News Intel Sapphire Rapids 56-Core ES CPU Hits 3.3 GHz at 420W

[Admin]

Twitter user exposes the specifications for an Intel Sapphire Rapids-SP ES CPU with 56 cores and 112 threads.

Intel Sapphire Rapids 56-Core ES CPU Hits 3.3 GHz at 420W : Read more

 

[escksu]

Power consumption is the main reason why these multicore server cpus have very low clockspeeds...

 

[jacob249358]

420... nice

 

[Kamen Rider Blade]

Power consumption is the main reason why these multicore server cpus have very low clockspeeds...

I thought it was a combination of Power Consumption & Thermals.

 

[jacob249358]

I thought it was a combination of Power Consumption & Thermals.

well power consumption is the cause of thermals so its kinda interchangeable in some scenarios.

 

[rluker5]

I'll just quote the article here:

" The Sapphire Rapids-SP processor from YuuKi-AnS is an engineering sample (ES), so don't take the clock speeds seriously. "

Also how many watts should 56 cores running 512 at 3.3ghz burn? That's 7.5w/core or 60w for an 8 core. And does that include the 64Gb HBM2e?

 

[jeremyj_83]

I'll just quote the article here:

Also how many watts should 56 cores running 512 at 3.3ghz burn? That's 7.5w/core or 60w for an 8 core. And does that include the 64Gb HBM2e?

We don't know if that is using AVX512 that will cause a 420W TDP. Sure 7.5w/core doesn't sound bad but is 3.3GHz the all core boost or is it something lower. On top of that the 64c Epyc is only 280W TDP or 4.38W/core. That means that SPR is using 70% more power to hit the same clocks. We don't know what Zen 4 will have for TDP but that will be the main competition for SPR and not Zen 3 Epyc.

 

[jeremyj_83]

Power consumption is the main reason why these multicore server cpus have very low clockspeeds...

Depends on the SKU. There are Gen 3 Epyc SKUs (7XF3) that are high clock speed designs that range from 8-32 cores. The 16c variant has a 3.5GHz base and 4.0GHz boost. The 32c has a nice 2.95GHz base and 4GHz boost. There will always be a tradeoff though going more cores. If you need more cores you are going to be sacrificing base clock speed but they are usually able to keep the boost to 3.5GHz+.

 

[pixelpusher220]

pair this with a RTX 3090 and you get a free level 2 charger installed.

sorry, I meant need a level 2 charger

 

[PCWarrior]

We don't know if that is using AVX512 that will cause a 420W TDP. Sure 7.5w/core doesn't sound bad but is 3.3GHz the all core boost or is it something lower. On top of that the 64c Epyc is only 280W TDP or 4.38W/core. That means that SPR is using 70% more power to hit the same clocks. We don't know what Zen 4 will have for TDP but that will be the main competition for SPR and not Zen 3 Epyc.

The latest 64C Epyc (the 7H12) has a max turbo boost of 3.3Ghz. That is however the single thread boost. Not the all-core boost. The other information we have is its base clock which is 2.6Ghz. This is its all-core “boost” under a really heavy all-core AVX2 workload. The TDP of 280W is rated at 2.6GHz under such a heavy workload. Under a lighter SSE all-core workload, it can probably boost a bit higher than base, though definitely below 3.3Ghz and likely 2.8GHz max. This Sapphire Rapids CPU on the other hand has a base clock of 3.3Ghz. The TDP for Intel cpus is defined at base frequency under an extremely intense Intel-defined workload which even though not an AVX-512 workload is much heavier than any AVX2 workload in terms of power consumption. So if we assume that under the same workload the AMD cpu would have to default to its base clock at 280W, then we are talking about AMD 64C@2.6GHz@280W Vs Intel 56C@3.3Ghz@420W.

A far more useful metric than the watts/core is performance per core. Or, if you prefer, watts/core for the same performance per core. With 27% higher frequency we can theoretically have 27% higher performance if IPC were equal. Then comes IPC which is a per workload metric. In workloads like Cinebench R23, Golden Cove cores have a 17-17.5% better IPC than Zen 3 cores. So in that workload we have a total of 48.6% (=1.27x1.17) higher performance per core in that workload. Factoring this we have 5W/core for the same performance, not 7.5. So not so far off from the 4.38W/core of Epyc. Then there are other things to consider like the fact that this is an engineering sample and final silicon will be better. It might as well get down to 4-4.5W/core for the same performance. Zen 4 Epyc will be launching next year. By that point (or within 3-4 months) Intel will be launching SPR's replacement. SPR's delayed launch won't affect (or at least not much) the timeline of its replacement.

 

[jeremyj_83]

The latest 64C Epyc (the 7H12) has a max turbo boost of 3.3Ghz. That is however the single thread boost. Not the all-core boost. The other information we have is its base clock which is 2.6Ghz. This is its all-core “boost” under a really heavy all-core AVX2 workload. The TDP of 280W is rated at 2.6GHz under such a heavy workload. Under a lighter SSE all-core workload, it can probably boost a bit higher than base, though definitely below 3.3Ghz and likely 2.8GHz max. This Sapphire Rapids CPU on the other hand has a base clock of 3.3Ghz. The TDP for Intel cpus is defined at base frequency under an extremely intense Intel-defined workload which even though not an AVX-512 workload is much heavier than any AVX2 workload in terms of power consumption. So if we assume that under the same workload the AMD cpu would have to default to its base clock at 280W, then we are talking about AMD 64C@2.6GHz@280W Vs Intel 56C@3.3Ghz@420W.

A far more useful metric than the watts/core is performance per core. Or, if you prefer, watts/core for the same performance per core. With 27% higher frequency we can theoretically have 27% higher performance if IPC were equal. Then comes IPC which is a per workload metric. In workloads like Cinebench R23, Golden Cove cores have a 17-17.5% better IPC than Zen 3 cores. So in that workload we have a total of 48.6% (=1.27x1.17) higher performance per core in that workload. Factoring this we have 5W/core for the same performance, not 7.5. So not so far off from the 4.38W/core of Epyc. Then there are other things to consider like the fact that this is a an engineering sample so final silicon will be better. It might well get to 4-4.5W/core for the same performance. Zen 4 Epyc will be launching next year. By that point (or within 3-4 months) Intel will be launching SPR's replacement. SPR's delayed launch won't affect (or at least not much) the timeline of its replacement.

Wow there is a lot of wrong information in that post PCWarrior.

1. The 7H12 is not the latest 64c Epyc. The 7H12 was released in September 2019, is a high frequency CPU, and is a Rome uArch design. Whereas the 7763 is a Milan CPU that was released in March 2021. The 7763 has a 2.45GHz base clock, 3.0GHz all core boost, and a 3.5GHz single core boost.
2. AMD does not base their TDP around the base clock speed. AMD uses total package power and their server CPUs will top out a 280W regardless of instruction set being used. Intel on the other hand bases their TDP around the base clock speed. This is why we see their CPUs say "120W TDP" but under load will draw 300W and can continue to do that if the Tau is set to infinity.
3. That SPR engineering sample has a 1.9GHz base clock and a 3.3GHz boost clock. From the article "Thus far, the 56-core part operates with a 1.9 GHz base clock and a 3.3 GHz boost clock. According to the report, the single-core boost clock escalates up to 3.7 GHz."
4. If that max draw is 420W under AVX-512 then at best the CPU will be running at the base clock of 1.9GHz. Previous versions of Intel CPUs that ran AVX-512 instructions could not run at their base clock with all cores running in AVX-512. It wasn't uncommon to see up to 500MHz reduction in clock speed from the base clock when all cores were running AVX-512. Intel seems to have gotten around some of this clock speed reduction by making the TDP in the PL2 being MUCH higher.
5. Again the frequency of this SPR engineering sample is only a 1.9GHz base clock. That puts the SPR 29% slower in the frequency which would negate your "17% higher IPC." When normalized for score per GHz in Cineabench R23, the Golden Cove with its best setup (Win 10 + DDR4) is 13% faster than Zen 3 in the single threaded benchmark. Golden Cove @ 2008 = 386.2 per GHz vs Zen 3 @ 1636 = 340.8 per GHz. https://www.anandtech.com/bench/product/2674?vs=2894 That higher performance won't matter if the clock speeds aren't high enough. This is the same reason that Ice Lake Xeon isn't always faster than Cascade Lake Xeon.
6. SPR has been delayed to Q3 this year. As of January 2022, Zen 4 was scheduled to come out end Q3 or beginning Q4 2022. The odds of Intel releasing Emerald Rapids all of 6ish months after SPR is released almost nil.

 

[PCWarrior]

Wow there is a lot of wrong information in that post PCWarrior.

1. The 7H12 is not the latest 64c Epyc. The 7H12 was released in September 2019, is a high frequency CPU, and is a Rome uArch design. Whereas the 7763 is a Milan CPU that was released in March 2021. The 7763 has a 2.45GHz base clock, 3.0GHz all core boost, and a 3.5GHz single core boost.
2. AMD does not base their TDP around the base clock speed. AMD uses total package power and their server CPUs will top out a 280W regardless of instruction set being used. Intel on the other hand bases their TDP around the base clock speed. This is why we see their CPUs say "120W TDP" but under load will draw 300W and can continue to do that if the Tau is set to infinity.
3. That SPR engineering sample has a 1.9GHz base clock and a 3.3GHz boost clock. From the article "Thus far, the 56-core part operates with a 1.9 GHz base clock and a 3.3 GHz boost clock. According to the report, the single-core boost clock escalates up to 3.7 GHz."
4. If that max draw is 420W under AVX-512 then at best the CPU will be running at the base clock of 1.9GHz. Previous versions of Intel CPUs that ran AVX-512 instructions could not run at their base clock with all cores running in AVX-512. It wasn't uncommon to see up to 500MHz reduction in clock speed from the base clock when all cores were running AVX-512. Intel seems to have gotten around some of this clock speed reduction by making the TDP in the PL2 being MUCH higher.
5. Again the frequency of this SPR engineering sample is only a 1.9GHz base clock. That puts the SPR 29% slower in the frequency which would negate your "17% higher IPC." When normalized for score per GHz in Cineabench R23, the Golden Cove with its best setup (Win 10 + DDR4) is 13% faster than Zen 3 in the single threaded benchmark. Golden Cove @ 2008 = 386.2 per GHz vs Zen 3 @ 1636 = 340.8 per GHz. https://www.anandtech.com/bench/product/2674?vs=2894 That higher performance won't matter if the clock speeds aren't high enough. This is the same reason that Ice Lake Xeon isn't always faster than Cascade Lake Xeon.
6. SPR has been delayed to Q3 this year. As of January 2022, Zen 4 was scheduled to come out end Q3 or beginning Q4 2022. The odds of Intel releasing Emerald Rapids all of 6ish months after SPR is released almost nil.

1. I didn’t reference any benchmarks of the 7H12 so the name of the latest 64C epyc is not that relevant. The frequency is incorrect but not to AMD’s favour as the 7763 Milan has an even lower base clock (2.45GHz) than the 7H12 (2.6GHz). Not sure from where you get the all-core boosts of these cpus (you say 3GHz) but that is not something advertised by AMD. AMD (here) just lists the base clock (2.45GHz) and max single thread clock (3.5GHz) and I haven't seen (e.g. here or here) any promotional material or review referencing or testing the all core turbo frequency.
2. Intel deviates from the TDP only when boosting. Not at base clocks. Actually if you get a non-AVX512 cpu with a certain TDP and run it at its base clock you will find out that you won’t be able to make it consume anywhere near its TDP. I did this experiment once on a 9900K and ran various intense workloads (handbrake, prime95 AVX, etc) while running it at a fixed 3.6GHz (which is equal to its bse). The highest power consumption I could get it to was 65W, despite the cpu having a TDP of 95W. I have since repeated it on other cpus and the outcome is always the same. That can only mean two things. Intel’s testing workload is really intense. It may not be as intense as an AVX-512 power virus but is more intense than any AVX2 workload or power virus we can throw to a cpu. The other thing is that Intel must account for the worst possible silicon as well as for the worst motherboard and cooling settings. In other words the cpu that consumes as much as the TDP while running at base clock is the worst of the worst in the worst scenario.
3. I know very well how AMD defines their TDP. And I know that AMD server cpus will consume at max as much power as their defined TDP (280W). But the all-core boost of any cpu and of the 7763 in this case, depends on the workload. On one extreme (under a light all-core workload such as Cinebench R15) the cpu will boost to its maximum all-core frequency (e.g. 3GHz). And it may not even consume the full power budget of 280W while doing so. On the other extreme however (when REALLY heavily stressed akin to how Intel stresses their cpus for tdp defining purposes) the cpu will clock to as low as its base speed. Otherwise, there wouldn’t be the need for AMD to define a lower base clock than its all-core turbo (if the all-core turbo -which is not advertised anyway- could be guaranteed under any workload while also remaining under the power rating of tdp). So I think it is a safe assumption to say that under the same intense power virus that Intel uses to define the tdp of their own cpus, an AMD cpu would not have the power headroom to boost higher than its base clock.
4. The TDP of SPR is 350W (PL1). So under the Intel-defined power virus it runs all cores at 1.9GHz and consumes 350W (on the worst silicon). Then it says that PL2 is 420W. So under the same Intel-defined power virus it can do an all-core 3.3GHz turbo and consume 420W (which is what I used in my previous post for comparison). Then it specifies that the maximum power limit in the BIOS is 764W. That is likely in order to allow the cpu to boost to the same all-core frequency (3.3GHz) while running an intense AVX-512 workload.
5. The IPC advantage of Golden Cove versus Zen 3, normalised at 4.9Ghz is 17.4%. The 12900K at 5GHz (see here) scores 1963. The 5950X at 4.9GHz scores 1638. That’s 19.8% lead. Accounting for the 100MHz difference we get 17.4%.
6. Zen 4 will come out end of Q3/beginning of Q4 but that won’t necessarily be the server chips. These may very well come out at the very end of the year (December 2022). Intel has released server cpus within 9 months before (Cooper lake and Icelake SP). The latest delay for SPR I have read mentioned H2 2022, not Q3 2022. So it may well be a June announcement and July launch. Adding 9 months for Emerald Rapids brings us to April 2023 (4 months after Zen 4 Epyc).

 

[gruffi]

Nonetheless, it'll be exciting to see Golden Cove go up against Zen 3 in the data center space

Zen 3? LOL. When Sapphire Rapids finally hits the market it will already face Zen 4 aka Genoa.

The IPC advantage of Golden Cove versus Zen 3, normalised at 4.9Ghz is 17.4%. The 12900K at 5GHz (see here) scores 1963. The 5950X at 4.9GHz scores 1638. That’s 19.8% lead. Accounting for the 100MHz difference we get 17.4%.

That's wrong. IPC advantage of Golden Cove over Zen 3 is something like ~10%. Making IPC claims on the basis of a single app is pointless. Referencing TPU is pointless as well. They are using DDR5-6000 for Intel, 25% over the supported specs. Normal SR servers won't use such memory. They also seem to use a bad AM4 system. The 5950X doesn't even score 26k in CB23. On a good configured AM4 system the 5950X can score around 29k in CB23.

Zen 4 will come out end of Q3/beginning of Q4 but that won’t necessarily be the server chips. These may very well come out at the very end of the year (December 2022).

Unlikely. Genoa ES are floating around for quite some time. I guess it will be launched very closely to Raphael. Which means somewhere in Q3. It makes not much sense to launch Genoa at the end of the year when Bergamo is expected early next year.

 

[Lorien Silmaril]

just fyi there is a leaked benchmark.

Intel Sapphire Rapids-SP Xeon Platinum ES CPUs Fail To Impress Against EPYC Milan (X) & Last-Gen Xeons In Leaked Benchmarks

Latest benchmarks of Intel's Sapphire Rapids-SP Xeon ES CPU shows lackluster performance against AMD's EPYC Milan processors.

wccftech.com

spoiler: it's not looking too good.

 

[Deleted member 14196]

Yeah all these guys can seay all the glowing things they want about the Intel processors but they’re very inefficient and they consume tons of power and this isn’t changing. So if they can’t Win by the power metric they change the rules and try to change all the metrics so to make it look better

Fact is they can’t touch AMD in the server workspace the epyc processors blow them out of the water and will continue to do so year after year. There’s a reason why everyone switching to epyc processors in the data center

 

[penguinslovebananas]

[

So if they can’t Win by the power metric they change the rules and try to change all the metrics so to make it look better

This is the correct state of the intel v. and landscape right now. The true facts of hardware are very well known in the enterprise realm and the vast amount of market share intel has lost to amd, IIRC they were up to around 95% at one time, bears out this info, and yes I know the most recent quarterly has Intel up, but that is 5% in the CONSUMER space, and had much more to do with supply issues than anything. How about the fact Intel has not produced an HEDT product since cascade lake, I would like to know your “facts” behind that. I understand everyone having their favorites, I am a tried and true AMD fan, using them in every build, the only intel pc I ever owned was a laptop that was a gift. If you are simply an Intel die hard just come and say it, however spreading false information that can easily be proven wrong of out of date help no one. At worst it can be taken as fact by consumers and hurt their purchasing choices or is simply ignored by those knowledgeable in hardware.

 

[PCWarrior]

Metrics are designed to convey information in an easy to compare way. But when the wrong metrics are used they can be misleading. When former Anandtech writer Ian Cuttress was doing energy comparisons and was showing AMD in good light, the AMD fbs were cheering. Now they don't like it and complain about changing metrics. In any case using power to claim efficiency is incorrect. Energy is what is important here and Energy=Power x time. In Layman’s terms if my CPU uses twice the power than yours but (due to higher clockspeed and higher IPC) manages to complete the task in half the time, it will consume the exact same energy as yours while also saving me half the time. Now you may be wondering where power is relevant. Power is relevant in cooling. If two cpus have the same die size and the same cooling solution, the higher the power of a cpu the hotter it will run (thereby necessitating the use of a better cooler). If you look at the physics of heat transfer or simply a thermal equivalent circuit, you will understand why.

That's wrong. IPC advantage of Golden Cove over Zen 3 is something like ~10%. Making IPC claims on the basis of a single app is pointless. Referencing TPU is pointless as well. They are using DDR5-6000 for Intel, 25% over the supported specs. Normal SR servers won't use such memory. They also seem to use a bad AM4 system. The 5950X doesn't even score 26k in CB23. On a good configured AM4 system the 5950X can score around 29k in CB23.

1. I already said in my original post that IPC is a per workload metric. I used Cinebench as it is a popular benchmark that its performance is almost completely invariable of RAM configuration. I find it funny how AMD fbs are now rejecting Cinebench as a benchmark and suggest looking at other benchmarks while for years Cinebench was THE TRUTH and were making fun of Intel when Intel said to look at other benchmarks. Double standard hypocrites of the highest order.

2. You are quoting my follow-up post which was a response to the claim that the best configuration for Alderlake in Cinebench is with Win10 and DDR4 and that it was supposedly only showing 13% difference. Which demonstrably is not true. As for Techpowerup’s testing in particular, the memory configuration used was the best for each platform for a dual channel 2x16GB setup. On AMD they used 2x16GB DDR4 3600C16, also above spec. And dual rank memory which benefits AMD.

3. For Intel, Cinebench scores do not vary with RAM almost at all. If in doubt look at benchmarks in various reviews that did both testing with DDR4 and DDR5 as well as with different RAM frequencies. We are talking about variation of 50-60 points at most and Cinebench scores vary a little from run to run for other reasons anyway. As for the MT score of the 5950X every good reviewer, in their benchmarks of the 5950X had its MT score in CB R23 around 26K. The 28K or higher scores you are seeing in some benchmark databases are with PBO enabled or OCing. It is not stock performance.

Unlikely. Genoa ES are floating around for quite some time. I guess it will be launched very closely to Raphael. Which means somewhere in Q3. It makes not much sense to launch Genoa at the end of the year when Bergamo is expected early next year.

We will see whose prediction is right. As for Genoa ES floating around and all I should remind you that at the beginning of June 2021 AMD demonstrated a working 3DVcache 5900X implying it will come later in the year only to never come to market and eventually launch a 5800X3D 9.5 months later. AMD may announce/show Genoa in Computex but I doubt we will see an actual launch prior to November-December. Certainly not in high volume. And that's based on TSMC's 5nm yields and capacity.

just fyi there is a leaked benchmark.

Intel Sapphire Rapids-SP Xeon Platinum ES CPUs Fail To Impress Against EPYC Milan (X) & Last-Gen Xeons In Leaked Benchmarks

Latest benchmarks of Intel's Sapphire Rapids-SP Xeon ES CPU shows lackluster performance against AMD's EPYC Milan processors.

wccftech.com

spoiler: it's not looking too good.

The same benchmarks also show SPR performing 25% worse in MT than Icelake despite having 40% more cores. It is obvious that there is an issue with the ES SPR two-socket system tested. The scores look as if only one CPU is enabled. We will see.

 

[jeremyj_83]

1. I didn’t reference any benchmarks of the 7H12 so the name of the latest 64C epyc is not that relevant. The frequency is incorrect but not to AMD’s favour as the 7763 Milan has an even lower base clock (2.45GHz) than the 7H12 (2.6GHz). Not sure from where you get the all-core boosts of these cpus (you say 3GHz) but that is not something advertised by AMD. AMD (here) just lists the base clock (2.45GHz) and max single thread clock (3.5GHz) and I haven't seen (e.g. here or here) any promotional material or review referencing or testing the all core turbo frequency.
2. Intel deviates from the TDP only when boosting. Not at base clocks. Actually if you get a non-AVX512 cpu with a certain TDP and run it at its base clock you will find out that you won’t be able to make it consume anywhere near its TDP. I did this experiment once on a 9900K and ran various intense workloads (handbrake, prime95 AVX, etc) while running it at a fixed 3.6GHz (which is equal to its bse). The highest power consumption I could get it to was 65W, despite the cpu having a TDP of 95W. I have since repeated it on other cpus and the outcome is always the same. That can only mean two things. Intel’s testing workload is really intense. It may not be as intense as an AVX-512 power virus but is more intense than any AVX2 workload or power virus we can throw to a cpu. The other thing is that Intel must account for the worst possible silicon as well as for the worst motherboard and cooling settings. In other words the cpu that consumes as much as the TDP while running at base clock is the worst of the worst in the worst scenario.
3. I know very well how AMD defines their TDP. And I know that AMD server cpus will consume at max as much power as their defined TDP (280W). But the all-core boost of any cpu and of the 7763 in this case, depends on the workload. On one extreme (under a light all-core workload such as Cinebench R15) the cpu will boost to its maximum all-core frequency (e.g. 3GHz). And it may not even consume the full power budget of 280W while doing so. On the other extreme however (when REALLY heavily stressed akin to how Intel stresses their cpus for tdp defining purposes) the cpu will clock to as low as its base speed. Otherwise, there wouldn’t be the need for AMD to define a lower base clock than its all-core turbo (if the all-core turbo -which is not advertised anyway- could be guaranteed under any workload while also remaining under the power rating of tdp). So I think it is a safe assumption to say that under the same intense power virus that Intel uses to define the tdp of their own cpus, an AMD cpu would not have the power headroom to boost higher than its base clock.
4. The TDP of SPR is 350W (PL1). So under the Intel-defined power virus it runs all cores at 1.9GHz and consumes 350W (on the worst silicon). Then it says that PL2 is 420W. So under the same Intel-defined power virus it can do an all-core 3.3GHz turbo and consume 420W (which is what I used in my previous post for comparison). Then it specifies that the maximum power limit in the BIOS is 764W. That is likely in order to allow the cpu to boost to the same all-core frequency (3.3GHz) while running an intense AVX-512 workload.
5. The IPC advantage of Golden Cove versus Zen 3, normalised at 4.9Ghz is 17.4%. The 12900K at 5GHz (see here) scores 1963. The 5950X at 4.9GHz scores 1638. That’s 19.8% lead. Accounting for the 100MHz difference we get 17.4%.
6. Zen 4 will come out end of Q3/beginning of Q4 but that won’t necessarily be the server chips. These may very well come out at the very end of the year (December 2022). Intel has released server cpus within 9 months before (Cooper lake and Icelake SP). The latest delay for SPR I have read mentioned H2 2022, not Q3 2022. So it may well be a June announcement and July launch. Adding 9 months for Emerald Rapids brings us to April 2023 (4 months after Zen 4 Epyc).

1. You are bringing up clock speeds to make a point but the 7763 with a lower base but higher boost clock is still faster than the 7H12 you tried to pass of as the "latest 64c Epyc (&H12). I did read the information incorrectly in regards to the all core boost. That was for a different CPU in a review of the 7763.
2. Yes Intel's TDP is ONLY for when all cores are running at base clock. This causes a problem for people building their own desktops because they think their 125W 10c/20t 10900k will be fine with a 150W TDP cooler. Then they install that cooler and find out the cooler cannot keep up with the heat being produced by the CPU (eg CPU running at 90C+). The issue is the motherboards, and Intel doesn't tell them to not do this, are preset to run at infinite boost duration. This is one of the reasons Intel CPUs have been able to keep up because they are boosting for longer time than they should based on power draw. We then see these CPUs drawing 280W of power and that 150W cooler cannot keep up and the CPU starts to throttle which kills performance. Whereas AMD says the total package power for their 105W TDP is 142W without using PBO. Now a 150W cooler still won't keep something like that very cool but you won't have the same throttling issues.
3. We just covered power so no need to go through it again.
4. You do not know what the all core boost for the SPR ES. The only information is it has a 1.9GHz base and 3.3GHz boost (note it doesn't say ALL CORE boost). That is most likely the single core boost and the all core on the ES is going to be less than 3GHz (2.8GHz would be my guess). We do know that the Intel CPUs are not power efficient and they need that power to clock high enough to be competitive. If I were to guess the clocks of the 56c/112t SPR will be at best similar to the 40c/80t Ice Lake 2.3GHz base with a 3.4GHz boost. However, in order to accommodate those clocks the max power will be substantially higher (350W vs 270W) without even getting to the PL2.
5. The site I used is Anandtech's bench which is done in stock settings. The 12900k has a peak 5.2GHz boost and the configuration that got the highest score (you can choose different ones and I looked at them all) was the DDR4 + Win 10 for CineBench R23. The 12900k got a score of 2008 in that configuration. The DDR5 + Win 11 = 1994 & DDR5 + Win 10 = 1973. When I said that was the best configuration, I wasn't lying. "the best configuration for Alderlake in Cinebench is with Win10 and DDR4 and that it was supposedly only showing 13% difference" is 100% accurate when you normalize over the clock speed. I used the 5900X for my example and it has a max boost of 4.8GHz. From Ananadtech's bench the 5900X get 1636 in ST CineBench R23 and the 5950X get 1655. Even if we use the 2030 score from Techpowerup for the 12900k at 5.2 GHz that still only gives a 14.6% improvement and NOT 17.4% when using the numbers from Ananadtech for the Ryzen 5900X. Now the numbers could be reduced for the Ryzens by a bit on Techpowerup as they were using straight Win 11.
6. Genoa is set to be released in Sept/Oct which is the server CPU. Does that mean day 1 availability, you guess is as good as mine. Even if Intel releases SPR on July 1 (1st day of Q3), there isn't any guarantee there will be availability. It took a couple months from the release of Ice Lake to have servers available to buy. Yes Intel released Cooper Lake and Ice Lake within a year of each other. However, they do not compete in the same space. Cooper Lake is 4 & 8 socket whereas Ice Lake is 2 socket only. My guess for Emerald Rapids will be 12 months minimum after release of SPR.

 

[gruffi]

1. I already said in my original post that IPC is a per workload metric. I used Cinebench as it is a popular benchmark that its performance is almost completely invariable of RAM configuration. I find it funny how AMD fbs are now rejecting Cinebench as a benchmark and suggest looking at other benchmarks while for years Cinebench was THE TRUTH and were making fun of Intel when Intel said to look at other benchmarks. Double standard hypocrites of the highest order.

Calling people fanboys, which is a blatant lie by the way, makes your "arguments" just look even more silly. No one said Cinebench is a bad benchmark. Can you read? I said IPC is not measured with a single app. IPC hugely differs from app to app. I could also reference Super Pi which was the IPC benchmark king in the "old days" and could claim Alder Lake's IPC is not higher than Rocket Lake and even lower than Zen 3. That's why IPC claims based on a single app is completely pointless and misleading. You have to benchmark a lot more apps with different types of workloads to get a more meaningful average. And from what I can see in reviews Alder Lake's IPC advantage over Zen 3 is ~10%.

As for Techpowerup’s testing in particular, the memory configuration used was the best for each platform for a dual channel 2x16GB setup. On AMD they used 2x16GB DDR4 3600C16, also above spec. And dual rank memory which benefits AMD.

DDR4-3600 is not the best for AMD. Just like DDR5-6000 is not the best for Intel. But that doesn't change any of my arguments. It's still not officially supported memory. So, it's not quite representative. And the used memory for Intel still offers a lot more bandwidth over officially supported memory compared to AMD (25% vs 12.5%). Which makes it an unfair comparison. Using DDR4-3600 vs DDR5-5200/5600 or DDR4-4000 vs DDR5-6000 would have been more fair. But again, testing at supported memory specs would have been the best and most representative scenario. Which means DDR4-3200 for AMD and DDR4-3200/DDR5-4800 for Intel.

3. For Intel, Cinebench scores do not vary with RAM almost at all. If in doubt look at benchmarks in various reviews that did both testing with DDR4 and DDR5 as well as with different RAM frequencies. We are talking about variation of 50-60 points at most and Cinebench scores vary a little from run to run for other reasons anyway.

Of course the variation of 12900K is much less. Because it can suck almost 250W. That's why it is so bad and power inefficient in such scenarios. It still doesn't make your arguments better if you reference one of the most Intel biased reviews.

As for the MT score of the 5950X every good reviewer, in their benchmarks of the 5950X had its MT score in CB R23 around 26K. The 28K or higher scores you are seeing in some benchmark databases are with PBO enabled or OCing. It is not stock performance.

That's wrong. Every good review sees scores of more than 26k. AT STOCK, NO OC. Good reviews also show CB15 and CB20 scores. Which don't favor Intel that much as CB23.

We will see whose prediction is right. As for Genoa ES floating around and all I should remind you that at the beginning of June 2021 AMD demonstrated a working 3DVcache 5900X implying it will come later in the year only to never come to market and eventually launch a 5800X3D 9.5 months later.

That's irrelevant. We are talking about Genoa ES. Which is the basis for the real launched SKUs. The demonstration of 5900X with 3D V-Cache is just what it was, a demonstration. AMD never announced a 5900X with 3D V-Cache. I'm sure AMD will ship Genoa soon to first customers. So, server Zen 4 will "come out" a lot earlier than end of this year. When you will be able to buy one at the retailer of your choice is another story.