AMD Claims Dragon Range, Rembrandt-R CPUs Are Faster Than Alder Lake

jeremyj_83 said:

Ummm that isn't true at all. AMD cannot make enough Threadripper Pros to meet demand. Movie studios buy those immediately because they are the fastest things for content creation. Servers Intel is only at Milan level of performance with SPR. Mega compiles LOVE extra threads and once again Intel is lacking in that department.

Ummm no. You're comparing hedt to standard desktop parts. Apples and oranges. Like comparing a mustang to a rally car.

And mega compiles are throughout limited. It's also based on the number of parallel threads you can execute at the same time. With dependencies the most I ever seen with chromium compile is ~20. But it's dipped as low as 3 concurrent compile link stages. Now if you a are running a gated build server running multiple projects that's another issue. I will admit 5800x3d did hold a slight advantage over 5800x in compiles. That extra cache is good for symbol tables. But Intel still wins. Efficiency cores do exceptionally well here.

Adobe and quick sync do exceptionally well in premiere, and Photoshop.

13th gen Intel standard desktop is just faster. And it delivers a slightly better platform value. I'll reserve judgement for games till I see what x3d does.

And before you call me a blind Intel fanboi, check my sig on my systems.

digitalgriffin said:

Ummm no. You're comparing hedt to standard desktop parts. Apples and oranges. Like comparing a mustang to a rally car.

And mega compiles are throughout limited. It's also based on the number of parallel threads you can execute at the same time. With dependencies the most I ever seen with chromium compile is ~20. But it's dipped as low as 3 concurrent compile link stages. Now if you a are running a gated build server running multiple projects that's another issue. I will admit 5800x3d did hold a slight advantage over 5800x in compiles. That extra cache is good for symbol tables. But Intel still wins. Efficiency cores do exceptionally well here.

Adobe and quick sync do exceptionally well in premiere, and Photoshop.

13th gen Intel standard desktop is just faster. And it delivers a slightly better platform value. I'll reserve judgement for games till I see what x3d does.

And before you call me a blind Intel fanboi, check my sig on my systems.

digitalgriffin said:

Ummm no. You're comparing hedt to standard desktop parts. Apples and oranges. Like comparing a mustang to a rally car.

And mega compiles are throughout limited. It's also based on the number of parallel threads you can execute at the same time. With dependencies the most I ever seen with chromium compile is ~20. But it's dipped as low as 3 concurrent compile link stages. Now if you a are running a gated build server running multiple projects that's another issue. I will admit 5800x3d did hold a slight advantage over 5800x in compiles. That extra cache is good for symbol tables. But Intel still wins. Efficiency cores do exceptionally well here.

Adobe and quick sync do exceptionally well in premiere, and Photoshop.

13th gen Intel standard desktop is just faster. And it delivers a slightly better platform value. I'll reserve judgement for games till I see what x3d does.

And before you call me a blind Intel fanboi, check my sig on my systems.

You brought up more than just desktops which is why I said what I said. You cannot now move the goalpost to fit your narrative.

While Chromium compiles might max at 20 threads, Linux kernel compiles don't.

Only reason the 13th gen is faster is Intel has thrown power out the window for the last several generations. Anandtech did a look at a 13900k and 7950X at different TDPs. What they found is Zen is far more efficient than Core. In fact there were multiple instances that Intel needed 125W to beat the AMD at 65W. Limit both platforms to their states TDP and Intel is slower.

Sure Quick sync does work well in Premier and Photoshop, but the movie companies are still buying the AMD because they need more horsepower for things other than exporting.

Never called you a fanboi so don't put words in my mouth.

jeremyj_83 said:

You brought up more than just desktops which is why I said what I said. You cannot now move the goalpost to fit your narrative.

While Chromium compiles might max at 20 threads, Linux kernel compiles don't.

Only reason the 13th gen is faster is Intel has thrown power out the window for the last several generations. Anandtech did a look at a 13900k and 7950X at different TDPs. What they found is Zen is far more efficient than Core. In fact there were multiple instances that Intel needed 125W to beat the AMD at 65W. Limit both platforms to their states TDP and Intel is slower.

Sure Quick sync does work well in Premier and Photoshop, but the movie companies are still buying the AMD because they need more horsepower for things other than exporting.

Never called you a fanboi so don't put words in my mouth.

I moved no such goalpost. I wanted to play football. You wanted to throw in soccer for comparison. Your argument was a non starter to begin with. Heck you might as well bring in clustered computing on blades. An average user could use that right?

You can't compare $8000 parts to $700 parts. And at the end of the day for standard desktop parts that don't cost you a spleen, Intel wins on cost to performance and sheer performance.

Sure for server markets or workstation pro market, a build machine based on thread ripper would eat intels lunch. But we are are comparing laptop and standard desktop parts here. Joe Smoe average user stuff. Even at the largest of institutions as a principal software engineer I never got more than a desktop for the last 27 years. And the last 17 were laptops.

Power is a secondary concern when you care about speed. Let's say your average editor makes $40/hour. ($80k/year). Let's say Intel is 10% faster. (That's conservative with quick sync) That's 6 minutes saved. Or $4 in editor pay. Are you spending $4 extra per hour on Intel's higher power? Not even close.

I made the same argument for bulldozer. I'm not so worried about the extra power as that's pennies per hour at full tilt. Just as long as it doesn't sound like a blow dryer (easy to avoid with liquid cooling), you are good to go.

Either way I'm done. Be my guest and say all you want.

digitalgriffin said:

I moved no such goalpost. I wanted to play football. You wanted to throw in soccer for comparison. Your argument was a non starter to begin with. Heck you might as well bring in clustered computing on blades. An average user could use that right?

You said "For desktop Intel is still the way to go. Content creators, servers, renderers, and mega compiles all do better on Intel." I never said that desktop Intel isn't better I said "AMD cannot make enough Threadripper Pros to meet demand. Movie studios buy those immediately because they are the fastest things for content creation." I was getting at medium and larger content creators not someone with a small YouTube channel. You never specified one way or the other in your first sentence. Then in the next sentence you said content creators, servers, ... at that point you are onto a different thought and talking about more large scale things. FYI Ian Cutress was saying about a year ago that system integrators had been waiting for Zen 3 Threadripper because demand is VERY high for them. Medium and larger content creation companies, like movie studios, want/need as many cores as possible.

digitalgriffin said:

Sure for server markets or workstation pro market, a build machine based on thread ripper would eat intels lunch. But we are are comparing laptop and standard desktop parts here.

Again you never specified and specifically said servers in the same sentence as content creators. That would lead someone to believe that you are talking about massive workstations.

digitalgriffin said:

Power is a secondary concern when you care about speed. Let's say your average editor makes $40/hour. ($80k/year). Let's say Intel is 10% faster. (That's conservative with quick sync) That's 6 minutes saved. Or $4 in editor pay. Are you spending $4 extra per hour on Intel's higher power? Not even close.

IIRC Quick Sync is only used when exporting to your timeline. Most of the computations in Premier are still done on the CPU. Also unless Intel did some more work on Quick Sync it used to be the worst quality exporter. If you are doing a 480p YouTube video no one cares. However, if you are doing a movie then people will care about the quality. The nVidia and CPU exporters provide the best quality so you do have to take that into account. Sure you might save some money but if the quality isn't there then whats the point?

DaveLTX said:

"For those interested in integrated graphics performance, the Ryzen 9 7945HX is substantially faster than the Ryzen 9 6900HX."
Did anyone even proofread this? In what universe is a 6900HX igpu faster than 2 puny rDNA2 CUs
No matter how much faster the Zen 4 cpu is, it still wouldn't make 2 CUs faster than a whole 12 CUs with higher clocks
🤦‍♂️

My exact same thoughts: the 7000HX parts are basically 7000X parts with a lower TDP and a different die carrier, no significant iGPU power to speak of, certainly nothing to match a 12 CU APU.

And what current iGPUs are able to squeeze out of ordinary DRAM is already amazing enough: to double that doesn't just require more transistors in the iGPU, you need either multiplying memory channels (Apple) or a very high bandwidth memory type to match that.

I've always thought one could replace one of the CCDs with some type of HBM for doubling the bandwidth for an iGPU. But then you still need some die area for a fat iGPU and the overall power requirements might not be worth the gains.

Some Apple-like play on replicating CPU/GPU/RAM blocks has obviously been proven possible and attractive with the fruity faction, but if you still need to support external RAM, PCIe etc. all those chunky high-power transistors currently on the IOD may just eat too much space to make that possible in a notebook form factor.

AMD tries to reuse their chip designs to the max and the TAM for something really fancy probably just isn't there for the budget prices we'd want to afford them.

abufrejoval said:

My exact same thoughts: the 7000HX parts are basically 7000X parts with a lower TDP and a different die carrier, no significant iGPU power to speak of, certainly nothing to match a 12 CU APU.

And what current iGPUs are able to squeeze out of ordinary DRAM is already amazing enough: to double that doesn't just require more transistors in the iGPU, you need either multiplying memory channels (Apple) or a very high bandwidth memory type to match that.

I've always thought one could replace one of the CCDs with some type of HBM for doubling the bandwidth for an iGPU. But then you still need some die area for a fat iGPU and the overall power requirements might not be worth the gains.

Some Apple-like play on replicating CPU/GPU/RAM blocks has obviously been proven possible and attractive with the fruity faction, but if you still need to support external RAM, PCIe etc. all those chunky high-power transistors currently on the IOD may just eat too much space to make that possible in a notebook form factor.

AMD tries to reuse their chip designs to the max and the TAM for something really fancy probably just isn't there for the budget prices we'd want to afford them.

Yeah and external HBM if even possible would make them not so high bandwidth anymore. Or efficient.
Right now having all the GPU memory with the CPU reduces data movement heavily so they may actually be way more efficient staying with one memory type unless they use HBM as a L4 of sorts. But HBM is graphics optimized so while they are high bandwidth their latency is pretty poor so it is bad as L4. But there is substantial advantage to having more cache with iGPUs if the extra cache is accessible by the iGPU (fwiw I'm not sure the amd apus do, they have their own caches and don't go back to the L3 in the CCX)