News AMD crowns the Ryzen 7 9800X3D a ‘gaming legend’ in a surprise announcement — chipmaker claims $479 Zen 5 3D V-Cache chip is up to an average 20% f...

[Avro Arrow]

Maybe, but I don't buy chips based off of marketing, I buy them based off of cold hard benchmarks and game performance results from third parties.

I think that most people do, which makes the dishonesty all the more pointless.

 

[-Fran-]

Of course they are using slow ram for Intel when their testing their chips with 6000c30. Man, just try to be objective for once, it ain't gonna harm you.

I'll half side with you on this one.

AMD is testing using EXPO 6000MT/s and using CU-DIMM (I'll assume) JEDEC 6400MT/s on the 285K. The fine print did not mention timing for that specific config, so I don't know where you got that from though.

Still, AMD being ever so scummy by using "OC" in their RAM (mind you; still technically SLOWER) vs "stock" on Intel. If they were using regular U-DIMM (non clocked), then it's a different story. Intel would be using XMP and even faster RAM, so that would still be a worse look on the Intel parts.

Regards.

 

[aberkae]

Only if AMD had a 5090 killer to pair with the 9800X3D, it would have been sweeter. Lisa loves giving Nvidia free advertisement!

 

[jeremyj_83]

Of course they are using slow ram for Intel when their testing their chips with 6000c30. Man, just try to be objective for once, it ain't gonna harm you.

There is no evidence of using CL30 DDR5-6000 as they don't provide the timings in the notes. That said there is almost no difference in performance (less than 1%) between CL30 and CL34. We have seen this time over time with RAM gen over gen due to the difference in CL being percentage wise smaller. In DDR1 era going CL3 over CL5 was a big performance difference by DDR4 CL16 was only 1-2% faster than CL18 at best. This keeps going with DDR5.

AMD only said that the 285k had 6400MHz RAM. You are immediately jumping to a conclusion with no proof that it is automatically the slowest RAM possible. The only evidence we have is they are using the fastest guaranteed MHz. We don't know the CL nor if it is CU-DIMM or UDIMM. We do know that the 8200(I think that is what Tom's used) only added on average 2% over 7200MHz UDIMMs. They didn't test with official top stock specs. However, once again I must remind you that you are NOT guaranteed to get anything beyond 6400MHz.

Note it would be nicer if the tests were stock speed vs stock speed and not using any RAM OC. So in this case 5600 on AMD and 6400 on Intel.

Seriously I have only posted facts based on available data. Just because the facts go against your beliefs of what was used doesn't mean I am not being objective. Between the two of us it is you who isn't being objective and are making wild claims against what the facts at hand show.

 

[halfcharlie]

AMD and Intel both have marketing specifically mentioning gaming. Anyone who falls for it is embarrassing, nobody is being bottlenecked by their CPU. Someone gaming on a ridiculously low 1080 monitor isn't the same person buying a top-end CPU for starters. And as has been proven in recent benchmarks, at 4K there's barely a 3-5FPS difference across 20 CPU's from the past half decade down to i5. Your CPU ain't gonna do crap for your 'gaming'. GPU, M.2 HDD, cooling, RAM, monitor, are all more important. If you're also using your PC for productivity that's when the CPU matters, if it's just an internet/gaming machine then it doesn't.

 

[YSCCC]

AMD and Intel both have marketing specifically mentioning gaming. Anyone who falls for it is embarrassing, nobody is being bottlenecked by their CPU. Someone gaming on a ridiculously low 1080 monitor isn't the same person buying a top-end CPU for starters. And as has been proven in recent benchmarks, at 4K there's barely a 3-5FPS difference across 20 CPU's from the past half decade down to i5. Your CPU ain't gonna do crap for your 'gaming'. GPU, M.2 HDD, cooling, RAM, monitor, are all more important. If you're also using your PC for productivity that's when the CPU matters, if it's just an internet/gaming machine then it doesn't.

Firstly, there are ppl who have space limitation to use a 1080P monitor for gaming (24") but use the best hardware possible, ability to max out all settings without DLSS is a big plus in that case.

And for those who buy a top end CPU, chances are slim they will cheap out on GPU, SSD, cooling and Ram, so that isn't an issue either.

FPS wise, that is another issue, firstly though not as common as GPU bottlenecking, there are games that need massive CPU calculations for background stuffs like flight simulation, which is usually main thread limited, and a lot without CPU limitation at all. But the main thing is, if one only build a PC for gaming and not doing any rendering or so, how "future proof" the CPU is, will be a big concern and money saver in medium term, see how much performance the 4090 have been uplifted compared to 2080? A CPU that can last for gaming only for 2 more generations will be a plus for those users, for office work, one can still do fine literally on a Core 2 Duo with a SATA SSD without major issue, we have long passed the era where new hardware is needed for daily office work, the "future proof" side sometimes just make sense, and of course, when one is not in the market for that, there's no harm if you put say an i5 or even i3 with 4090 and enjoy your game.

 

[YSCCC]

Of course they are using slow ram for Intel when their testing their chips with 6000c30. Man, just try to be objective for once, it ain't gonna harm you.

This from you is... hum... costed me a spilled coffee When you don't even have proof for that.

But yea, promotion stuffs usually are exaggeratted, I don't believe in actual gaming there could be 159% over 285k in cyber punk, maybe in a scene or two, but it does make sense overall being much faster than the ARL in gaming, changing ram alone won't get anywhere near 20% improvement, let alone 59%

 

[Amdlova]

World best gaming cpu... what games? Only see flopped games and other studios closing...
At this rate will get a gtx 980 and windows xp copy.

 

[TheHerald]

I'll half side with you on this one.

AMD is testing using EXPO 6000MT/s and using CU-DIMM (I'll assume) JEDEC 6400MT/s on the 285K. The fine print did not mention timing for that specific config, so I don't know where you got that from though.

Still, AMD being ever so scummy by using "OC" in their RAM (mind you; still technically SLOWER) vs "stock" on Intel. If they were using regular U-DIMM (non clocked), then it's a different story. Intel would be using XMP and even faster RAM, so that would still be a worse look on the Intel parts.

Regards.

Helper 800 asked me to explain the differences between amds numbers and reviews. The first huge difference between amds review and the 3rd party ones are the memory speeds. We can argue back and forth about what is and isn't faster but realistically 6400mhz is like entry level memory for Intel. Not just for core ultra, even for rpl or alderlake, that's your baseline. I don't think guy with a 40/5090 gpu and a high end cpu is going to be running 6400 unless they are doing productivity and need the stability, in which case gaming performance is kinda irrelevant.

I have no problem with amds numbers BTW, dunno why people get so defensive. I was just pointing out the huge delta in memory speeds compared to 3rd party reviews.

 

[Lucky_SLS]

Well, I am happy that the performance lead is less than 10%. I am still a happy 7800X3D owner. Now, would this be the same case for Nvidia?

 

[YSCCC]

Well, I am happy that the performance lead is less than 10%. I am still a happy 7800X3D owner. Now, would this be the same case for Nvidia?

Somehow I feel Nvidia could have a larger gain, since the most demanding part of modern gaming are the rendering and RT/path tracing using the Nv promoted hardware, CPU mostly handles background info calculation.

But well, it's Nvidia, judging how they upnamed the 70 and 80 class GPU compared to the TOTL 90 tier in the 40X0 series, maybe they will just do it again and give you a 60 class core and call it 80 class and sell you at $1000, unless RDNA4 really as promised (I doubt it myself) being really cheap and yet performs well in the main stream 70 class segment

 

[TheHerald]

Well, I am happy that the performance lead is less than 10%. I am still a happy 7800X3D owner. Now, would this be the same case for Nvidia?

4090 ---> 5090 won't be less than 50%. I'd be shocked if it is.

 

[Thunder64]

4090 ---> 5090 won't be less than 50%. I'd be shocked if it is.

Nvidia doesn't give a crap about gamers anymore and with AMD opting out of the high end this time they are free to price them as they wish. Hopefully AMD/Intel will have good price/performance in the mid range.

 

[TheHerald]

Nvidia doesn't give a crap about gamers anymore and with AMD opting out of the high end this time they are free to price them as they wish. Hopefully AMD/Intel will have good price/performance in the mid range.

There was nothing stopping nvidia from pricing as they wish the last couple of generations anyways, since amd is just going to follow nvidias prices with a 10$ discount. If the 5080 is 1600$ amds competitor will be 1549$.

 

[Lucky_SLS]

the XX90 cards are the only outliers. Last gen cards moved the performance by a tier. But asked for more money at the same tier. So i dont know what to expect for blackwell.

 

[ottonis]

To all the people in this thread who were wondering about which RAM speeds were used in testing the new AMD vs Intel chips: does RAM speed even matter in a chip that already uses huge amounts of ultra-fast cash on-die?

 

[-Fran-]

To all the people in this thread who were wondering about which RAM speeds were used in testing the new AMD vs Intel chips: does RAM speed even matter in a chip that already uses huge amounts of ultra-fast cash on-die?

On AMD it still matters because of the IF to RAM speed ratio. Some ratios can actually hurt your overall performance due to, let's call them, timing shenanigans in the SoC.

The recommended (consensus) ratio for Infinity Fabric (FCLK) to memory clocks (MCLK) and BUS clock (BCLK) is 2:3:3. So, for instance. AMD testing with 6000MT/s is not a coincidence, since this is where their overclocked fabric usually lands on an attainable speed for most chips landing on the 2:3:3 "golden" ratio: 6000MT/s -> 3Ghz MCLK -> 3Ghz BCLK -> 2Ghz FCLK. These speeds, while considered overclocking (they technically ARE), it'll be rare for an AMD CPU to not achieve those 3 clocks. Higher speed RAM, to hit all three in that particular ratio, makes it harder and, like I said, different (dissimilar) ratios can actually worsen your performance.

Plenty of sweaty study has gone through in order to achieve that "truth" about AMD and a similar thing happens with Intel, but they handle their ratios very differently to AMD. They also call them "gears". I'm guessing to keep it simple, but it's the same idea behind: a "gear" defines the ratios at which the clocks in the CPU will work at against the memory clocks. I believe Arrow Lake defines up to "gear 4" this generation with CU-DIMMs, but I have no idea what ratios behind the scenes are used for those.

Now, specifically for the VCache'd siblings: the bigger L3 helps hide the latency aspect, but it doesn't eliminate the overall worse latency in AMD CPUs compared to Intel. Also true for bandwidth, which is something I've hated when they call it "effective bandwidth". Such a stupid thing. Anyway, it matters, but ever so slightly less than non-VCache'd CPUs and with the Intel counterpart. Once the L3 is full, then you're at the mercy of the RAM latency and bandwidth to retrieve whatever you need for the calculations.

Regards.

 

[TheHerald]

To all the people in this thread who were wondering about which RAM speeds were used in testing the new AMD vs Intel chips: does RAM speed even matter in a chip that already uses huge amounts of ultra-fast cash on-die?

If you mean the x3d, it still matters but much less than normal chips. Or to put it better, normal chips scale better with ram than the x3d does. You'll see 15-20% better performance by ram tuning on x3d vs 40% on non 3d parts.

 

[TheHerald]

On AMD it still matters because of the IF to RAM speed ratio. Some ratios can actually hurt your overall performance due to, let's call them, timing shenanigans in the SoC.

The recommended (consensus) ratio for Infinity Fabric (FCLK) to memory clocks (MCLK) and BUS clock (BCLK) is 2:3:3. So, for instance. AMD testing with 6000MT/s is not a coincidence, since this is where their overclocked fabric usually lands on an attainable speed for most chips landing on the 2:3:3 "golden" ratio: 6000MT/s -> 3Ghz MCLK -> 3Ghz BCLK -> 2Ghz FCLK. These speeds, while considered overclocking (they technically ARE), it'll be rare for an AMD CPU to not achieve those 3 clocks. Higher speed RAM, to hit all three in that particular ratio, makes it harder and, like I said, different (dissimilar) ratios can actually worsen your performance.

Plenty of sweaty study has gone through in order to achieve that "truth" about AMD and a similar thing happens with Intel, but they handle their ratios very differently to AMD. They also call them "gears". I'm guessing to keep it simple, but it's the same idea behind: a "gear" defines the ratios at which the clocks in the CPU will work at against the memory clocks. I believe Arrow Lake defines up to "gear 4" this generation with CU-DIMMs, but I have no idea what ratios behind the scenes are used for those.

Now, specifically for the VCache'd siblings: the bigger L3 helps hide the latency aspect, but it doesn't eliminate the overall worse latency in AMD CPUs compared to Intel. Also true for bandwidth, which is something I've hated when they call it "effective bandwidth". Such a stupid thing. Anyway, it matters, but ever so slightly less than non-VCache'd CPUs and with the Intel counterpart. Once the L3 is full, then you're at the mercy of the RAM latency and bandwidth to retrieve whatever you need for the calculations.

Regards.

Gear 2 halves the imc speed compared to ram, and gear 4 runs at 1/4th.

So gear 2 with 8000MT memory runs the imc at 2000mhz and gear 4 at 1000.

 

[ottonis]

On AMD it still matters because of the IF to RAM speed ratio. Some ratios can actually hurt your overall performance due to, let's call them, timing shenanigans in the SoC.

The recommended (consensus) ratio for Infinity Fabric (FCLK) to memory clocks (MCLK) and BUS clock (BCLK) is 2:3:3. So, for instance. AMD testing with 6000MT/s is not a coincidence, since this is where their overclocked fabric usually lands on an attainable speed for most chips landing on the 2:3:3 "golden" ratio: 6000MT/s -> 3Ghz MCLK -> 3Ghz BCLK -> 2Ghz FCLK. These speeds, while considered overclocking (they technically ARE), it'll be rare for an AMD CPU to not achieve those 3 clocks. Higher speed RAM, to hit all three in that particular ratio, makes it harder and, like I said, different (dissimilar) ratios can actually worsen your performance.

Plenty of sweaty study has gone through in order to achieve that "truth" about AMD and a similar thing happens with Intel, but they handle their ratios very differently to AMD. They also call them "gears". I'm guessing to keep it simple, but it's the same idea behind: a "gear" defines the ratios at which the clocks in the CPU will work at against the memory clocks. I believe Arrow Lake defines up to "gear 4" this generation with CU-DIMMs, but I have no idea what ratios behind the scenes are used for those.

Now, specifically for the VCache'd siblings: the bigger L3 helps hide the latency aspect, but it doesn't eliminate the overall worse latency in AMD CPUs compared to Intel. Also true for bandwidth, which is something I've hated when they call it "effective bandwidth". Such a stupid thing. Anyway, it matters, but ever so slightly less than non-VCache'd CPUs and with the Intel counterpart. Once the L3 is full, then you're at the mercy of the RAM latency and bandwidth to retrieve whatever you need for the calculations.

Regards.

I am impressed by your comprehensive explanation. Thanks a lot!

 

[-Fran-]

I am impressed by your comprehensive explanation. Thanks a lot!

No problem.

I didn't touch RAM timings on my explanation, in case someone wants to point that out, because it's a small rabbit hole as even with the same clocks, some sub-timings can have a huge impact on the memory-bound software and doesn't go against what I pointed out. Buildzoid has over 40 hours of videos going into very long-winded rants and really good explanations of this (including the ratios), so I rather not go there myself.

Also, there's other ratios that can work ifyou want or need the extra bandwidth of faster RAM, but the general rule of thumb for AMD CPUs is to try and increase all speeds in a way the ratios "make sense". AMD doesn't lock them, but that freedom to use disagregated clocks has the penalty of shooting yourself in the foot very easily.

Regards.

 

[helper800]

No problem.

I didn't touch RAM timings on my explanation, in case someone wants to point that out, because it's a small rabbit hole as even with the same clocks, some sub-timings can have a huge impact on the memory-bound software and doesn't go against what I pointed out. Buildzoid has over 40 hours of videos going into very long-winded rants and really good explanations of this (including the ratios), so I rather not go there myself.

Also, there's other ratios that can work ifyou want or need the extra bandwidth of faster RAM, but the general rule of thumb for AMD CPUs is to try and increase all speeds in a way the ratios "make sense". AMD doesn't lock them, but that freedom to use disagregated clocks has the penalty of shooting yourself in the foot very easily.

Regards.

I was under the impression that a 2:1:1 ratio was optimal. Was that with the 7000 CPUs and what you are explaining is for the 9000 CPUs? I know for a fact that 5000 and prior CPUs a 1:1:1 ratio was best and is what I am currently running with my 5800X3D, though, that's on DDR4 RAM compared to DDR5 RAM.

 

[-Fran-]

I was under the impression that a 2:1:1 ratio was optimal. Was that with the 7000 CPUs and what you are explaining is for the 9000 CPUs? I know for a fact that 5000 and prior CPUs a 1:1:1 ratio was best and is what I am currently running with my 5800X3D, though, that's on DDR4 RAM compared to DDR5 RAM.

Yes. Zen4 is different from Zen3 in that regard. In fact, in Zen3 they didn't separate IF from BUS clocks, so it was a 2 clock ratio instead and you'd just say 2:1. So me explaining the 3rd ratio implies Zen4 and Zen5.

Regards.

 

[helper800]

Yes. Zen4 is different from Zen3 in that regard. In fact, in Zen3 they didn't separate IF from BUS clocks, so it was a 2 clock ratio instead and you'd just say 2:1. So me explaining the 3rd ratio implies Zen4 and Zen5.

Regards.

From what I understand for Zen4-5 the best profiles are either very high MT/s kits and worse ratios or a 6000-6400 MT/s kit at 2:1:1. I am deferring to TPU here, and if you can explain what I am missing with what you are saying in regards to a 2:3:3 ratio being "best" that would be a great help. I have dabbled in RAM OCing with timings and sub timings, IF clocks, and tweaking the VDDC, but by no means am I an expert, I just want to understand what you mean.

 

[-Fran-]

From what I understand for Zen4-5 the best profiles are either very high MT/s kits and worse ratios or a 6000-6400 MT/s kit at 2:1:1. I am deferring to TPU here, and if you can explain what I am missing with what you are saying in regards to a 2:3:3 ratio being "best" that would be a great help. I have dabbled in RAM OCing with timings and sub timings, IF clocks, and tweaking the VDDC, but by no means am I an expert, I just want to understand what you mean.

View: https://www.youtube.com/watch?v=Xcn_nvWGj7U


Buildzoid explained it better.

Also, the chart I love giving to everyone:

Regards.