1 performs Intel vs AMD, total slaughter by AMD I can have relative trust in them when they say they have improved AVX2 and doubled it's throughput capabilities (throughput != performance). Even if it's only on par the price would make anything from Intel a joke.

Yup. We'll see how the AVX units operate at 256bit, but the 128bit performance was really good compared to Intel. If that gets normalized, there's just so many cores more. Plus, there's going to be companies that want PCIe 4.0 bandwidth and only AMD will be able to provide that for a while.

[LukeSavenije]

2 minutes ago, leadeater said:

14nm supply says what? 

14 nm is like the only thing intel knows. 7nm? 10nm? they never heared about that

[GoldenLag]

1 minute ago, leadeater said:

14nm supply says what? 

hold up, we have 22nm to help with that. 

 

intel: hey TSMC you wouldnt mind makin some 14nm chips would you?

[Taf the Ghost]

1 minute ago, GoldenLag said:

hold up, we have 22nm to help with that. 

 

intel: hey TSMC you wouldnt mind makin some 14nm chips would you?

The under-reported aspect to the Intel Supply crunch is it appears that yields still aren't brilliant.

[leadeater]

1 minute ago, Taf the Ghost said:

Plus, there's going to be companies that want PCIe 4.0 bandwidth and only AMD will be able to provide that for a while.

Not showing much love to IBM, IBM CPU lives matter too. Only x86 PCIe 4.0.

[Taf the Ghost]

Just now, leadeater said:

Not showing much love to IBM, IBM CPU lives matter too. Only x86 PCIe 4.0.

True.

[LukeSavenije]

1 minute ago, leadeater said:

Not showing much love to IBM, IBM CPU lives matter too. Only x86 PCIe 4.0.

ibm still makes cpu's? good to know